xref: /dflybsd-src/sys/vfs/hammer2/hammer2_vnops.c (revision 37149c60342ce21064029fedf00ed32f4cce006b)
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 	 * WARNING: nvtruncbuf() can only be safely called without the inode
108 	 *	    lock held due to the way our write thread works.
109 	 */
110 	if (ip->flags & HAMMER2_INODE_ISUNLINKED) {
111 		hammer2_key_t lbase;
112 		int nblksize;
113 
114 		/*
115 		 * Detect updates to the embedded data which may be
116 		 * synchronized by the strategy code.  Simply mark the
117 		 * inode modified so it gets picked up by our normal flush.
118 		 */
119 		nblksize = hammer2_calc_logical(ip, 0, &lbase, NULL);
120 		nvtruncbuf(vp, 0, nblksize, 0, 0);
121 		vrecycle(vp);
122 	}
123 	LOCKSTOP;
124 	return (0);
125 }
126 
127 /*
128  * Reclaim a vnode so that it can be reused; after the inode is
129  * disassociated, the filesystem must manage it alone.
130  */
131 static
132 int
133 hammer2_vop_reclaim(struct vop_reclaim_args *ap)
134 {
135 	hammer2_inode_t *ip;
136 	hammer2_pfs_t *pmp;
137 	struct vnode *vp;
138 
139 	LOCKSTART;
140 	vp = ap->a_vp;
141 	ip = VTOI(vp);
142 	if (ip == NULL) {
143 		LOCKSTOP;
144 		return(0);
145 	}
146 	pmp = ip->pmp;
147 
148 	/*
149 	 * The final close of a deleted file or directory marks it for
150 	 * destruction.  The DELETED flag allows the flusher to shortcut
151 	 * any modified blocks still unflushed (that is, just ignore them).
152 	 *
153 	 * HAMMER2 usually does not try to optimize the freemap by returning
154 	 * deleted blocks to it as it does not usually know how many snapshots
155 	 * might be referencing portions of the file/dir.
156 	 */
157 	vp->v_data = NULL;
158 	ip->vp = NULL;
159 
160 	/*
161 	 * NOTE! We do not attempt to flush chains here, flushing is
162 	 *	 really fragile and could also deadlock.
163 	 */
164 	vclrisdirty(vp);
165 
166 	/*
167 	 * Once reclaimed the inode is disconnected from the normal flush
168 	 * mechanism and must be tracked
169 	 *
170 	 * A reclaim can occur at any time so we cannot safely start a
171 	 * transaction to handle reclamation of unlinked files.  Instead,
172 	 * the ip is left with a reference and placed on a linked list and
173 	 * handled later on.
174 	 */
175 	if (ip->flags & HAMMER2_INODE_ISUNLINKED) {
176 		hammer2_inode_unlink_t *ipul;
177 
178 		ipul = kmalloc(sizeof(*ipul), pmp->minode, M_WAITOK | M_ZERO);
179 		ipul->ip = ip;
180 
181 		hammer2_spin_ex(&pmp->list_spin);
182 		TAILQ_INSERT_TAIL(&pmp->unlinkq, ipul, entry);
183 		hammer2_spin_unex(&pmp->list_spin);
184 		/* retain ref from vp for ipul */
185 	} else {
186 		hammer2_inode_drop(ip);			/* vp ref */
187 	}
188 
189 	/*
190 	 * XXX handle background sync when ip dirty, kernel will no longer
191 	 * notify us regarding this inode because there is no longer a
192 	 * vnode attached to it.
193 	 */
194 
195 	LOCKSTOP;
196 	return (0);
197 }
198 
199 static
200 int
201 hammer2_vop_fsync(struct vop_fsync_args *ap)
202 {
203 	hammer2_inode_t *ip;
204 	struct vnode *vp;
205 
206 	LOCKSTART;
207 	vp = ap->a_vp;
208 	ip = VTOI(vp);
209 
210 #if 0
211 	/* XXX can't do this yet */
212 	hammer2_trans_init(ip->pmp, HAMMER2_TRANS_ISFLUSH);
213 	vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
214 #endif
215 	hammer2_trans_init(ip->pmp, 0);
216 	vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
217 
218 	/*
219 	 * Calling chain_flush here creates a lot of duplicative
220 	 * COW operations due to non-optimal vnode ordering.
221 	 *
222 	 * Only do it for an actual fsync() syscall.  The other forms
223 	 * which call this function will eventually call chain_flush
224 	 * on the volume root as a catch-all, which is far more optimal.
225 	 */
226 	hammer2_inode_lock(ip, 0);
227 	if (ip->flags & HAMMER2_INODE_MODIFIED)
228 		hammer2_inode_fsync(ip);
229 	hammer2_inode_unlock(ip);
230 	hammer2_trans_done(ip->pmp);
231 
232 	LOCKSTOP;
233 	return (0);
234 }
235 
236 static
237 int
238 hammer2_vop_access(struct vop_access_args *ap)
239 {
240 	hammer2_inode_t *ip = VTOI(ap->a_vp);
241 	uid_t uid;
242 	gid_t gid;
243 	int error;
244 
245 	LOCKSTART;
246 	hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
247 	uid = hammer2_to_unix_xid(&ip->meta.uid);
248 	gid = hammer2_to_unix_xid(&ip->meta.gid);
249 	error = vop_helper_access(ap, uid, gid, ip->meta.mode, ip->meta.uflags);
250 	hammer2_inode_unlock(ip);
251 
252 	LOCKSTOP;
253 	return (error);
254 }
255 
256 static
257 int
258 hammer2_vop_getattr(struct vop_getattr_args *ap)
259 {
260 	hammer2_pfs_t *pmp;
261 	hammer2_inode_t *ip;
262 	struct vnode *vp;
263 	struct vattr *vap;
264 	hammer2_chain_t *chain;
265 	int i;
266 
267 	LOCKSTART;
268 	vp = ap->a_vp;
269 	vap = ap->a_vap;
270 
271 	ip = VTOI(vp);
272 	pmp = ip->pmp;
273 
274 	hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
275 
276 	vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0];
277 	vap->va_fileid = ip->meta.inum;
278 	vap->va_mode = ip->meta.mode;
279 	vap->va_nlink = ip->meta.nlinks;
280 	vap->va_uid = hammer2_to_unix_xid(&ip->meta.uid);
281 	vap->va_gid = hammer2_to_unix_xid(&ip->meta.gid);
282 	vap->va_rmajor = 0;
283 	vap->va_rminor = 0;
284 	vap->va_size = ip->meta.size;	/* protected by shared lock */
285 	vap->va_blocksize = HAMMER2_PBUFSIZE;
286 	vap->va_flags = ip->meta.uflags;
287 	hammer2_time_to_timespec(ip->meta.ctime, &vap->va_ctime);
288 	hammer2_time_to_timespec(ip->meta.mtime, &vap->va_mtime);
289 	hammer2_time_to_timespec(ip->meta.mtime, &vap->va_atime);
290 	vap->va_gen = 1;
291 	vap->va_bytes = 0;
292 	for (i = 0; i < ip->cluster.nchains; ++i) {
293 		if ((chain = ip->cluster.array[i].chain) != NULL) {
294 			if (vap->va_bytes < chain->bref.data_count)
295 				vap->va_bytes = chain->bref.data_count;
296 		}
297 	}
298 	vap->va_type = hammer2_get_vtype(ip->meta.type);
299 	vap->va_filerev = 0;
300 	vap->va_uid_uuid = ip->meta.uid;
301 	vap->va_gid_uuid = ip->meta.gid;
302 	vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
303 			  VA_FSID_UUID_VALID;
304 
305 	hammer2_inode_unlock(ip);
306 
307 	LOCKSTOP;
308 	return (0);
309 }
310 
311 static
312 int
313 hammer2_vop_setattr(struct vop_setattr_args *ap)
314 {
315 	hammer2_inode_t *ip;
316 	struct vnode *vp;
317 	struct vattr *vap;
318 	int error;
319 	int kflags = 0;
320 	uint64_t ctime;
321 
322 	LOCKSTART;
323 	vp = ap->a_vp;
324 	vap = ap->a_vap;
325 	hammer2_update_time(&ctime);
326 
327 	ip = VTOI(vp);
328 
329 	if (ip->pmp->ronly) {
330 		LOCKSTOP;
331 		return(EROFS);
332 	}
333 
334 	hammer2_pfs_memory_wait(ip->pmp);
335 	hammer2_trans_init(ip->pmp, 0);
336 	hammer2_inode_lock(ip, 0);
337 	error = 0;
338 
339 	if (vap->va_flags != VNOVAL) {
340 		u_int32_t flags;
341 
342 		flags = ip->meta.uflags;
343 		error = vop_helper_setattr_flags(&flags, vap->va_flags,
344 				     hammer2_to_unix_xid(&ip->meta.uid),
345 				     ap->a_cred);
346 		if (error == 0) {
347 			if (ip->meta.uflags != flags) {
348 				hammer2_inode_modify(ip);
349 				ip->meta.uflags = flags;
350 				ip->meta.ctime = ctime;
351 				kflags |= NOTE_ATTRIB;
352 			}
353 			if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
354 				error = 0;
355 				goto done;
356 			}
357 		}
358 		goto done;
359 	}
360 	if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
361 		error = EPERM;
362 		goto done;
363 	}
364 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
365 		mode_t cur_mode = ip->meta.mode;
366 		uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
367 		gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
368 		uuid_t uuid_uid;
369 		uuid_t uuid_gid;
370 
371 		error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
372 					 ap->a_cred,
373 					 &cur_uid, &cur_gid, &cur_mode);
374 		if (error == 0) {
375 			hammer2_guid_to_uuid(&uuid_uid, cur_uid);
376 			hammer2_guid_to_uuid(&uuid_gid, cur_gid);
