xref: /dflybsd-src/sys/vfs/hammer2/hammer2_vnops.c (revision 7a27faded6bccbda68815bb82b3be97f731409b9)
1 /*
2  * Copyright (c) 2011-2013 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  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  *
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in
16  *    the documentation and/or other materials provided with the
17  *    distribution.
18  * 3. Neither the name of The DragonFly Project nor the names of its
19  *    contributors may be used to endorse or promote products derived
20  *    from this software without specific, prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
25  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
26  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
27  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
28  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
29  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
30  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
31  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
32  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  */
35 /*
36  * Kernel Filesystem interface
37  *
38  * NOTE! local ipdata pointers must be reloaded on any modifying operation
39  *	 to the inode as its underlying chain may have changed.
40  */
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/kernel.h>
45 #include <sys/fcntl.h>
46 #include <sys/buf.h>
47 #include <sys/proc.h>
48 #include <sys/namei.h>
49 #include <sys/mount.h>
50 #include <sys/vnode.h>
51 #include <sys/mountctl.h>
52 #include <sys/dirent.h>
53 #include <sys/uio.h>
54 
55 #include "hammer2.h"
56 
57 #define ZFOFFSET	(-2LL)
58 
59 static int hammer2_read_file(hammer2_inode_t *ip, struct uio *uio,
60 				int seqcount);
61 static int hammer2_write_file(hammer2_trans_t *trans, hammer2_inode_t *ip,
62 				hammer2_chain_t **parentp,
63 				struct uio *uio, int ioflag, int seqcount);
64 static void hammer2_write_bp(hammer2_chain_t *chain, struct buf *bp,
65 				int ioflag);
66 static hammer2_chain_t *hammer2_assign_physical(hammer2_trans_t *trans,
67 				hammer2_inode_t *ip, hammer2_chain_t **parentp,
68 				hammer2_key_t lbase, int lblksize,
69 				int *errorp);
70 static void hammer2_extend_file(hammer2_trans_t *trans, hammer2_inode_t *ip,
71 				hammer2_chain_t **parentp, hammer2_key_t nsize);
72 static void hammer2_truncate_file(hammer2_trans_t *trans, hammer2_inode_t *ip,
73 				hammer2_chain_t **parentp, hammer2_key_t nsize);
74 
75 static __inline
76 void
77 hammer2_knote(struct vnode *vp, int flags)
78 {
79 	if (flags)
80 		KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
81 }
82 
83 /*
84  * Last reference to a vnode is going away but it is still cached.
85  */
86 static
87 int
88 hammer2_vop_inactive(struct vop_inactive_args *ap)
89 {
90 	hammer2_inode_t *ip;
91 	hammer2_chain_t *parent;
92 	struct vnode *vp;
93 
94 	vp = ap->a_vp;
95 	ip = VTOI(vp);
96 
97 	/*
98 	 * Degenerate case
99 	 */
100 	if (ip == NULL) {
101 		vrecycle(vp);
102 		return (0);
103 	}
104 
105 	/*
106 	 * Detect updates to the embedded data which may be synchronized by
107 	 * the strategy code.  Simply mark the inode modified so it gets
108 	 * picked up by our normal flush.
109 	 */
110 	parent = hammer2_inode_lock_ex(ip);
111 	KKASSERT(parent);
112 
113 	/*
114 	 * Check for deleted inodes and recycle immediately.
115 	 */
116 	if (parent->flags & HAMMER2_CHAIN_DELETED) {
117 		hammer2_inode_unlock_ex(ip, parent);
118 		vrecycle(vp);
119 	} else {
120 		hammer2_inode_unlock_ex(ip, parent);
121 	}
122 	return (0);
123 }
124 
125 /*
126  * Reclaim a vnode so that it can be reused; after the inode is
127  * disassociated, the filesystem must manage it alone.
128  */
129 static
130 int
131 hammer2_vop_reclaim(struct vop_reclaim_args *ap)
132 {
133 	hammer2_chain_t *chain;
134 	hammer2_inode_t *ip;
135 #if 0
136 	hammer2_trans_t trans;
137 #endif
138 	struct vnode *vp;
139 
140 	vp = ap->a_vp;
141 	ip = VTOI(vp);
142 	if (ip == NULL)
143 		return(0);
144 
145 	/*
146 	 * Set SUBMODIFIED so we can detect and propagate the DESTROYED
147 	 * bit in the flush code.
148 	 *
149 	 * ip->chain might be stale, correct it before checking as older
150 	 * versions of the chain are likely marked deleted even if the
151 	 * file hasn't been.  XXX ip->chain should never be stale on
152 	 * reclaim.
153 	 */
154 	chain = hammer2_inode_lock_ex(ip);
155 #if 0
156 	if (chain->next_parent)
157 		kprintf("RECLAIM DUPLINKED IP: %p ip->ch=%p ch=%p np=%p\n",
158 			ip, ip->chain, chain, chain->next_parent);
159 #endif
160 
161 	/*
162 	 * The final close of a deleted file or directory marks it for
163 	 * destruction.  The DESTROYED flag allows the flusher to shortcut
164 	 * any modified blocks still unflushed (that is, just ignore them).
165 	 *
166 	 * HAMMER2 usually does not try to optimize the freemap by returning
167 	 * deleted blocks to it as it does not usually know how many snapshots
168 	 * might be referencing portions of the file/dir.  XXX TODO.
169 	 *
170 	 * XXX TODO - However, any modified file as-of when a snapshot is made
171 	 *	      cannot use this optimization as some of the modifications
172 	 *	      may wind up being part of the snapshot.
173 	 */
174 	vp->v_data = NULL;
175 	ip->vp = NULL;
176 	if (chain->flags & HAMMER2_CHAIN_DELETED) {
177 		KKASSERT(chain->flags & HAMMER2_CHAIN_DELETED);
178 		atomic_set_int(&chain->flags, HAMMER2_CHAIN_DESTROYED |
179 					      HAMMER2_CHAIN_SUBMODIFIED);
180 	}
181 #if 0
182 	/*
183 	 * XXX chains will be flushed on sync, no need to do it here.