377 			if (bcmp(&uuid_uid, &ip->meta.uid, sizeof(uuid_uid)) ||
378 			    bcmp(&uuid_gid, &ip->meta.gid, sizeof(uuid_gid)) ||
379 			    ip->meta.mode != cur_mode
380 			) {
381 				hammer2_inode_modify(ip);
382 				ip->meta.uid = uuid_uid;
383 				ip->meta.gid = uuid_gid;
384 				ip->meta.mode = cur_mode;
385 				ip->meta.ctime = ctime;
386 			}
387 			kflags |= NOTE_ATTRIB;
388 		}
389 	}
390 
391 	/*
392 	 * Resize the file
393 	 */
394 	if (vap->va_size != VNOVAL && ip->meta.size != vap->va_size) {
395 		switch(vp->v_type) {
396 		case VREG:
397 			if (vap->va_size == ip->meta.size)
398 				break;
399 			if (vap->va_size < ip->meta.size) {
400 				hammer2_truncate_file(ip, vap->va_size);
401 			} else {
402 				hammer2_extend_file(ip, vap->va_size);
403 			}
404 			hammer2_inode_modify(ip);
405 			ip->meta.mtime = ctime;
406 			break;
407 		default:
408 			error = EINVAL;
409 			goto done;
410 		}
411 	}
412 #if 0
413 	/* atime not supported */
414 	if (vap->va_atime.tv_sec != VNOVAL) {
415 		hammer2_inode_modify(ip);
416 		ip->meta.atime = hammer2_timespec_to_time(&vap->va_atime);
417 		kflags |= NOTE_ATTRIB;
418 	}
419 #endif
420 	if (vap->va_mode != (mode_t)VNOVAL) {
421 		mode_t cur_mode = ip->meta.mode;
422 		uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
423 		gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
424 
425 		error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
426 					 cur_uid, cur_gid, &cur_mode);
427 		if (error == 0 && ip->meta.mode != cur_mode) {
428 			hammer2_inode_modify(ip);
429 			ip->meta.mode = cur_mode;
430 			ip->meta.ctime = ctime;
431 			kflags |= NOTE_ATTRIB;
432 		}
433 	}
434 
435 	if (vap->va_mtime.tv_sec != VNOVAL) {
436 		hammer2_inode_modify(ip);
437 		ip->meta.mtime = hammer2_timespec_to_time(&vap->va_mtime);
438 		kflags |= NOTE_ATTRIB;
439 	}
440 
441 done:
442 	/*
443 	 * If a truncation occurred we must call inode_fsync() now in order
444 	 * to trim the related data chains, otherwise a later expansion can
445 	 * cause havoc.
446 	 *
447 	 * If an extend occured that changed the DIRECTDATA state, we must
448 	 * call inode_fsync now in order to prepare the inode's indirect
449 	 * block table.
450 	 */
451 	if (ip->flags & HAMMER2_INODE_RESIZED)
452 		hammer2_inode_fsync(ip);
453 
454 	/*
455 	 * Cleanup.
456 	 */
457 	hammer2_inode_unlock(ip);
458 	hammer2_trans_done(ip->pmp);
459 	hammer2_knote(ip->vp, kflags);
460 
461 	LOCKSTOP;
462 	return (error);
463 }
464 
465 static
466 int
467 hammer2_vop_readdir(struct vop_readdir_args *ap)
468 {
469 	hammer2_xop_readdir_t *xop;
470 	hammer2_blockref_t bref;
471 	hammer2_inode_t *ip;
472 	hammer2_tid_t inum;
473 	hammer2_key_t lkey;
474 	struct uio *uio;
475 	off_t *cookies;
476 	off_t saveoff;
477 	int cookie_index;
478 	int ncookies;
479 	int error;
480 	int eofflag;
481 	int dtype;
482 	int r;
483 
484 	LOCKSTART;
485 	ip = VTOI(ap->a_vp);
486 	uio = ap->a_uio;
487 	saveoff = uio->uio_offset;
488 	eofflag = 0;
489 	error = 0;
490 
491 	/*
492 	 * Setup cookies directory entry cookies if requested
493 	 */
494 	if (ap->a_ncookies) {
495 		ncookies = uio->uio_resid / 16 + 1;
496 		if (ncookies > 1024)
497 			ncookies = 1024;
498 		cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
499 	} else {
500 		ncookies = -1;
501 		cookies = NULL;
502 	}
503 	cookie_index = 0;
504 
505 	hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
506 
507 	/*
508 	 * Handle artificial entries.  To ensure that only positive 64 bit
509 	 * quantities are returned to userland we always strip off bit 63.
510 	 * The hash code is designed such that codes 0x0000-0x7FFF are not
511 	 * used, allowing us to use these codes for articial entries.
512 	 *
513 	 * Entry 0 is used for '.' and entry 1 is used for '..'.  Do not
514 	 * allow '..' to cross the mount point into (e.g.) the super-root.
515 	 */
516 	if (saveoff == 0) {
517 		inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
518 		r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
519 		if (r)
520 			goto done;
521 		if (cookies)
522 			cookies[cookie_index] = saveoff;
523 		++saveoff;
524 		++cookie_index;
525 		if (cookie_index == ncookies)
526 			goto done;
527 	}
528 
529 	if (saveoff == 1) {
530 		/*
531 		 * Be careful with lockorder when accessing ".."
532 		 *
533 		 * (ip is the current dir. xip is the parent dir).
534 		 */
535 		inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
536 		if (ip->pip && ip != ip->pmp->iroot)
537 			inum = ip->pip->meta.inum & HAMMER2_DIRHASH_USERMSK;
538 		r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
539 		if (r)
540 			goto done;
541 		if (cookies)
542 			cookies[cookie_index] = saveoff;
543 		++saveoff;
544 		++cookie_index;
545 		if (cookie_index == ncookies)
546 			goto done;
547 	}
548 
549 	lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
550 	if (hammer2_debug & 0x0020)
551 		kprintf("readdir: lkey %016jx\n", lkey);
552 	if (error)
553 		goto done;
554 
555 	/*
556 	 * Use XOP for cluster scan.
557 	 *
558 	 * parent is the inode cluster, already locked for us.  Don't
559 	 * double lock shared locks as this will screw up upgrades.
560 	 */
561 	xop = hammer2_xop_alloc(ip, 0);
562 	xop->lkey = lkey;
563 	hammer2_xop_start(&xop->head, hammer2_xop_readdir);
564 
565 	for (;;) {
566 		const hammer2_inode_data_t *ripdata;
567 
568 		error = hammer2_xop_collect(&xop->head, 0);
569 		if (error)
570 			break;
571 		if (cookie_index == ncookies)
572 			break;
573 		if (hammer2_debug & 0x0020)
574 		kprintf("cluster chain %p %p\n",
575 			xop->head.cluster.focus,
576 			(xop->head.cluster.focus ?
577 			 xop->head.cluster.focus->data : (void *)-1));
578 		ripdata = &hammer2_cluster_rdata(&xop->head.cluster)->ipdata;
579 		hammer2_cluster_bref(&xop->head.cluster, &bref);
580 		if (bref.type == HAMMER2_BREF_TYPE_INODE) {
581 			dtype = hammer2_get_dtype(ripdata);
582 			saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
583 			r = vop_write_dirent(&error, uio,
584 					     ripdata->meta.inum &
585 					      HAMMER2_DIRHASH_USERMSK,
586 					     dtype,
587 					     ripdata->meta.name_len,
588 					     ripdata->filename);
589 			if (r)
590 				break;
591 			if (cookies)
592 				cookies[cookie_index] = saveoff;
593 			++cookie_index;
594 		} else {
595 			/* XXX chain error */
596 			kprintf("bad chain type readdir %d\n", bref.type);
597 		}
598 	}
599 	hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
600 	if (error == ENOENT) {
601 		error = 0;
602 		eofflag = 1;
603 		saveoff = (hammer2_key_t)-1;
604 	} else {
605 		saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
606 	}
607 done:
608 	hammer2_inode_unlock(ip);
609 	if (ap->a_eofflag)
610 		*ap->a_eofflag = eofflag;
611 	if (hammer2_debug & 0x0020)
612 		kprintf("readdir: done at %016jx\n", saveoff);
613 	uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
614 	if (error && cookie_index == 0) {
615 		if (cookies) {
616 			kfree(cookies, M_TEMP);
617 			*ap->a_ncookies = 0;
618 			*ap->a_cookies = NULL;
619 		}
620 	} else {
621 		if (cookies) {
622 			*ap->a_ncookies = cookie_index;
623 			*ap->a_cookies = cookies;
624 		}
625 	}
626 	LOCKSTOP;
627 	return (error);
628 }
629 
630 /*
631  * hammer2_vop_readlink { vp, uio, cred }
632  */
633 static
634 int
635 hammer2_vop_readlink(struct vop_readlink_args *ap)
636 {
637 	struct vnode *vp;
638 	hammer2_inode_t *ip;
639 	int error;
640 
641 	vp = ap->a_vp;
642 	if (vp->v_type != VLNK)
643 		return (EINVAL);
644 	ip = VTOI(vp);
645 
646 	error = hammer2_read_file(ip, ap->a_uio, 0);
647 	return (error);
648 }
649 
650 static
651 int
652 hammer2_vop_read(struct vop_read_args *ap)
653 {
654 	struct vnode *vp;
655 	hammer2_inode_t *ip;
656 	struct uio *uio;
657 	int error;
658 	int seqcount;
659 	int bigread;
660 
661 	/*
662 	 * Read operations supported on this vnode?
663 	 */
664 	vp = ap->a_vp;
665 	if (vp->v_type != VREG)
666 		return (EINVAL);
667 
668 	/*
669 	 * Misc
670 	 */
671 	ip = VTOI(vp);
672 	uio = ap->a_uio;
673 	error = 0;
674 
675 	seqcount = ap->a_ioflag >> 16;
676 	bigread = (uio->uio_resid > 100 * 1024 * 1024);
677 
678 	error = hammer2_read_file(ip, uio, seqcount);
679 	return (error);
680 }
681 
682 static
683 int
684 hammer2_vop_write(struct vop_write_args *ap)
685 {
686 	hammer2_inode_t *ip;
687 	thread_t td;
688 	struct vnode *vp;
689 	struct uio *uio;
690 	int error;
691 	int seqcount;
692 
693 	/*
694 	 * Read operations supported on this vnode?
695 	 */
696 	vp = ap->a_vp;
697 	if (vp->v_type != VREG)
698 		return (EINVAL);
699 
700 	/*
701 	 * Misc
702 	 */
703 	ip = VTOI(vp);
704 	uio = ap->a_uio;
705 	error = 0;
706 	if (ip->pmp->ronly) {
707 		return (EROFS);
708 	}
709 
710 	seqcount = ap->a_ioflag >> 16;
711 
712 	/*
713 	 * Check resource limit
714 	 */
715 	if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc &&
716 	    uio->uio_offset + uio->uio_resid >
717 	     td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
718 		lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
719 		return (EFBIG);
720 	}
721 
722 	/*
723 	 * The transaction interlocks against flushes initiations
724 	 * (note: but will run concurrently with the actual flush).
725 	 */
726 	hammer2_trans_init(ip->pmp, 0);
727 	error = hammer2_write_file(ip, uio, ap->a_ioflag, seqcount);
728 	hammer2_trans_done(ip->pmp);
729 
730 	return (error);
731 }
732 
733 /*
734  * Perform read operations on a file or symlink given an UNLOCKED
735  * inode and uio.
736  *
737  * The passed ip is not locked.
738  */
739 static
740 int
741 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
742 {
743 	hammer2_off_t size;
744 	struct buf *bp;
745 	int error;
746 
747 	error = 0;
748 
749 	/*
750 	 * UIO read loop.
751 	 *
752 	 * WARNING! Assumes that the kernel interlocks size changes at the
753 	 *	    vnode level.
754 	 */
755 	hammer2_mtx_sh(&ip->lock);
756 	size = ip->meta.size;
757 	hammer2_mtx_unlock(&ip->lock);
758 
759 	while (uio->uio_resid > 0 && uio->uio_offset < size) {
760 		hammer2_key_t lbase;
761 		hammer2_key_t leof;
762 		int lblksize;
763 		int loff;
764 		int n;
765 
766 		lblksize = hammer2_calc_logical(ip, uio->uio_offset,
767 						&lbase, &leof);
768 
769 		error = cluster_read(ip->vp, leof, lbase, lblksize,
770 				     uio->uio_resid, seqcount * BKVASIZE,
771 				     &bp);
772 
773 		if (error)
774 			break;
775 		loff = (int)(uio->uio_offset - lbase);
776 		n = lblksize - loff;
777 		if (n > uio->uio_resid)
778 			n = uio->uio_resid;
779 		if (n > size - uio->uio_offset)
780 			n = (int)(size - uio->uio_offset);
781 		bp->b_flags |= B_AGE;
782 		uiomove((char *)bp->b_data + loff, n, uio);
783 		bqrelse(bp);
784 	}
785 	return (error);
786 }
787 
788 /*
789  * Write to the file represented by the inode via the logical buffer cache.
790  * The inode may represent a regular file or a symlink.
791  *
792  * The inode must not be locked.
793  */
794 static
795 int
796 hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
797 		   int ioflag, int seqcount)
798 {
799 	hammer2_key_t old_eof;
800 	hammer2_key_t new_eof;
801 	struct buf *bp;
802 	int kflags;
803 	int error;
804 	int modified;
805 
806 	/*
807 	 * Setup if append
808 	 *
809 	 * WARNING! Assumes that the kernel interlocks size changes at the
810 	 *	    vnode level.
811 	 */
812 	hammer2_mtx_ex(&ip->lock);
813 	if (ioflag & IO_APPEND)
814 		uio->uio_offset = ip->meta.size;
815 	old_eof = ip->meta.size;
816 
817 	/*
818 	 * Extend the file if necessary.  If the write fails at some point
819 	 * we will truncate it back down to cover as much as we were able
820 	 * to write.
821 	 *
822 	 * Doing this now makes it easier to calculate buffer sizes in
823 	 * the loop.
824 	 */
825 	kflags = 0;
826 	error = 0;
827 	modified = 0;
828 
829 	if (uio->uio_offset + uio->uio_resid > old_eof) {
830 		new_eof = uio->uio_offset + uio->uio_resid;
831 		modified = 1;
832 		hammer2_extend_file(ip, new_eof);
833 		kflags |= NOTE_EXTEND;
834 	} else {
835 		new_eof = old_eof;
836 	}
837 	hammer2_mtx_unlock(&ip->lock);
838 
839 	/*
840 	 * UIO write loop
841 	 */
842 	while (uio->uio_resid > 0) {
843 		hammer2_key_t lbase;
844 		int trivial;
845 		int endofblk;
846 		int lblksize;
847 		int loff;
848 		int n;
849 
850 		/*
851 		 * Don't allow the buffer build to blow out the buffer
852 		 * cache.
853 		 */
854 		if ((ioflag & IO_RECURSE) == 0)
855 			bwillwrite(HAMMER2_PBUFSIZE);
856 
857 		/*
858 		 * This nominally tells us how much we can cluster and
859 		 * what the logical buffer size needs to be.  Currently
860 		 * we don't try to cluster the write and just handle one
861 		 * block at a time.
862 		 */
863 		lblksize = hammer2_calc_logical(ip, uio->uio_offset,
864 						&lbase, NULL);
865 		loff = (int)(uio->uio_offset - lbase);
866 
867 		KKASSERT(lblksize <= 65536);
868 
869 		/*
870 		 * Calculate bytes to copy this transfer and whether the
871 		 * copy completely covers the buffer or not.
872 		 */
873 		trivial = 0;
874 		n = lblksize - loff;
875 		if (n > uio->uio_resid) {
876 			n = uio->uio_resid;
877 			if (loff == lbase && uio->uio_offset + n == new_eof)
878 				trivial = 1;
879 			endofblk = 0;
880 		} else {
881 			if (loff == 0)
882 				trivial = 1;
883 			endofblk = 1;
884 		}
885 
886 		/*
887 		 * Get the buffer
888 		 */
889 		if (uio->uio_segflg == UIO_NOCOPY) {
890 			/*
891 			 * Issuing a write with the same data backing the
892 			 * buffer.  Instantiate the buffer to collect the
893 			 * backing vm pages, then read-in any missing bits.
894 			 *
895 			 * This case is used by vop_stdputpages().
896 			 */
897 			bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
898 			if ((bp->b_flags & B_CACHE) == 0) {
899 				bqrelse(bp);
900 				error = bread(ip->vp, lbase, lblksize, &bp);
901 			}
902 		} else if (trivial) {
903 			/*
904 			 * Even though we are entirely overwriting the buffer
905 			 * we may still have to zero it out to avoid a
906 			 * mmap/write visibility issue.
907 			 */
908 			bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
909 			if ((bp->b_flags & B_CACHE) == 0)
910 				vfs_bio_clrbuf(bp);
911 		} else {
912 			/*
913 			 * Partial overwrite, read in any missing bits then
914 			 * replace the portion being written.