184 	 */
185 	if (chain->flags & (HAMMER2_CHAIN_MODIFIED |
186 			    HAMMER2_CHAIN_DELETED |
187 			    HAMMER2_CHAIN_SUBMODIFIED)) {
188 		hammer2_trans_init(&trans, ip->pmp, HAMMER2_TRANS_ISFLUSH);
189 		hammer2_chain_flush(&trans, chain);
190 		hammer2_trans_done(&trans);
191 	}
192 #endif
193 	hammer2_inode_unlock_ex(ip, chain);		/* unlock */
194 	hammer2_inode_drop(ip);				/* vp ref */
195 	/* chain no longer referenced */
196 	/* chain = NULL; not needed */
197 
198 	/*
199 	 * XXX handle background sync when ip dirty, kernel will no longer
200 	 * notify us regarding this inode because there is no longer a
201 	 * vnode attached to it.
202 	 */
203 
204 	return (0);
205 }
206 
207 static
208 int
209 hammer2_vop_fsync(struct vop_fsync_args *ap)
210 {
211 	hammer2_inode_t *ip;
212 	hammer2_trans_t trans;
213 	hammer2_chain_t *chain;
214 	struct vnode *vp;
215 
216 	vp = ap->a_vp;
217 	ip = VTOI(vp);
218 
219 	hammer2_trans_init(&trans, ip->pmp, HAMMER2_TRANS_ISFLUSH);
220 	chain = hammer2_inode_lock_ex(ip);
221 
222 	vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
223 
224 	/*
225 	 * Calling chain_flush here creates a lot of duplicative
226 	 * COW operations due to non-optimal vnode ordering.
227 	 *
228 	 * Only do it for an actual fsync() syscall.  The other forms
229 	 * which call this function will eventually call chain_flush
230 	 * on the volume root as a catch-all, which is far more optimal.
231 	 */
232 	atomic_clear_int(&ip->flags, HAMMER2_INODE_MODIFIED);
233 	if (ap->a_flags & VOP_FSYNC_SYSCALL) {
234 		hammer2_chain_flush(&trans, chain);
235 	}
236 	hammer2_inode_unlock_ex(ip, chain);
237 	hammer2_trans_done(&trans);
238 
239 	return (0);
240 }
241 
242 static
243 int
244 hammer2_vop_access(struct vop_access_args *ap)
245 {
246 	hammer2_inode_t *ip = VTOI(ap->a_vp);
247 	hammer2_inode_data_t *ipdata;
248 	hammer2_chain_t *chain;
249 	uid_t uid;
250 	gid_t gid;
251 	int error;
252 
253 	chain = hammer2_inode_lock_sh(ip);
254 	ipdata = &chain->data->ipdata;
255 	uid = hammer2_to_unix_xid(&ipdata->uid);
256 	gid = hammer2_to_unix_xid(&ipdata->gid);
257 	error = vop_helper_access(ap, uid, gid, ipdata->mode, ipdata->uflags);
258 	hammer2_inode_unlock_sh(ip, chain);
259 
260 	return (error);
261 }
262 
263 static
264 int
265 hammer2_vop_getattr(struct vop_getattr_args *ap)
266 {
267 	hammer2_inode_data_t *ipdata;
268 	hammer2_chain_t *chain;
269 	hammer2_pfsmount_t *pmp;
270 	hammer2_inode_t *ip;
271 	struct vnode *vp;
272 	struct vattr *vap;
273 
274 	vp = ap->a_vp;
275 	vap = ap->a_vap;
276 
277 	ip = VTOI(vp);
278 	pmp = ip->pmp;
279 
280 	chain = hammer2_inode_lock_sh(ip);
281 	ipdata = &chain->data->ipdata;
282 
283 	vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0];
284 	vap->va_fileid = ipdata->inum;
285 	vap->va_mode = ipdata->mode;
286 	vap->va_nlink = ipdata->nlinks;
287 	vap->va_uid = hammer2_to_unix_xid(&ipdata->uid);
288 	vap->va_gid = hammer2_to_unix_xid(&ipdata->gid);
289 	vap->va_rmajor = 0;
290 	vap->va_rminor = 0;
291 	vap->va_size = ipdata->size;
292 	vap->va_blocksize = HAMMER2_PBUFSIZE;
293 	vap->va_flags = ipdata->uflags;
294 	hammer2_time_to_timespec(ipdata->ctime, &vap->va_ctime);
295 	hammer2_time_to_timespec(ipdata->mtime, &vap->va_mtime);
296 	hammer2_time_to_timespec(ipdata->mtime, &vap->va_atime);
297 	vap->va_gen = 1;
298 	vap->va_bytes = vap->va_size;	/* XXX */
299 	vap->va_type = hammer2_get_vtype(chain);
300 	vap->va_filerev = 0;
301 	vap->va_uid_uuid = ipdata->uid;
302 	vap->va_gid_uuid = ipdata->gid;
303 	vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
304 			  VA_FSID_UUID_VALID;
305 
306 	hammer2_inode_unlock_sh(ip, chain);
307 
308 	return (0);
309 }
310 
311 static
312 int
313 hammer2_vop_setattr(struct vop_setattr_args *ap)
314 {
315 	hammer2_inode_data_t *ipdata;
316 	hammer2_inode_t *ip;
317 	hammer2_chain_t *chain;
318 	hammer2_trans_t trans;
319 	struct vnode *vp;
320 	struct vattr *vap;
321 	int error;
322 	int kflags = 0;
323 	int domtime = 0;
324 	uint64_t ctime;
325 
326 	vp = ap->a_vp;
327 	vap = ap->a_vap;
328 	hammer2_update_time(&ctime);
329 
330 	ip = VTOI(vp);
331 
332 	if (ip->pmp->ronly)
333 		return(EROFS);
334 
335 	hammer2_trans_init(&trans, ip->pmp, 0);
336 	chain = hammer2_inode_lock_ex(ip);
337 	ipdata = &chain->data->ipdata;
338 	error = 0;
339 
340 	if (vap->va_flags != VNOVAL) {
341 		u_int32_t flags;
342 
343 		flags = ipdata->uflags;
344 		error = vop_helper_setattr_flags(&flags, vap->va_flags,
345 					 hammer2_to_unix_xid(&ipdata->uid),
346 					 ap->a_cred);
347 		if (error == 0) {
348 			if (ipdata->uflags != flags) {
349 				ipdata = hammer2_chain_modify_ip(&trans, ip,
350 								 &chain, 0);
351 				ipdata->uflags = flags;
352 				ipdata->ctime = ctime;
353 				kflags |= NOTE_ATTRIB;
354 			}
355 			if (ipdata->uflags & (IMMUTABLE | APPEND)) {