915 			 *
916 			 * (The strategy code will detect zero-fill physical
917 			 * blocks for this case).
918 			 */
919 			error = bread(ip->vp, lbase, lblksize, &bp);
920 			if (error == 0)
921 				bheavy(bp);
922 		}
923 
924 		if (error) {
925 			brelse(bp);
926 			break;
927 		}
928 
929 		/*
930 		 * Ok, copy the data in
931 		 */
932 		error = uiomove(bp->b_data + loff, n, uio);
933 		kflags |= NOTE_WRITE;
934 		modified = 1;
935 		if (error) {
936 			brelse(bp);
937 			break;
938 		}
939 
940 		/*
941 		 * WARNING: Pageout daemon will issue UIO_NOCOPY writes
942 		 *	    with IO_SYNC or IO_ASYNC set.  These writes
943 		 *	    must be handled as the pageout daemon expects.
944 		 */
945 		if (ioflag & IO_SYNC) {
946 			bwrite(bp);
947 		} else if ((ioflag & IO_DIRECT) && endofblk) {
948 			bawrite(bp);
949 		} else if (ioflag & IO_ASYNC) {
950 			bawrite(bp);
951 		} else {
952 			bdwrite(bp);
953 		}
954 	}
955 
956 	/*
957 	 * Cleanup.  If we extended the file EOF but failed to write through
958 	 * the entire write is a failure and we have to back-up.
959 	 */
960 	if (error && new_eof != old_eof) {
961 		hammer2_mtx_ex(&ip->lock);
962 		hammer2_truncate_file(ip, old_eof);
963 		if (ip->flags & HAMMER2_INODE_MODIFIED)
964 			hammer2_inode_fsync(ip);
965 		hammer2_mtx_unlock(&ip->lock);
966 	} else if (modified) {
967 		hammer2_mtx_ex(&ip->lock);
968 		hammer2_inode_modify(ip);
969 		hammer2_update_time(&ip->meta.mtime);
970 		if (ip->flags & HAMMER2_INODE_MODIFIED)
971 			hammer2_inode_fsync(ip);
972 		hammer2_mtx_unlock(&ip->lock);
973 		hammer2_knote(ip->vp, kflags);
974 	}
975 	hammer2_trans_assert_strategy(ip->pmp);
976 
977 	return error;
978 }
979 
980 /*
981  * Truncate the size of a file.  The inode must not be locked.
982  *
983  * We must unconditionally set HAMMER2_INODE_RESIZED to properly
984  * ensure that any on-media data beyond the new file EOF has been destroyed.
985  *
986  * WARNING: nvtruncbuf() can only be safely called without the inode lock
987  *	    held due to the way our write thread works.  If the truncation
988  *	    occurs in the middle of a buffer, nvtruncbuf() is responsible
989  *	    for dirtying that buffer and zeroing out trailing bytes.
990  *
991  * WARNING! Assumes that the kernel interlocks size changes at the
992  *	    vnode level.
993  *
994  * WARNING! Caller assumes responsibility for removing dead blocks
995  *	    if INODE_RESIZED is set.
996  */
997 static
998 void
999 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1000 {
1001 	hammer2_key_t lbase;
1002 	int nblksize;
1003 
1004 	LOCKSTART;
1005 	hammer2_mtx_unlock(&ip->lock);
1006 	if (ip->vp) {
1007 		nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1008 		nvtruncbuf(ip->vp, nsize,
1009 			   nblksize, (int)nsize & (nblksize - 1),
1010 			   0);
1011 	}
1012 	hammer2_mtx_ex(&ip->lock);
1013 	KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1014 	ip->osize = ip->meta.size;
1015 	ip->meta.size = nsize;
1016 	atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED |
1017 				   HAMMER2_INODE_RESIZED);
1018 	LOCKSTOP;
1019 }
1020 
1021 /*
1022  * Extend the size of a file.  The inode must not be locked.
1023  *
1024  * Even though the file size is changing, we do not have to set the
1025  * INODE_RESIZED bit unless the file size crosses the EMBEDDED_BYTES
1026  * boundary.  When this occurs a hammer2_inode_fsync() is required
1027  * to prepare the inode cluster's indirect block table.
1028  *
1029  * WARNING! Assumes that the kernel interlocks size changes at the
1030  *	    vnode level.
1031  *
1032  * WARNING! Caller assumes responsibility for transitioning out
1033  *	    of the inode DIRECTDATA mode if INODE_RESIZED is set.
1034  */
1035 static
1036 void
1037 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1038 {
1039 	hammer2_key_t lbase;
1040 	hammer2_key_t osize;
1041 	int oblksize;
1042 	int nblksize;
1043 
1044 	LOCKSTART;
1045 
1046 	KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1047 	osize = ip->meta.size;
1048 	ip->osize = osize;
1049 	ip->meta.size = nsize;
1050 	atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED);
1051 
1052 	if (osize <= HAMMER2_EMBEDDED_BYTES && nsize > HAMMER2_EMBEDDED_BYTES)
1053 		atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1054 
1055 	hammer2_mtx_unlock(&ip->lock);
1056 	if (ip->vp) {
1057 		oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL);
1058 		nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1059 		nvextendbuf(ip->vp,
1060 			    osize, nsize,
1061 			    oblksize, nblksize,
1062 			    -1, -1, 0);
1063 	}
1064 	hammer2_mtx_ex(&ip->lock);
1065 
1066 	LOCKSTOP;
1067 }
1068 
1069 static
1070 int
1071 hammer2_vop_nresolve(struct vop_nresolve_args *ap)
1072 {
1073 	hammer2_xop_nresolve_t *xop;
1074 	hammer2_inode_t *ip;
1075 	hammer2_inode_t *dip;
1076 	struct namecache *ncp;
1077 	struct vnode *vp;
1078 	int error;
1079 
1080 	LOCKSTART;
1081 	dip = VTOI(ap->a_dvp);
1082 	xop = hammer2_xop_alloc(dip, 0);
1083 
1084 	ncp = ap->a_nch->ncp;
1085 	hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1086 
1087 	/*
1088 	 * Note: In DragonFly the kernel handles '.' and '..'.
1089 	 */
1090 	hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1091 	hammer2_xop_start(&xop->head, hammer2_xop_nresolve);
1092 
1093 	error = hammer2_xop_collect(&xop->head, 0);
1094 	if (error) {
1095 		ip = NULL;
1096 	} else {
1097 		ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1098 	}
1099 	hammer2_inode_unlock(dip);
1100 
1101 	/*
1102 	 * Acquire the related vnode
1103 	 *
1104 	 * NOTE: For error processing, only ENOENT resolves the namecache
1105 	 *	 entry to NULL, otherwise we just return the error and
1106 	 *	 leave the namecache unresolved.
1107 	 *
1108 	 * NOTE: multiple hammer2_inode structures can be aliased to the
1109 	 *	 same chain element, for example for hardlinks.  This
1110 	 *	 use case does not 'reattach' inode associations that
1111 	 *	 might already exist, but always allocates a new one.
1112 	 *
1113 	 * WARNING: inode structure is locked exclusively via inode_get
1114 	 *	    but chain was locked shared.  inode_unlock()
1115 	 *	    will handle it properly.
1116 	 */
1117 	if (ip) {
1118 		vp = hammer2_igetv(ip, &error);
1119 		if (error == 0) {
1120 			vn_unlock(vp);
1121 			cache_setvp(ap->a_nch, vp);
1122 		} else if (error == ENOENT) {
1123 			cache_setvp(ap->a_nch, NULL);
1124 		}
1125 		hammer2_inode_unlock(ip);
1126 
1127 		/*
1128 		 * The vp should not be released until after we've disposed
1129 		 * of our locks, because it might cause vop_inactive() to
1130 		 * be called.