356 				error = 0;
357 				goto done;
358 			}
359 		}
360 		goto done;
361 	}
362 	if (ipdata->uflags & (IMMUTABLE | APPEND)) {
363 		error = EPERM;
364 		goto done;
365 	}
366 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
367 		mode_t cur_mode = ipdata->mode;
368 		uid_t cur_uid = hammer2_to_unix_xid(&ipdata->uid);
369 		gid_t cur_gid = hammer2_to_unix_xid(&ipdata->gid);
370 		uuid_t uuid_uid;
371 		uuid_t uuid_gid;
372 
373 		error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
374 					 ap->a_cred,
375 					 &cur_uid, &cur_gid, &cur_mode);
376 		if (error == 0) {
377 			hammer2_guid_to_uuid(&uuid_uid, cur_uid);
378 			hammer2_guid_to_uuid(&uuid_gid, cur_gid);
379 			if (bcmp(&uuid_uid, &ipdata->uid, sizeof(uuid_uid)) ||
380 			    bcmp(&uuid_gid, &ipdata->gid, sizeof(uuid_gid)) ||
381 			    ipdata->mode != cur_mode
382 			) {
383 				ipdata = hammer2_chain_modify_ip(&trans, ip,
384 								 &chain, 0);
385 				ipdata->uid = uuid_uid;
386 				ipdata->gid = uuid_gid;
387 				ipdata->mode = cur_mode;
388 				ipdata->ctime = ctime;
389 			}
390 			kflags |= NOTE_ATTRIB;
391 		}
392 	}
393 
394 	/*
395 	 * Resize the file
396 	 */
397 	if (vap->va_size != VNOVAL && ipdata->size != vap->va_size) {
398 		switch(vp->v_type) {
399 		case VREG:
400 			if (vap->va_size == ipdata->size)
401 				break;
402 			if (vap->va_size < ipdata->size) {
403 				hammer2_truncate_file(&trans, ip,
404 						      &chain, vap->va_size);
405 			} else {
406 				hammer2_extend_file(&trans, ip,
407 						    &chain, vap->va_size);
408 			}
409 			ipdata = &chain->data->ipdata; /* RELOAD */
410 			domtime = 1;
411 			break;
412 		default:
413 			error = EINVAL;
414 			goto done;
415 		}
416 	}
417 #if 0
418 	/* atime not supported */
419 	if (vap->va_atime.tv_sec != VNOVAL) {
420 		ipdata = hammer2_chain_modify_ip(&trans, ip, &chain, 0);
421 		ipdata->atime = hammer2_timespec_to_time(&vap->va_atime);
422 		kflags |= NOTE_ATTRIB;
423 	}
424 #endif
425 	if (vap->va_mtime.tv_sec != VNOVAL) {
426 		ipdata = hammer2_chain_modify_ip(&trans, ip, &chain, 0);
427 		ipdata->mtime = hammer2_timespec_to_time(&vap->va_mtime);
428 		kflags |= NOTE_ATTRIB;
429 	}
430 	if (vap->va_mode != (mode_t)VNOVAL) {
431 		mode_t cur_mode = ipdata->mode;
432 		uid_t cur_uid = hammer2_to_unix_xid(&ipdata->uid);
433 		gid_t cur_gid = hammer2_to_unix_xid(&ipdata->gid);
434 
435 		error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
436 					 cur_uid, cur_gid, &cur_mode);
437 		if (error == 0 && ipdata->mode != cur_mode) {
438 			ipdata = hammer2_chain_modify_ip(&trans, ip, &chain, 0);
439 			ipdata->mode = cur_mode;
440 			ipdata->ctime = ctime;
441 			kflags |= NOTE_ATTRIB;
442 		}
443 	}
444 done:
445 	hammer2_inode_unlock_ex(ip, chain);
446 	hammer2_trans_done(&trans);
447 	return (error);
448 }
449 
450 static
451 int
452 hammer2_vop_readdir(struct vop_readdir_args *ap)
453 {
454 	hammer2_inode_data_t *ipdata;
455 	hammer2_inode_t *ip;
456 	hammer2_inode_t *xip;
457 	hammer2_chain_t *parent;
458 	hammer2_chain_t *chain;
459 	hammer2_chain_t *xchain;
460 	hammer2_tid_t inum;
461 	hammer2_key_t lkey;
462 	struct uio *uio;
463 	off_t *cookies;
464 	off_t saveoff;
465 	int cookie_index;
466 	int ncookies;
467 	int error;
468 	int dtype;
469 	int r;
470 
471 	ip = VTOI(ap->a_vp);
472 	uio = ap->a_uio;
473 	saveoff = uio->uio_offset;
474 
475 	/*
476 	 * Setup cookies directory entry cookies if requested
477 	 */
478 	if (ap->a_ncookies) {
479 		ncookies = uio->uio_resid / 16 + 1;
480 		if (ncookies > 1024)
481 			ncookies = 1024;
482 		cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
483 	} else {
484 		ncookies = -1;
485 		cookies = NULL;
486 	}
487 	cookie_index = 0;
488 
489 	parent = hammer2_inode_lock_sh(ip);
490 	ipdata = &parent->data->ipdata;
491 
492 	/*
493 	 * Handle artificial entries.  To ensure that only positive 64 bit
494 	 * quantities are returned to userland we always strip off bit 63.
495 	 * The hash code is designed such that codes 0x0000-0x7FFF are not
496 	 * used, allowing us to use these codes for articial entries.
497 	 *
498 	 * Entry 0 is used for '.' and entry 1 is used for '..'.  Do not
499 	 * allow '..' to cross the mount point into (e.g.) the super-root.
500 	 */
501 	error = 0;
502 	chain = (void *)(intptr_t)-1;	/* non-NULL for early goto done case */
503 
504 	if (saveoff == 0) {
505 		inum = ipdata->inum & HAMMER2_DIRHASH_USERMSK;
506 		r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
507 		if (r)
508 			goto done;
509 		if (cookies)
510 			cookies[cookie_index] = saveoff;
511 		++saveoff;
512 		++cookie_index;
513 		if (cookie_index == ncookies)
514 			goto done;
515 	}
516 
517 	if (saveoff == 1) {
518 		/*
519 		 * Be careful with lockorder when accessing ".."
520 		 *
521 		 * (ip is the current dir. xip is the parent dir).
522 		 */
523 		inum = ipdata->inum & HAMMER2_DIRHASH_USERMSK;
524 		while (ip->pip != NULL && ip != ip->pmp->iroot) {
525 			xip = ip->pip;
526 			hammer2_inode_ref(xip);
527 			hammer2_inode_unlock_sh(ip, parent);
528 			xchain = hammer2_inode_lock_sh(xip);
529 			parent = hammer2_inode_lock_sh(ip);
530 			hammer2_inode_drop(xip);
531 			if (xip == ip->pip) {
532 				inum = xchain->data->ipdata.inum &
533 				       HAMMER2_DIRHASH_USERMSK;
534 				hammer2_inode_unlock_sh(xip, xchain);
535 				break;
536 			}
537 			hammer2_inode_unlock_sh(xip, xchain);
538 		}
539 		r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
540 		if (r)
541 			goto done;
542 		if (cookies)
543 			cookies[cookie_index] = saveoff;
544 		++saveoff;
545 		++cookie_index;
546 		if (cookie_index == ncookies)
547 			goto done;
548 	}
549 
550 	lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
551 
552 	/*
553 	 * parent is the inode chain, already locked for us.  Don't
554 	 * double lock shared locks as this will screw up upgrades.
555 	 */
556 	if (error) {
557 		goto done;
558 	}
559 	chain = hammer2_chain_lookup(&parent, lkey, lkey,
560 				     HAMMER2_LOOKUP_SHARED);
561 	if (chain == NULL) {
562 		chain = hammer2_chain_lookup(&parent,
563 					     lkey, (hammer2_key_t)-1,
564 					     HAMMER2_LOOKUP_SHARED);
565 	}
566 	while (chain) {
567 		if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) {
568 			dtype = hammer2_get_dtype(chain);
569 			saveoff = chain->bref.key & HAMMER2_DIRHASH_USERMSK;
570 			r = vop_write_dirent(&error, uio,
571 					     chain->data->ipdata.inum &
572 					      HAMMER2_DIRHASH_USERMSK,
573 					     dtype,
574 					     chain->data->ipdata.name_len,
575 					     chain->data->ipdata.filename);
576 			if (r)
577 				break;
578 			if (cookies)
579 				cookies[cookie_index] = saveoff;
580 			++cookie_index;
581 		} else {
582 			/* XXX chain error */
583 			kprintf("bad chain type readdir %d\n",
584 				chain->bref.type);
585 		}
586 
587 		/*
588 		 * Keys may not be returned in order so once we have a
589 		 * placemarker (chain) the scan must allow the full range
590 		 * or some entries will be missed.
591 		 */
592 		chain = hammer2_chain_next(&parent, chain,
593 					   HAMMER2_DIRHASH_VISIBLE,
594 					   (hammer2_key_t)-1,
595 					   HAMMER2_LOOKUP_SHARED);
596 		if (chain) {
597 			saveoff = (chain->bref.key &
598 				   HAMMER2_DIRHASH_USERMSK) + 1;
599 		} else {
600 			saveoff = (hammer2_key_t)-1;
601 		}
602 		if (cookie_index == ncookies)
603 			break;
604 	}
605 	if (chain)
606 		hammer2_chain_unlock(chain);
607 done:
608 	hammer2_inode_unlock_sh(ip, parent);
609 	if (ap->a_eofflag)
610 		*ap->a_eofflag = (chain == NULL);
611 	uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
612 	if (error && cookie_index == 0) {
613 		if (cookies) {
614 			kfree(cookies, M_TEMP);
615 			*ap->a_ncookies = 0;
616 			*ap->a_cookies = NULL;
617 		}
618 	} else {
619 		if (cookies) {
620 			*ap->a_ncookies = cookie_index;
621 			*ap->a_cookies = cookies;
622 		}
623 	}
624 	return (error);
625 }
626 
627 /*
628  * hammer2_vop_readlink { vp, uio, cred }
629  */
630 static
631 int
632 hammer2_vop_readlink(struct vop_readlink_args *ap)
633 {
634 	struct vnode *vp;
635 	hammer2_inode_t *ip;
636 	int error;
637 
638 	vp = ap->a_vp;
639 	if (vp->v_type != VLNK)
640 		return (EINVAL);
641 	ip = VTOI(vp);
642 
643 	error = hammer2_read_file(ip, ap->a_uio, 0);
644 	return (error);
645 }
646 
647 static
648 int
649 hammer2_vop_read(struct vop_read_args *ap)
650 {
651 	struct vnode *vp;
652 	hammer2_inode_t *ip;
653 	struct uio *uio;
654 	int error;
655 	int seqcount;
656 	int bigread;
657 
658 	/*
659 	 * Read operations supported on this vnode?
660 	 */
661 	vp = ap->a_vp;
662 	if (vp->v_type != VREG)
663 		return (EINVAL);
664 
665 	/*
666 	 * Misc
667 	 */
668 	ip = VTOI(vp);
669 	uio = ap->a_uio;
670 	error = 0;
671 
672 	seqcount = ap->a_ioflag >> 16;
673 	bigread = (uio->uio_resid > 100 * 1024 * 1024);
674 
675 	error = hammer2_read_file(ip, uio, seqcount);
676 	return (error);
677 }
678 
679 static
680 int
681 hammer2_vop_write(struct vop_write_args *ap)
682 {
683 	hammer2_inode_t *ip;
684 	hammer2_trans_t trans;
685 	hammer2_chain_t *parent;
686 	thread_t td;
687 	struct vnode *vp;
688 	struct uio *uio;
689 	int error;
690 	int seqcount;
691 	int bigwrite;
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 	seqcount = ap->a_ioflag >> 16;
710 	bigwrite = (uio->uio_resid > 100 * 1024 * 1024);
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 	bigwrite = (uio->uio_resid > 100 * 1024 * 1024);
723 
724 	/*
725 	 * ip must be locked if extending the file.
726 	 * ip must be locked to avoid racing a truncation.
727 	 *
728 	 * ip must be marked modified, particularly because the write
729 	 * might wind up being copied into the embedded data area.
730 	 */
731 	hammer2_trans_init(&trans, ip->pmp, 0);
732 	parent = hammer2_inode_lock_ex(ip);
733 	error = hammer2_write_file(&trans, ip, &parent,
734 				   uio, ap->a_ioflag, seqcount);
735 	hammer2_inode_unlock_ex(ip, parent);
736 	hammer2_trans_done(&trans);
737 
738 	return (error);
739 }
740 
741 /*
742  * Perform read operations on a file or symlink given an UNLOCKED
743  * inode and uio.
744  *
745  * The passed ip is not locked.
746  */
747 static
748 int
749 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
750 {
751 	hammer2_off_t size;
752 	hammer2_chain_t *parent;
753 	struct buf *bp;
754 	int error;
755 
756 	error = 0;
757 
758 	/*
759 	 * UIO read loop.