1131 		 */
1132 		if (vp)
1133 			vrele(vp);
1134 	} else {
1135 		error = ENOENT;
1136 		cache_setvp(ap->a_nch, NULL);
1137 	}
1138 	hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1139 	KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
1140 		("resolve error %d/%p ap %p\n",
1141 		 error, ap->a_nch->ncp->nc_vp, ap));
1142 	LOCKSTOP;
1143 
1144 	return error;
1145 }
1146 
1147 static
1148 int
1149 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
1150 {
1151 	hammer2_inode_t *dip;
1152 	hammer2_inode_t *ip;
1153 	int error;
1154 
1155 	LOCKSTART;
1156 	dip = VTOI(ap->a_dvp);
1157 
1158 	if ((ip = dip->pip) == NULL) {
1159 		*ap->a_vpp = NULL;
1160 		LOCKSTOP;
1161 		return ENOENT;
1162 	}
1163 	hammer2_inode_lock(ip, 0);
1164 	*ap->a_vpp = hammer2_igetv(ip, &error);
1165 	hammer2_inode_unlock(ip);
1166 
1167 	LOCKSTOP;
1168 	return error;
1169 }
1170 
1171 static
1172 int
1173 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
1174 {
1175 	hammer2_inode_t *dip;
1176 	hammer2_inode_t *nip;
1177 	struct namecache *ncp;
1178 	const uint8_t *name;
1179 	size_t name_len;
1180 	int error;
1181 
1182 	LOCKSTART;
1183 	dip = VTOI(ap->a_dvp);
1184 	if (dip->pmp->ronly) {
1185 		LOCKSTOP;
1186 		return (EROFS);
1187 	}
1188 
1189 	ncp = ap->a_nch->ncp;
1190 	name = ncp->nc_name;
1191 	name_len = ncp->nc_nlen;
1192 
1193 	hammer2_pfs_memory_wait(dip->pmp);
1194 	hammer2_trans_init(dip->pmp, 0);
1195 	nip = hammer2_inode_create(dip, ap->a_vap, ap->a_cred,
1196 				   name, name_len, 0,
1197 				   hammer2_trans_newinum(dip->pmp), 0, 0,
1198 				   0, &error);
1199 	if (error) {
1200 		KKASSERT(nip == NULL);
1201 		*ap->a_vpp = NULL;
1202 	} else {
1203 		*ap->a_vpp = hammer2_igetv(nip, &error);
1204 		hammer2_inode_unlock(nip);
1205 	}
1206 	hammer2_trans_done(dip->pmp);
1207 
1208 	if (error == 0) {
1209 		cache_setunresolved(ap->a_nch);
1210 		cache_setvp(ap->a_nch, *ap->a_vpp);
1211 	}
1212 	LOCKSTOP;
1213 	return error;
1214 }
1215 
1216 static
1217 int
1218 hammer2_vop_open(struct vop_open_args *ap)
1219 {
1220 	return vop_stdopen(ap);
1221 }
1222 
1223 /*
1224  * hammer2_vop_advlock { vp, id, op, fl, flags }
1225  */
1226 static
1227 int
1228 hammer2_vop_advlock(struct vop_advlock_args *ap)
1229 {
1230 	hammer2_inode_t *ip = VTOI(ap->a_vp);
1231 	hammer2_off_t size;
1232 
1233 	size = ip->meta.size;
1234 	return (lf_advlock(ap, &ip->advlock, size));
1235 }
1236 
1237 static
1238 int
1239 hammer2_vop_close(struct vop_close_args *ap)
1240 {
1241 	return vop_stdclose(ap);
1242 }
1243 
1244 /*
1245  * hammer2_vop_nlink { nch, dvp, vp, cred }
1246  *
1247  * Create a hardlink from (vp) to {dvp, nch}.
1248  */
1249 static
1250 int
1251 hammer2_vop_nlink(struct vop_nlink_args *ap)
1252 {
1253 	hammer2_xop_nlink_t *xop1;
1254 	hammer2_inode_t *fdip;	/* target directory to create link in */
1255 	hammer2_inode_t *tdip;	/* target directory to create link in */
1256 	hammer2_inode_t *cdip;	/* common parent directory */
1257 	hammer2_inode_t *ip;	/* inode we are hardlinking to */
1258 	struct namecache *ncp;
1259 	const uint8_t *name;
1260 	size_t name_len;
1261 	int error;
1262 
1263 	LOCKSTART;
1264 	tdip = VTOI(ap->a_dvp);
1265 	if (tdip->pmp->ronly) {
1266 		LOCKSTOP;
1267 		return (EROFS);
1268 	}
1269 
1270 	ncp = ap->a_nch->ncp;
1271 	name = ncp->nc_name;
1272 	name_len = ncp->nc_nlen;
1273 
1274 	/*
1275 	 * ip represents the file being hardlinked.  The file could be a
1276 	 * normal file or a hardlink target if it has already been hardlinked.
1277 	 * If ip is a hardlinked target then ip->pip represents the location
1278 	 * of the hardlinked target, NOT the location of the hardlink pointer.
1279 	 *
1280 	 * Bump nlinks and potentially also create or move the hardlink
1281 	 * target in the parent directory common to (ip) and (tdip).  The
1282 	 * consolidation code can modify ip->cluster and ip->pip.  The
1283 	 * returned cluster is locked.
1284 	 */
1285 	ip = VTOI(ap->a_vp);
1286 	hammer2_pfs_memory_wait(ip->pmp);
1287 	hammer2_trans_init(ip->pmp, 0);
1288 
1289 	/*
1290 	 * The common parent directory must be locked first to avoid deadlocks.
1291 	 * Also note that fdip and/or tdip might match cdip.
1292 	 */
1293 	fdip = ip->pip;
1294 	cdip = hammer2_inode_common_parent(fdip, tdip);
1295 	hammer2_inode_lock(cdip, 0);
1296 	hammer2_inode_lock(fdip, 0);
1297 	hammer2_inode_lock(tdip, 0);
1298 	hammer2_inode_lock(ip, 0);
1299 	error = 0;
1300 
1301 	/*
1302 	 * If ip is not a hardlink target we must convert it to a hardlink.
1303 	 * If fdip != cdip we must shift the inode to cdip.
1304 	 */
1305 	if (fdip != cdip || (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
1306 		xop1 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
1307 		hammer2_xop_setip2(&xop1->head, ip);
1308 		hammer2_xop_setip3(&xop1->head, cdip);
1309 
1310 		hammer2_xop_start(&xop1->head, hammer2_xop_nlink);
1311 		error = hammer2_xop_collect(&xop1->head, 0);
1312 		hammer2_xop_retire(&xop1->head, HAMMER2_XOPMASK_VOP);
1313 		if (error == ENOENT)
1314 			error = 0;
1315 	}
1316 
1317 	/*
1318 	 * Must synchronize original inode whos chains are now a hardlink
1319 	 * target.  We must match what the backend XOP did to the
1320 	 * chains.
1321 	 */
1322 	if (error == 0 && (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
1323 		hammer2_inode_modify(ip);
1324 		ip->meta.name_key = ip->meta.inum;
1325 		ip->meta.name_len = 18;	/* "0x%016jx" */
1326 	}
1327 
1328 	/*
1329 	 * Create the hardlink target and bump nlinks.
1330 	 */
1331 	if (error == 0) {
1332 		hammer2_inode_create(tdip, NULL, NULL,
1333 				     name, name_len, 0,
1334 				     ip->meta.inum,
1335 				     HAMMER2_OBJTYPE_HARDLINK, ip->meta.type,
1336 				     0, &error);
1337 		hammer2_inode_modify(ip);
1338 		++ip->meta.nlinks;
1339 	}
1340 	if (error == 0) {
1341 		cache_setunresolved(ap->a_nch);
1342 		cache_setvp(ap->a_nch, ap->a_vp);
1343 	}
1344 	hammer2_inode_unlock(ip);
1345 	hammer2_inode_unlock(tdip);
1346 	hammer2_inode_unlock(fdip);
1347 	hammer2_inode_unlock(cdip);
1348 	hammer2_inode_drop(cdip);
1349 	hammer2_trans_done(ip->pmp);
1350 
1351 	LOCKSTOP;
1352 	return error;
1353 }
1354 
1355 /*
1356  * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1357  *
1358  * The operating system has already ensured that the directory entry
1359  * does not exist and done all appropriate namespace locking.