760 	 */
761 	parent = hammer2_inode_lock_sh(ip);
762 	size = ip->chain->data->ipdata.size;
763 
764 	while (uio->uio_resid > 0 && uio->uio_offset < size) {
765 		hammer2_key_t lbase;
766 		hammer2_key_t leof;
767 		int lblksize;
768 		int loff;
769 		int n;
770 
771 		lblksize = hammer2_calc_logical(ip, uio->uio_offset,
772 						&lbase, &leof);
773 
774 		error = cluster_read(ip->vp, leof, lbase, lblksize,
775 				     uio->uio_resid, seqcount * BKVASIZE,
776 				     &bp);
777 
778 		if (error)
779 			break;
780 		loff = (int)(uio->uio_offset - lbase);
781 		n = lblksize - loff;
782 		if (n > uio->uio_resid)
783 			n = uio->uio_resid;
784 		if (n > size - uio->uio_offset)
785 			n = (int)(size - uio->uio_offset);
786 		bp->b_flags |= B_AGE;
787 		uiomove((char *)bp->b_data + loff, n, uio);
788 		bqrelse(bp);
789 	}
790 	hammer2_inode_unlock_sh(ip, parent);
791 	return (error);
792 }
793 
794 /*
795  * Called with a locked (ip) to do the underlying write to a file or
796  * to build the symlink target.
797  */
798 static
799 int
800 hammer2_write_file(hammer2_trans_t *trans, hammer2_inode_t *ip,
801 		   hammer2_chain_t **parentp,
802 		   struct uio *uio, int ioflag, int seqcount)
803 {
804 	hammer2_inode_data_t *ipdata;
805 	hammer2_key_t old_eof;
806 	struct buf *bp;
807 	int kflags;
808 	int error;
809 	int modified = 0;
810 
811 	/*
812 	 * Setup if append
813 	 */
814 	ipdata = hammer2_chain_modify_ip(trans, ip, parentp, 0);
815 	if (ioflag & IO_APPEND)
816 		uio->uio_offset = ipdata->size;
817 	kflags = 0;
818 	error = 0;
819 
820 	/*
821 	 * vfs_sync visibility.  Interlocked by the inode ex lock so we
822 	 * shouldn't have to reassert it multiple times if the ip->chain
823 	 * is modified/flushed multiple times during the write, except
824 	 * when we release/reacquire the inode ex lock.
825 	 */
826 	atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED);
827 
828 	/*
829 	 * Extend the file if necessary.  If the write fails at some point
830 	 * we will truncate it back down to cover as much as we were able
831 	 * to write.
832 	 *
833 	 * Doing this now makes it easier to calculate buffer sizes in
834 	 * the loop.
835 	 */
836 	KKASSERT(ipdata->type != HAMMER2_OBJTYPE_HARDLINK);
837 	old_eof = ipdata->size;
838 	if (uio->uio_offset + uio->uio_resid > ipdata->size) {
839 		modified = 1;
840 		hammer2_extend_file(trans, ip, parentp,
841 				    uio->uio_offset + uio->uio_resid);
842 		ipdata = &ip->chain->data->ipdata;	/* RELOAD */
843 		kflags |= NOTE_EXTEND;
844 	}
845 	KKASSERT(ipdata->type != HAMMER2_OBJTYPE_HARDLINK);
846 
847 	/*
848 	 * UIO write loop
849 	 */
850 	while (uio->uio_resid > 0) {
851 		hammer2_chain_t *chain;
852 		hammer2_key_t lbase;
853 		hammer2_key_t leof;
854 		int trivial;
855 		int lblksize;
856 		int loff;
857 		int n;
858 
859 		/*
860 		 * Don't allow the buffer build to blow out the buffer
861 		 * cache.
862 		 */
863 		if ((ioflag & IO_RECURSE) == 0) {
864 			/*
865 			 * XXX should try to leave this unlocked through
866 			 *	the whole loop
867 			 */
868 			hammer2_inode_unlock_ex(ip, *parentp);
869 			bwillwrite(HAMMER2_PBUFSIZE);
870 			*parentp = hammer2_inode_lock_ex(ip);
871 			atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED);
872 			ipdata = &ip->chain->data->ipdata;	/* reload */
873 		}
874 
875 		/* XXX bigwrite & signal check test */
876 
877 		/*
878 		 * This nominally tells us how much we can cluster and
879 		 * what the logical buffer size needs to be.  Currently
880 		 * we don't try to cluster the write and just handle one
881 		 * block at a time.
882 		 */
883 		lblksize = hammer2_calc_logical(ip, uio->uio_offset,
884 						&lbase, &leof);
885 		loff = (int)(uio->uio_offset - lbase);
886 
887 		/*
888 		 * Calculate bytes to copy this transfer and whether the
889 		 * copy completely covers the buffer or not.
890 		 */
891 		trivial = 0;
892 		n = lblksize - loff;
893 		if (n > uio->uio_resid) {
894 			n = uio->uio_resid;
895 			if (loff == lbase &&
896 			    uio->uio_offset + n == ipdata->size)
897 				trivial = 1;
898 		} else if (loff == 0) {
899 			trivial = 1;
900 		}
901 
902 		/*
903 		 * Get the buffer
904 		 */
905 		if (uio->uio_segflg == UIO_NOCOPY) {
906 			/*
907 			 * Issuing a write with the same data backing the
908 			 * buffer.  Instantiate the buffer to collect the
909 			 * backing vm pages, then read-in any missing bits.
910 			 *
911 			 * This case is used by vop_stdputpages().
912 			 */
913 			bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
914 			if ((bp->b_flags & B_CACHE) == 0) {
915 				bqrelse(bp);
916 				error = bread(ip->vp, lbase, lblksize, &bp);
917 			}
918 		} else if (trivial) {
919 			/*
920 			 * Even though we are entirely overwriting the buffer
921 			 * we may still have to zero it out to avoid a
922 			 * mmap/write visibility issue.
923 			 */
924 			bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
925 			if ((bp->b_flags & B_CACHE) == 0)
926 				vfs_bio_clrbuf(bp);
927 		} else {
928 			/*
929 			 * Partial overwrite, read in any missing bits then
930 			 * replace the portion being written.
931 			 *
932 			 * (The strategy code will detect zero-fill physical
933 			 * blocks for this case).
934 			 */
935 			error = bread(ip->vp, lbase, lblksize, &bp);
936 			if (error == 0)
937 				bheavy(bp);
938 		}
939 
940 		if (error) {
941 			brelse(bp);
942 			break;
943 		}
944 
945 		/*
946 		 * Ok, copy the data in
947 		 */
948 		hammer2_inode_unlock_ex(ip, *parentp);
949 		error = uiomove(bp->b_data + loff, n, uio);
950 		*parentp = hammer2_inode_lock_ex(ip);
951 		atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED);
952 		ipdata = &ip->chain->data->ipdata;	/* reload */
953 		kflags |= NOTE_WRITE;
954 		modified = 1;
955 		if (error) {
956 			brelse(bp);
957 			break;
958 		}
959 
960 		/*
961 		 * We have to assign physical storage to the buffer we intend
962 		 * to dirty or write now to avoid deadlocks in the strategy
963 		 * code later.
964 		 *
965 		 * This can return NOOFFSET for inode-embedded data.  The
966 		 * strategy code will take care of it in that case.
967 		 */
968 		chain = hammer2_assign_physical(trans, ip, parentp,
969 						lbase, lblksize, &error);
970 		ipdata = &ip->chain->data->ipdata;	/* RELOAD */
971 
972 		if (error) {
973 			KKASSERT(chain == NULL);
974 			brelse(bp);
975 			break;
976 		}
977 
978 		/* XXX update ip_data.mtime */
979 
980 		/*
981 		 * Once we dirty a buffer any cached offset becomes invalid.
982 		 *
983 		 * NOTE: For cluster_write() always use the trailing block
984 		 *	 size, which is HAMMER2_PBUFSIZE.  lblksize is the
985 		 *	 eof-straddling blocksize and is incorrect.
986 		 */
987 		bp->b_flags |= B_AGE;
988 		hammer2_write_bp(chain, bp, ioflag);
989 		hammer2_chain_unlock(chain);
990 	}
991 
992 	/*
993 	 * Cleanup.  If we extended the file EOF but failed to write through
994 	 * the entire write is a failure and we have to back-up.
995 	 */
996 	if (error && ipdata->size != old_eof) {
997 		hammer2_truncate_file(trans, ip, parentp, old_eof);
998 		ipdata = &ip->chain->data->ipdata;	/* RELOAD */
999 	} else if (modified) {
1000 		ipdata = hammer2_chain_modify_ip(trans, ip, parentp, 0);
1001 		hammer2_update_time(&ipdata->mtime);
1002 	}
1003 	hammer2_knote(ip->vp, kflags);
1004 
1005 	return error;
1006 }
1007 
1008 /*
1009  * Write the logical file bp out.
1010  */
1011 static
1012 void
1013 hammer2_write_bp(hammer2_chain_t *chain, struct buf *bp, int ioflag)
1014 {
1015 	hammer2_off_t pbase;
1016 	hammer2_off_t pmask;
1017 	hammer2_off_t peof;
1018 	struct buf *dbp;
1019 	size_t boff;
1020 	size_t psize;
1021 
1022 	KKASSERT(chain->flags & HAMMER2_CHAIN_MODIFIED);
1023 
1024 	switch(chain->bref.type) {
1025 	case HAMMER2_BREF_TYPE_INODE:
1026 		KKASSERT(chain->data->ipdata.op_flags &
1027 			 HAMMER2_OPFLAG_DIRECTDATA);
1028 		KKASSERT(bp->b_loffset == 0);
1029 		bcopy(bp->b_data, chain->data->ipdata.u.data,
1030 		      HAMMER2_EMBEDDED_BYTES);
1031 		break;
1032 	case HAMMER2_BREF_TYPE_DATA:
1033 		psize = hammer2_devblksize(chain->bytes);
1034 		pmask = (hammer2_off_t)psize - 1;
1035 		pbase = chain->bref.data_off & ~pmask;
1036 		boff = chain->bref.data_off & (HAMMER2_OFF_MASK & pmask);
1037 		peof = (pbase + HAMMER2_SEGMASK64) & ~HAMMER2_SEGMASK64;
1038 
1039 		dbp = getblk(chain->hmp->devvp, pbase, psize, 0, 0);
1040 		bcopy(bp->b_data, dbp->b_data + boff, chain->bytes);
1041 
1042 		if (ioflag & IO_SYNC) {
1043 			/*
1044 			 * Synchronous I/O requested.
1045 			 */
1046 			bwrite(dbp);
1047 		/*
1048 		} else if ((ioflag & IO_DIRECT) && loff + n == lblksize) {
1049 			bdwrite(dbp);
1050 		*/
1051 		} else if (ioflag & IO_ASYNC) {
1052 			bawrite(dbp);
1053 		} else if (hammer2_cluster_enable) {
1054 			cluster_write(dbp, peof, HAMMER2_PBUFSIZE, 4/*XXX*/);
1055 		} else {
1056 			bdwrite(dbp);
1057 		}
1058 		break;
1059 	default:
1060 		panic("hammer2_write_bp: bad chain type %d\n",
1061 		      chain->bref.type);
1062 		/* NOT REACHED */
1063 		break;
1064 	}
1065 	bqrelse(bp);
1066 }
1067 
1068 /*
1069  * Assign physical storage to a logical block.  This function creates the
1070  * related meta-data chains representing the data blocks and marks them
1071  * MODIFIED.  We could mark them MOVED instead but ultimately I need to
1072  * XXX code the flusher to check that the related logical buffer is
1073  * flushed.
1074  *
1075  * NOOFFSET is returned if the data is inode-embedded.  In this case the
1076  * strategy code will simply bcopy() the data into the inode.
1077  *
1078  * The inode's delta_dcount is adjusted.
1079  */
1080 static
1081 hammer2_chain_t *
1082 hammer2_assign_physical(hammer2_trans_t *trans,
1083 			hammer2_inode_t *ip, hammer2_chain_t **parentp,
1084 			hammer2_key_t lbase, int lblksize, int *errorp)
1085 {
1086 	hammer2_chain_t *parent;
1087 	hammer2_chain_t *chain;
1088 	hammer2_off_t pbase;
1089 
1090 	/*
1091 	 * Locate the chain associated with lbase, return a locked chain.
1092 	 * However, do not instantiate any data reference (which utilizes a
1093 	 * device buffer) because we will be using direct IO via the
1094 	 * logical buffer cache buffer.
1095 	 */
1096 	*errorp = 0;
1097 retry:
1098 	parent = *parentp;
1099 	hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS); /* extra lock */
1100 	chain = hammer2_chain_lookup(&parent,
1101 				     lbase, lbase,
1102 				     HAMMER2_LOOKUP_NODATA);
1103 
1104 	if (chain == NULL) {
1105 		/*
1106 		 * We found a hole, create a new chain entry.