1360  */
1361 static
1362 int
1363 hammer2_vop_ncreate(struct vop_ncreate_args *ap)
1364 {
1365 	hammer2_inode_t *dip;
1366 	hammer2_inode_t *nip;
1367 	struct namecache *ncp;
1368 	const uint8_t *name;
1369 	size_t name_len;
1370 	int error;
1371 
1372 	LOCKSTART;
1373 	dip = VTOI(ap->a_dvp);
1374 	if (dip->pmp->ronly) {
1375 		LOCKSTOP;
1376 		return (EROFS);
1377 	}
1378 
1379 	ncp = ap->a_nch->ncp;
1380 	name = ncp->nc_name;
1381 	name_len = ncp->nc_nlen;
1382 	hammer2_pfs_memory_wait(dip->pmp);
1383 	hammer2_trans_init(dip->pmp, 0);
1384 
1385 	nip = hammer2_inode_create(dip, ap->a_vap, ap->a_cred,
1386 				   name, name_len, 0,
1387 				   hammer2_trans_newinum(dip->pmp), 0, 0,
1388 				   0, &error);
1389 	if (error) {
1390 		KKASSERT(nip == NULL);
1391 		*ap->a_vpp = NULL;
1392 	} else {
1393 		*ap->a_vpp = hammer2_igetv(nip, &error);
1394 		hammer2_inode_unlock(nip);
1395 	}
1396 	hammer2_trans_done(dip->pmp);
1397 
1398 	if (error == 0) {
1399 		cache_setunresolved(ap->a_nch);
1400 		cache_setvp(ap->a_nch, *ap->a_vpp);
1401 	}
1402 	LOCKSTOP;
1403 	return error;
1404 }
1405 
1406 /*
1407  * Make a device node (typically a fifo)
1408  */
1409 static
1410 int
1411 hammer2_vop_nmknod(struct vop_nmknod_args *ap)
1412 {
1413 	hammer2_inode_t *dip;
1414 	hammer2_inode_t *nip;
1415 	struct namecache *ncp;
1416 	const uint8_t *name;
1417 	size_t name_len;
1418 	int error;
1419 
1420 	LOCKSTART;
1421 	dip = VTOI(ap->a_dvp);
1422 	if (dip->pmp->ronly) {
1423 		LOCKSTOP;
1424 		return (EROFS);
1425 	}
1426 
1427 	ncp = ap->a_nch->ncp;
1428 	name = ncp->nc_name;
1429 	name_len = ncp->nc_nlen;
1430 	hammer2_pfs_memory_wait(dip->pmp);
1431 	hammer2_trans_init(dip->pmp, 0);
1432 
1433 	nip = hammer2_inode_create(dip, ap->a_vap, ap->a_cred,
1434 				   name, name_len, 0,
1435 				   hammer2_trans_newinum(dip->pmp), 0, 0,
1436 				   0, &error);
1437 	if (error) {
1438 		KKASSERT(nip == NULL);
1439 		*ap->a_vpp = NULL;
1440 	} else {
1441 		*ap->a_vpp = hammer2_igetv(nip, &error);
1442 		hammer2_inode_unlock(nip);
1443 	}
1444 	hammer2_trans_done(dip->pmp);
1445 
1446 	if (error == 0) {
1447 		cache_setunresolved(ap->a_nch);
1448 		cache_setvp(ap->a_nch, *ap->a_vpp);
1449 	}
1450 	LOCKSTOP;
1451 	return error;
1452 }
1453 
1454 /*
1455  * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1456  */
1457 static
1458 int
1459 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
1460 {
1461 	hammer2_inode_t *dip;
1462 	hammer2_inode_t *nip;
1463 	struct namecache *ncp;
1464 	const uint8_t *name;
1465 	size_t name_len;
1466 	int error;
1467 
1468 	dip = VTOI(ap->a_dvp);
1469 	if (dip->pmp->ronly)
1470 		return (EROFS);
1471 
1472 	ncp = ap->a_nch->ncp;
1473 	name = ncp->nc_name;
1474 	name_len = ncp->nc_nlen;
1475 	hammer2_pfs_memory_wait(dip->pmp);
1476 	hammer2_trans_init(dip->pmp, 0);
1477 
1478 	ap->a_vap->va_type = VLNK;	/* enforce type */
1479 
1480 	nip = hammer2_inode_create(dip, ap->a_vap, ap->a_cred,
1481 				   name, name_len, 0,
1482 				   hammer2_trans_newinum(dip->pmp), 0, 0,
1483 				   0, &error);
1484 	if (error) {
1485 		KKASSERT(nip == NULL);
1486 		*ap->a_vpp = NULL;
1487 		hammer2_trans_done(dip->pmp);
1488 		return error;
1489 	}
1490 	*ap->a_vpp = hammer2_igetv(nip, &error);
1491 
1492 	/*
1493 	 * Build the softlink (~like file data) and finalize the namecache.
1494 	 */
1495 	if (error == 0) {
1496 		size_t bytes;
1497 		struct uio auio;
1498 		struct iovec aiov;
1499 
1500 		bytes = strlen(ap->a_target);
1501 
1502 		hammer2_inode_unlock(nip);
1503 		bzero(&auio, sizeof(auio));
1504 		bzero(&aiov, sizeof(aiov));
1505 		auio.uio_iov = &aiov;
1506 		auio.uio_segflg = UIO_SYSSPACE;
1507 		auio.uio_rw = UIO_WRITE;
1508 		auio.uio_resid = bytes;
1509 		auio.uio_iovcnt = 1;
1510 		auio.uio_td = curthread;
1511 		aiov.iov_base = ap->a_target;
1512 		aiov.iov_len = bytes;
1513 		error = hammer2_write_file(nip, &auio, IO_APPEND, 0);
1514 		/* XXX handle error */
1515 		error = 0;
1516 	} else {
1517 		hammer2_inode_unlock(nip);
1518 	}
1519 	hammer2_trans_done(dip->pmp);
1520 
1521 	/*
1522 	 * Finalize namecache
1523 	 */
1524 	if (error == 0) {
1525 		cache_setunresolved(ap->a_nch);
1526 		cache_setvp(ap->a_nch, *ap->a_vpp);
1527 		/* hammer2_knote(ap->a_dvp, NOTE_WRITE); */
1528 	}
1529 	return error;
1530 }
1531 
1532 /*
1533  * hammer2_vop_nremove { nch, dvp, cred }
1534  */
1535 static
1536 int
1537 hammer2_vop_nremove(struct vop_nremove_args *ap)
1538 {
1539 	hammer2_xop_unlink_t *xop;
1540 	hammer2_inode_t *dip;
1541 	hammer2_inode_t *ip;
1542 	struct namecache *ncp;
1543 	int error;
1544 	int isopen;
1545 
1546 	LOCKSTART;
1547 	dip = VTOI(ap->a_dvp);
1548 	if (dip->pmp->ronly) {
1549 		LOCKSTOP;
1550 		return(EROFS);
1551 	}
1552 
1553 	ncp = ap->a_nch->ncp;
1554 
1555 	hammer2_pfs_memory_wait(dip->pmp);
1556 	hammer2_trans_init(dip->pmp, 0);
1557 	hammer2_inode_lock(dip, 0);
1558 
1559 	/*
1560 	 * The unlink XOP unlinks the path from the directory and
1561 	 * locates and returns the cluster associated with the real inode.
1562 	 * We have to handle nlinks here on the frontend.
1563 	 */
1564 	xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1565 	hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1566 	isopen = cache_isopen(ap->a_nch);
1567 	xop->isdir = 0;
1568 	xop->dopermanent = isopen ?  0 : HAMMER2_DELETE_PERMANENT;
1569 	hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1570 
1571 	/*
1572 	 * Collect the real inode and adjust nlinks, destroy the real
1573 	 * inode if nlinks transitions to 0 and it was the real inode
1574 	 * (else it has already been removed).
1575 	 */
1576 	error = hammer2_xop_collect(&xop->head, 0);
1577 	hammer2_inode_unlock(dip);
1578 
1579 	if (error == 0) {
1580 		ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1581 		hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1582 		if (ip) {
1583 			hammer2_inode_unlink_finisher(ip, isopen);
1584 			hammer2_inode_unlock(ip);
1585 		}
1586 	} else {
1587 		hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1588 	}
1589 
1590 	hammer2_inode_run_unlinkq(dip->pmp);
1591 	hammer2_trans_done(dip->pmp);
1592 	if (error == 0)
1593 		cache_unlink(ap->a_nch);
1594 	LOCKSTOP;
1595 	return (error);
1596 }
1597 
1598 /*
1599  * hammer2_vop_nrmdir { nch, dvp, cred }
1600  */
1601 static
1602 int
1603 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
1604 {
1605 	hammer2_xop_unlink_t *xop;
1606 	hammer2_inode_t *dip;
1607 	hammer2_inode_t *ip;
1608 	struct namecache *ncp;
1609 	int isopen;
1610 	int error;
1611 
1612 	LOCKSTART;
1613 	dip = VTOI(ap->a_dvp);
1614 	if (dip->pmp->ronly) {
1615 		LOCKSTOP;
1616 		return(EROFS);
1617 	}
1618 
1619 	hammer2_pfs_memory_wait(dip->pmp);
1620 	hammer2_trans_init(dip->pmp, 0);
1621 	hammer2_inode_lock(dip, 0);
1622 
1623 	xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1624 
1625 	ncp = ap->a_nch->ncp;
1626 	hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1627 	isopen = cache_isopen(ap->a_nch);
1628 	xop->isdir = 1;
1629 	xop->dopermanent = isopen ?  0 : HAMMER2_DELETE_PERMANENT;
1630 	hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1631 
1632 	/*
1633 	 * Collect the real inode and adjust nlinks, destroy the real
1634 	 * inode if nlinks transitions to 0 and it was the real inode
1635 	 * (else it has already been removed).