1107 		 *
1108 		 * NOTE: DATA chains are created without device backing
1109 		 *	 store (nor do we want any).
1110 		 */
1111 		*errorp = hammer2_chain_create(trans, &parent, &chain,
1112 					       lbase, HAMMER2_PBUFRADIX,
1113 					       HAMMER2_BREF_TYPE_DATA,
1114 					       lblksize);
1115 		if (chain == NULL) {
1116 			hammer2_chain_lookup_done(parent);
1117 			panic("hammer2_chain_create: par=%p error=%d\n",
1118 				parent, *errorp);
1119 			goto retry;
1120 		}
1121 
1122 		pbase = chain->bref.data_off & ~HAMMER2_OFF_MASK_RADIX;
1123 		/*ip->delta_dcount += lblksize;*/
1124 	} else {
1125 		switch (chain->bref.type) {
1126 		case HAMMER2_BREF_TYPE_INODE:
1127 			/*
1128 			 * The data is embedded in the inode.  The
1129 			 * caller is responsible for marking the inode
1130 			 * modified and copying the data to the embedded
1131 			 * area.
1132 			 */
1133 			pbase = NOOFFSET;
1134 			break;
1135 		case HAMMER2_BREF_TYPE_DATA:
1136 			if (chain->bytes != lblksize) {
1137 				panic("hammer2_assign_physical: "
1138 				      "size mismatch %d/%d\n",
1139 				      lblksize, chain->bytes);
1140 			}
1141 			hammer2_chain_modify(trans, &chain,
1142 					     HAMMER2_MODIFY_OPTDATA);
1143 			pbase = chain->bref.data_off & ~HAMMER2_OFF_MASK_RADIX;
1144 			break;
1145 		default:
1146 			panic("hammer2_assign_physical: bad type");
1147 			/* NOT REACHED */
1148 			pbase = NOOFFSET;
1149 			break;
1150 		}
1151 	}
1152 
1153 	/*
1154 	 * Cleanup.  If chain wound up being the inode (i.e. DIRECTDATA),
1155 	 * we might have to replace *parentp.
1156 	 */
1157 	hammer2_chain_lookup_done(parent);
1158 	if (chain) {
1159 		if (*parentp != chain &&
1160 		    (*parentp)->core == chain->core) {
1161 			parent = *parentp;
1162 			*parentp = chain;		/* eats lock */
1163 			hammer2_chain_unlock(parent);
1164 			hammer2_chain_lock(chain, 0);	/* need another */
1165 		}
1166 		/* else chain already locked for return */
1167 	}
1168 	return (chain);
1169 }
1170 
1171 /*
1172  * Truncate the size of a file.
1173  *
1174  * This routine adjusts ipdata->size smaller, destroying any related
1175  * data beyond the new EOF and potentially resizing the block straddling
1176  * the EOF.
1177  *
1178  * The inode must be locked.
1179  */
1180 static
1181 void
1182 hammer2_truncate_file(hammer2_trans_t *trans, hammer2_inode_t *ip,
1183 		      hammer2_chain_t **parentp, hammer2_key_t nsize)
1184 {
1185 	hammer2_inode_data_t *ipdata;
1186 	hammer2_chain_t *parent;
1187 	hammer2_chain_t *chain;
1188 	hammer2_key_t lbase;
1189 	hammer2_key_t leof;
1190 	struct buf *bp;
1191 	int loff;
1192 	int error;
1193 	int oblksize;
1194 	int nblksize;
1195 
1196 	bp = NULL;
1197 	error = 0;
1198 	ipdata = hammer2_chain_modify_ip(trans, ip, parentp, 0);
1199 
1200 	/*
1201 	 * Destroy any logical buffer cache buffers beyond the file EOF.
1202 	 *
1203 	 * We call nvtruncbuf() w/ trivial == 1 to prevent it from messing
1204 	 * around with the buffer straddling EOF, because we need to assign
1205 	 * a new physical offset to it.
1206 	 */
1207 	if (ip->vp) {
1208 		nvtruncbuf(ip->vp, nsize,
1209 			   HAMMER2_PBUFSIZE, (int)nsize & HAMMER2_PBUFMASK,
1210 			   1);
1211 	}
1212 
1213 	/*
1214 	 * Setup for lookup/search
1215 	 */
1216 	parent = hammer2_chain_lookup_init(ip->chain, 0);
1217 
1218 	/*
1219 	 * Handle the case where a chain/logical-buffer straddles the new
1220 	 * EOF.  We told nvtruncbuf() above not to mess with the logical
1221 	 * buffer straddling the EOF because we need to reassign its storage
1222 	 * and can't let the strategy code do it for us.
1223 	 */
1224 	loff = (int)nsize & HAMMER2_PBUFMASK;
1225 	if (loff && ip->vp) {
1226 		oblksize = hammer2_calc_logical(ip, nsize, &lbase, &leof);
1227 		error = bread(ip->vp, lbase, oblksize, &bp);
1228 		KKASSERT(error == 0);
1229 	}
1230 	ipdata->size = nsize;
1231 	nblksize = hammer2_calc_logical(ip, nsize, &lbase, &leof);
1232 
1233 	/*
1234 	 * Fixup the chain element.  If we have a logical buffer in-hand
1235 	 * we don't want to create a conflicting device buffer.
1236 	 */
1237 	if (loff && bp) {
1238 		chain = hammer2_chain_lookup(&parent, lbase, lbase,
1239 					     HAMMER2_LOOKUP_NODATA);
1240 		if (chain) {
1241 			switch(chain->bref.type) {
1242 			case HAMMER2_BREF_TYPE_DATA:
1243 				hammer2_chain_resize(trans, ip, bp,
1244 					     parent, &chain,
1245 					     hammer2_getradix(nblksize),
1246 					     HAMMER2_MODIFY_OPTDATA);
1247 				allocbuf(bp, nblksize);
1248 				bzero(bp->b_data + loff, nblksize - loff);
1249 				bp->b_bio2.bio_caller_info1.ptr = chain->hmp;
1250 				bp->b_bio2.bio_offset = chain->bref.data_off &
1251 							HAMMER2_OFF_MASK;
1252 				break;
1253 			case HAMMER2_BREF_TYPE_INODE:
1254 				allocbuf(bp, nblksize);
1255 				bzero(bp->b_data + loff, nblksize - loff);
1256 				bp->b_bio2.bio_caller_info1.ptr = NULL;
1257 				bp->b_bio2.bio_offset = NOOFFSET;
1258 				break;
1259 			default:
1260 				panic("hammer2_truncate_file: bad type");
1261 				break;
1262 			}
1263 			hammer2_write_bp(chain, bp, 0);
1264 			hammer2_chain_unlock(chain);
1265 		} else {
1266 			/*
1267 			 * Destroy clean buffer w/ wrong buffer size.  Retain
1268 			 * backing store.