1636 	 */
1637 	error = hammer2_xop_collect(&xop->head, 0);
1638 	hammer2_inode_unlock(dip);
1639 
1640 	if (error == 0) {
1641 		ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1642 		hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1643 		if (ip) {
1644 			hammer2_inode_unlink_finisher(ip, isopen);
1645 			hammer2_inode_unlock(ip);
1646 		}
1647 	} else {
1648 		hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1649 	}
1650 	hammer2_inode_run_unlinkq(dip->pmp);
1651 	hammer2_trans_done(dip->pmp);
1652 	if (error == 0)
1653 		cache_unlink(ap->a_nch);
1654 	LOCKSTOP;
1655 	return (error);
1656 }
1657 
1658 /*
1659  * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1660  */
1661 static
1662 int
1663 hammer2_vop_nrename(struct vop_nrename_args *ap)
1664 {
1665 	struct namecache *fncp;
1666 	struct namecache *tncp;
1667 	hammer2_inode_t *cdip;
1668 	hammer2_inode_t *fdip;
1669 	hammer2_inode_t *tdip;
1670 	hammer2_inode_t *ip;
1671 	const uint8_t *fname;
1672 	size_t fname_len;
1673 	const uint8_t *tname;
1674 	size_t tname_len;
1675 	int error;
1676 	int tnch_error;
1677 	hammer2_key_t tlhc;
1678 
1679 	if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
1680 		return(EXDEV);
1681 	if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
1682 		return(EXDEV);
1683 
1684 	fdip = VTOI(ap->a_fdvp);	/* source directory */
1685 	tdip = VTOI(ap->a_tdvp);	/* target directory */
1686 
1687 	if (fdip->pmp->ronly)
1688 		return(EROFS);
1689 
1690 	LOCKSTART;
1691 	fncp = ap->a_fnch->ncp;		/* entry name in source */
1692 	fname = fncp->nc_name;
1693 	fname_len = fncp->nc_nlen;
1694 
1695 	tncp = ap->a_tnch->ncp;		/* entry name in target */
1696 	tname = tncp->nc_name;
1697 	tname_len = tncp->nc_nlen;
1698 
1699 	hammer2_pfs_memory_wait(tdip->pmp);
1700 	hammer2_trans_init(tdip->pmp, 0);
1701 
1702 	/*
1703 	 * ip is the inode being renamed.  If this is a hardlink then
1704 	 * ip represents the actual file and not the hardlink marker.
1705 	 */
1706 	ip = VTOI(fncp->nc_vp);
1707 
1708 	/*
1709 	 * The common parent directory must be locked first to avoid deadlocks.
1710 	 * Also note that fdip and/or tdip might match cdip.
1711 	 */
1712 	cdip = hammer2_inode_common_parent(ip->pip, tdip);
1713 	hammer2_inode_lock(cdip, 0);
1714 	hammer2_inode_lock(fdip, 0);
1715 	hammer2_inode_lock(tdip, 0);
1716 	hammer2_inode_ref(ip);		/* extra ref */
1717 	error = 0;
1718 
1719 	/*
1720 	 * If ip is a hardlink target and fdip != cdip we must shift the
1721 	 * inode to cdip.
1722 	 */
1723 	if (fdip != cdip &&
1724 	    (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0) {
1725 		hammer2_xop_nlink_t *xop1;
1726 
1727 		xop1 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
1728 		hammer2_xop_setip2(&xop1->head, ip);
1729 		hammer2_xop_setip3(&xop1->head, cdip);
1730 
1731 		hammer2_xop_start(&xop1->head, hammer2_xop_nlink);
1732 		error = hammer2_xop_collect(&xop1->head, 0);
1733 		hammer2_xop_retire(&xop1->head, HAMMER2_XOPMASK_VOP);
1734 	}
1735 
1736 	/*
1737 	 * Delete the target namespace.
1738 	 */
1739 	{
1740 		hammer2_xop_unlink_t *xop2;
1741 		hammer2_inode_t *tip;
1742 		int isopen;
1743 
1744 		/*
1745 		 * The unlink XOP unlinks the path from the directory and
1746 		 * locates and returns the cluster associated with the real
1747 		 * inode.  We have to handle nlinks here on the frontend.
1748 		 */
1749 		xop2 = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
1750 		hammer2_xop_setname(&xop2->head, tname, tname_len);
1751 		isopen = cache_isopen(ap->a_tnch);
1752 		xop2->isdir = -1;
1753 		xop2->dopermanent = isopen ?  0 : HAMMER2_DELETE_PERMANENT;
1754 		hammer2_xop_start(&xop2->head, hammer2_xop_unlink);
1755 
1756 		/*
1757 		 * Collect the real inode and adjust nlinks, destroy the real
1758 		 * inode if nlinks transitions to 0 and it was the real inode
1759 		 * (else it has already been removed).
1760 		 */
1761 		tnch_error = hammer2_xop_collect(&xop2->head, 0);
1762 		/* hammer2_inode_unlock(tdip); */
1763 
1764 		if (tnch_error == 0) {
1765 			tip = hammer2_inode_get(tdip->pmp, NULL,
1766 						&xop2->head.cluster, -1);
1767 			hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
1768 			if (tip) {
1769 				hammer2_inode_unlink_finisher(tip, isopen);
1770 				hammer2_inode_unlock(tip);
1771 			}
1772 		} else {
1773 			hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
1774 		}
1775 		/* hammer2_inode_lock(tdip, 0); */
1776 
1777 		if (tnch_error && tnch_error != ENOENT) {
1778 			error = tnch_error;
1779 			goto done2;
1780 		}
1781 	}
1782 
1783 	/*
1784 	 * Resolve the collision space for (tdip, tname, tname_len)
1785 	 *
1786 	 * tdip must be held exclusively locked to prevent races.
1787 	 */
1788 	{
1789 		hammer2_xop_scanlhc_t *sxop;
1790 		hammer2_tid_t lhcbase;
1791 
1792 		tlhc = hammer2_dirhash(tname, tname_len);
1793 		lhcbase = tlhc;
1794 		sxop = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
1795 		sxop->lhc = tlhc;
1796 		hammer2_xop_start(&sxop->head, hammer2_xop_scanlhc);
1797 		while ((error = hammer2_xop_collect(&sxop->head, 0)) == 0) {
1798 			if (tlhc != sxop->head.cluster.focus->bref.key)
1799 				break;
1800 			++tlhc;
1801 		}
1802 		hammer2_xop_retire(&sxop->head, HAMMER2_XOPMASK_VOP);
1803 
1804 		if (error) {
1805 			if (error != ENOENT)
1806 				goto done2;
1807 			++tlhc;
1808 			error = 0;
1809 		}
1810 		if ((lhcbase ^ tlhc) & ~HAMMER2_DIRHASH_LOMASK) {
1811 			error = ENOSPC;
1812 			goto done2;
1813 		}
1814 	}
1815 
1816 	/*
1817 	 * Everything is setup, do the rename.
1818 	 *
1819 	 * We have to synchronize ip->meta to the underlying operation.
1820 	 *
1821 	 * NOTE: To avoid deadlocks we cannot lock (ip) while we are
1822 	 *	 unlinking elements from their directories.  Locking
1823 	 *	 the nlinks field does not lock the whole inode.
1824 	 */
1825 	hammer2_inode_lock(ip, 0);
1826 	if (error == 0) {
1827 		hammer2_xop_nrename_t *xop4;
1828 
1829 		xop4 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
1830 		xop4->lhc = tlhc;
1831 		xop4->ip_key = ip->meta.name_key;
1832 		hammer2_xop_setip2(&xop4->head, ip);
1833 		hammer2_xop_setip3(&xop4->head, tdip);
1834 		hammer2_xop_setname(&xop4->head, fname, fname_len);
1835 		hammer2_xop_setname2(&xop4->head, tname, tname_len);
1836 		hammer2_xop_start(&xop4->head, hammer2_xop_nrename);
1837 
1838 		error = hammer2_xop_collect(&xop4->head, 0);
1839 		hammer2_xop_retire(&xop4->head, HAMMER2_XOPMASK_VOP);
1840 
1841 		if (error == ENOENT)
1842 			error = 0;
1843 		if (error == 0 &&
1844 		    (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
1845 			hammer2_inode_modify(ip);
1846 			ip->meta.name_len = tname_len;
1847 			ip->meta.name_key = tlhc;
1848 
1849 		}
1850 	}
1851 
1852 	/*
1853 	 * Fixup ip->pip if we were renaming the actual file and not a
1854 	 * hardlink pointer.
1855 	 */
1856 	if (error == 0 && (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
1857 		hammer2_inode_t *opip;
1858 
1859 		if (ip->pip != tdip) {
1860 			hammer2_inode_ref(tdip);
1861 			opip = ip->pip;
1862 			ip->pip = tdip;
1863 			if (opip)
1864 				hammer2_inode_drop(opip);
1865 		}
1866 	}
1867 	hammer2_inode_unlock(ip);
1868 done2:
1869 	hammer2_inode_unlock(tdip);
1870 	hammer2_inode_unlock(fdip);
1871 	hammer2_inode_unlock(cdip);
1872 	hammer2_inode_drop(ip);
1873 	hammer2_inode_drop(cdip);
1874 	hammer2_inode_run_unlinkq(fdip->pmp);
1875 	hammer2_trans_done(tdip->pmp);
1876 
1877 	/*
1878 	 * Issue the namecache update after unlocking all the internal
1879 	 * hammer structures, otherwise we might deadlock.
1880 	 */
1881 	if (tnch_error == 0) {
1882 		cache_unlink(ap->a_tnch);
1883 		cache_setunresolved(ap->a_tnch);
1884 	}
1885 	if (error == 0)
1886 		cache_rename(ap->a_fnch, ap->a_tnch);
1887 
1888 	LOCKSTOP;
1889 	return (error);
1890 }
1891 
1892 /*
1893  * hammer2_vop_ioctl { vp, command, data, fflag, cred }
1894  */
1895 static
1896 int
1897 hammer2_vop_ioctl(struct vop_ioctl_args *ap)
1898 {
1899 	hammer2_inode_t *ip;
1900 	int error;
1901 
1902 	LOCKSTART;
1903 	ip = VTOI(ap->a_vp);
1904 
1905 	error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
1906 			      ap->a_fflag, ap->a_cred);
1907 	LOCKSTOP;
1908 	return (error);
1909 }
1910 
1911 static
1912 int
1913 hammer2_vop_mountctl(struct vop_mountctl_args *ap)
1914 {
1915 	struct mount *mp;
1916 	hammer2_pfs_t *pmp;
1917 	int rc;
1918 
1919 	LOCKSTART;
1920 	switch (ap->a_op) {
1921 	case (MOUNTCTL_SET_EXPORT):
1922 		mp = ap->a_head.a_ops->head.vv_mount;
1923 		pmp = MPTOPMP(mp);
1924 
1925 		if (ap->a_ctllen != sizeof(struct export_args))
1926 			rc = (EINVAL);
1927 		else
1928 			rc = vfs_export(mp, &pmp->export,
1929 					(const struct export_args *)ap->a_ctl);
1930 		break;
1931 	default:
1932 		rc = vop_stdmountctl(ap);
1933 		break;
1934 	}
1935 	LOCKSTOP;
1936 	return (rc);
1937 }
1938 
1939 /*
1940  * KQFILTER
1941  */
1942 static void filt_hammer2detach(struct knote *kn);
1943 static int filt_hammer2read(struct knote *kn, long hint);
1944 static int filt_hammer2write(struct knote *kn, long hint);
1945 static int filt_hammer2vnode(struct knote *kn, long hint);
1946 
1947 static struct filterops hammer2read_filtops =
1948 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
1949 	  NULL, filt_hammer2detach, filt_hammer2read };
1950 static struct filterops hammer2write_filtops =
1951 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
1952 	  NULL, filt_hammer2detach, filt_hammer2write };
1953 static struct filterops hammer2vnode_filtops =
1954 	{ FILTEROP_ISFD | FILTEROP_MPSAFE,
1955 	  NULL, filt_hammer2detach, filt_hammer2vnode };
1956 
1957 static
1958 int
1959 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap)
1960 {
1961 	struct vnode *vp = ap->a_vp;
1962 	struct knote *kn = ap->a_kn;
1963 
1964 	switch (kn->kn_filter) {
1965 	case EVFILT_READ:
1966 		kn->kn_fop = &hammer2read_filtops;
1967 		break;
1968 	case EVFILT_WRITE:
1969 		kn->kn_fop = &hammer2write_filtops;
1970 		break;
1971 	case EVFILT_VNODE:
1972 		kn->kn_fop = &hammer2vnode_filtops;
1973 		break;
1974 	default:
1975 		return (EOPNOTSUPP);
1976 	}
1977 
1978 	kn->kn_hook = (caddr_t)vp;
1979 
1980 	knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1981 
1982 	return(0);
1983 }
1984 
1985 static void
1986 filt_hammer2detach(struct knote *kn)
1987 {
1988 	struct vnode *vp = (void *)kn->kn_hook;
1989 
1990 	knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1991 }
1992 
1993 static int
1994 filt_hammer2read(struct knote *kn, long hint)
1995 {
1996 	struct vnode *vp = (void *)kn->kn_hook;
1997 	hammer2_inode_t *ip = VTOI(vp);
1998 	off_t off;
1999 
2000 	if (hint == NOTE_REVOKE) {
2001 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2002 		return(1);
2003 	}
2004 	off = ip->meta.size - kn->kn_fp->f_offset;
2005 	kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2006 	if (kn->kn_sfflags & NOTE_OLDAPI)
2007 		return(1);
2008 	return (kn->kn_data != 0);
2009 }
2010 
2011 
2012 static int
2013 filt_hammer2write(struct knote *kn, long hint)
2014 {
2015 	if (hint == NOTE_REVOKE)
2016 		kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2017 	kn->kn_data = 0;
2018 	return (1);
2019 }
2020 
2021 static int
2022 filt_hammer2vnode(struct knote *kn, long hint)
2023 {
2024 	if (kn->kn_sfflags & hint)
2025 		kn->kn_fflags |= hint;
2026 	if (hint == NOTE_REVOKE) {
2027 		kn->kn_flags |= (EV_EOF | EV_NODATA);
2028 		return (1);
2029 	}
2030 	return (kn->kn_fflags != 0);
2031 }
2032 
2033 /*
2034  * FIFO VOPS
2035  */
2036 static
2037 int
2038 hammer2_vop_markatime(struct vop_markatime_args *ap)
2039 {
2040 	hammer2_inode_t *ip;
2041 	struct vnode *vp;
2042 
2043 	vp = ap->a_vp;
2044 	ip = VTOI(vp);
2045 
2046 	if (ip->pmp->ronly)
2047 		return(EROFS);
2048 	return(0);
2049 }
2050 
2051 static
2052 int
2053 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap)
2054 {
2055 	int error;
2056 
2057 	error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2058 	if (error)
2059 		error = hammer2_vop_kqfilter(ap);
2060 	return(error);
2061 }
2062 
2063 /*
2064  * VOPS vector
2065  */
2066 struct vop_ops hammer2_vnode_vops = {
2067 	.vop_default	= vop_defaultop,
2068 	.vop_fsync	= hammer2_vop_fsync,
2069 	.vop_getpages	= vop_stdgetpages,
2070 	.vop_putpages	= vop_stdputpages,
2071 	.vop_access	= hammer2_vop_access,
2072 	.vop_advlock	= hammer2_vop_advlock,
2073 	.vop_close	= hammer2_vop_close,
2074 	.vop_nlink	= hammer2_vop_nlink,
2075 	.vop_ncreate	= hammer2_vop_ncreate,
2076 	.vop_nsymlink	= hammer2_vop_nsymlink,
2077 	.vop_nremove	= hammer2_vop_nremove,
2078 	.vop_nrmdir	= hammer2_vop_nrmdir,
2079 	.vop_nrename	= hammer2_vop_nrename,
2080 	.vop_getattr	= hammer2_vop_getattr,
2081 	.vop_setattr	= hammer2_vop_setattr,
2082 	.vop_readdir	= hammer2_vop_readdir,
2083 	.vop_readlink	= hammer2_vop_readlink,
2084 	.vop_getpages	= vop_stdgetpages,
2085 	.vop_putpages	= vop_stdputpages,
2086 	.vop_read	= hammer2_vop_read,
2087 	.vop_write	= hammer2_vop_write,
2088 	.vop_open	= hammer2_vop_open,
2089 	.vop_inactive	= hammer2_vop_inactive,
2090 	.vop_reclaim 	= hammer2_vop_reclaim,
2091 	.vop_nresolve	= hammer2_vop_nresolve,
2092 	.vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
2093 	.vop_nmkdir 	= hammer2_vop_nmkdir,
2094 	.vop_nmknod 	= hammer2_vop_nmknod,
2095 	.vop_ioctl	= hammer2_vop_ioctl,
2096 	.vop_mountctl	= hammer2_vop_mountctl,
2097 	.vop_bmap	= hammer2_vop_bmap,
2098 	.vop_strategy	= hammer2_vop_strategy,
2099         .vop_kqfilter	= hammer2_vop_kqfilter
2100 };
2101 
2102 struct vop_ops hammer2_spec_vops = {
2103         .vop_default =          vop_defaultop,
2104         .vop_fsync =            hammer2_vop_fsync,
2105         .vop_read =             vop_stdnoread,
2106         .vop_write =            vop_stdnowrite,
2107         .vop_access =           hammer2_vop_access,
2108         .vop_close =            hammer2_vop_close,
2109         .vop_markatime =        hammer2_vop_markatime,
2110         .vop_getattr =          hammer2_vop_getattr,
2111         .vop_inactive =         hammer2_vop_inactive,
2112         .vop_reclaim =          hammer2_vop_reclaim,
2113         .vop_setattr =          hammer2_vop_setattr
2114 };
2115 
2116 struct vop_ops hammer2_fifo_vops = {
2117         .vop_default =          fifo_vnoperate,
2118         .vop_fsync =            hammer2_vop_fsync,
2119 #if 0
2120         .vop_read =             hammer2_vop_fiforead,
2121         .vop_write =            hammer2_vop_fifowrite,
2122 #endif
2123         .vop_access =           hammer2_vop_access,
2124 #if 0
2125         .vop_close =            hammer2_vop_fifoclose,
2126 #endif
2127         .vop_markatime =        hammer2_vop_markatime,
2128         .vop_getattr =          hammer2_vop_getattr,
2129         .vop_inactive =         hammer2_vop_inactive,
2130         .vop_reclaim =          hammer2_vop_reclaim,
2131         .vop_setattr =          hammer2_vop_setattr,
2132         .vop_kqfilter =         hammer2_vop_fifokqfilter
2133 };
2134 
2135