1269 			 */
1270 			bp->b_flags |= B_RELBUF;
1271 			KKASSERT(bp->b_bio2.bio_offset == NOOFFSET);
1272 			KKASSERT((bp->b_flags & B_DIRTY) == 0);
1273 			bqrelse(bp);
1274 		}
1275 	} else if (loff) {
1276 		/*
1277 		 * WARNING: This utilizes a device buffer for the data.
1278 		 *
1279 		 * This case should not occur because file truncations without
1280 		 * a vnode (and hence no logical buffer cache) should only
1281 		 * always truncate to 0-length.
1282 		 */
1283 		panic("hammer2_truncate_file: non-zero truncation, no-vnode");
1284 	}
1285 
1286 	/*
1287 	 * Clean up any fragmentory VM pages now that we have properly
1288 	 * resized the straddling buffer.  These pages are no longer
1289 	 * part of the buffer.
1290 	 */
1291 	if (ip->vp) {
1292 		nvtruncbuf(ip->vp, nsize,
1293 			   nblksize, (int)nsize & (nblksize - 1),
1294 			   1);
1295 	}
1296 
1297 	/*
1298 	 * Destroy any physical blocks after the new EOF point.
1299 	 */
1300 	lbase = (nsize + HAMMER2_PBUFMASK64) & ~HAMMER2_PBUFMASK64;
1301 	chain = hammer2_chain_lookup(&parent,
1302 				     lbase, (hammer2_key_t)-1,
1303 				     HAMMER2_LOOKUP_NODATA);
1304 	while (chain) {
1305 		/*
1306 		 * Degenerate embedded data case, nothing to loop on.
1307 		 */
1308 		if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) {
1309 			hammer2_chain_unlock(chain);
1310 			break;
1311 		}
1312 
1313 		/*
1314 		 * Delete physical data blocks past the file EOF.
1315 		 */
1316 		if (chain->bref.type == HAMMER2_BREF_TYPE_DATA) {
1317 			/*ip->delta_dcount -= chain->bytes;*/
1318 			hammer2_chain_delete(trans, chain, 0);
1319 		}
1320 		/* XXX check parent if empty indirect block & delete */
1321 		chain = hammer2_chain_next(&parent, chain,
1322 					   lbase, (hammer2_key_t)-1,
1323 					   HAMMER2_LOOKUP_NODATA);
1324 	}
1325 	hammer2_chain_lookup_done(parent);
1326 }
1327 
1328 /*
1329  * Extend the size of a file.  The inode must be locked.
1330  *
1331  * We may have to resize the block straddling the old EOF.
1332  */
1333 static
1334 void
1335 hammer2_extend_file(hammer2_trans_t *trans, hammer2_inode_t *ip,
1336 		    hammer2_chain_t **parentp, hammer2_key_t nsize)
1337 {
1338 	hammer2_inode_data_t *ipdata;
1339 	hammer2_chain_t *parent;
1340 	hammer2_chain_t *chain;
1341 	struct buf *bp;
1342 	hammer2_key_t osize;
1343 	hammer2_key_t obase;
1344 	hammer2_key_t nbase;
1345 	hammer2_key_t leof;
1346 	int oblksize;
1347 	int nblksize;
1348 	int nradix;
1349 	int error;
1350 
1351 	KKASSERT(ip->vp);
1352 
1353 	ipdata = hammer2_chain_modify_ip(trans, ip, parentp, 0);
1354 
1355 	/*
1356 	 * Nothing to do if the direct-data case is still intact
1357 	 */
1358 	if ((ipdata->op_flags & HAMMER2_OPFLAG_DIRECTDATA) &&
1359 	    nsize <= HAMMER2_EMBEDDED_BYTES) {
1360 		ipdata->size = nsize;
1361 		nvextendbuf(ip->vp,
1362 			    ipdata->size, nsize,
1363 			    0, HAMMER2_EMBEDDED_BYTES,
1364 			    0, (int)nsize,
1365 			    1);
1366 		/* ipdata = &ip->chain->data->ipdata; RELOAD */
1367 		return;
1368 	}
1369 
1370 	/*
1371 	 * Calculate the blocksize at the original EOF and resize the block
1372 	 * if necessary.  Adjust the file size in the inode.
1373 	 */
1374 	osize = ipdata->size;
1375 	oblksize = hammer2_calc_logical(ip, osize, &obase, &leof);
1376 	ipdata->size = nsize;
1377 	nblksize = hammer2_calc_logical(ip, osize, &nbase, &leof);
1378 
1379 	/*
1380 	 * Do all required vnode operations, but do not mess with the
1381 	 * buffer straddling the orignal EOF.
1382 	 */
1383 	nvextendbuf(ip->vp,
1384 		    ipdata->size, nsize,
1385 		    0, nblksize,
1386 		    0, (int)nsize & HAMMER2_PBUFMASK,
1387 		    1);
1388 	ipdata = &ip->chain->data->ipdata;
1389 
1390 	/*
1391 	 * Early return if we have no more work to do.
1392 	 */
1393 	if (obase == nbase && oblksize == nblksize &&
1394 	    (ipdata->op_flags & HAMMER2_OPFLAG_DIRECTDATA) == 0) {
1395 		return;
1396 	}
1397 
1398 	/*
1399 	 * We have work to do, including possibly resizing the buffer
1400 	 * at the previous EOF point and turning off DIRECTDATA mode.
1401 	 */
1402 	bp = NULL;
1403 	if (((int)osize & HAMMER2_PBUFMASK)) {
1404 		error = bread(ip->vp, obase, oblksize, &bp);
1405 		KKASSERT(error == 0);
1406 	}
1407 
1408 	/*
1409 	 * Disable direct-data mode by loading up a buffer cache buffer
1410 	 * with the data, then converting the inode data area into the
1411 	 * inode indirect block array area.
1412 	 */
1413 	if (ipdata->op_flags & HAMMER2_OPFLAG_DIRECTDATA) {
1414 		ipdata->op_flags &= ~HAMMER2_OPFLAG_DIRECTDATA;
1415 		bzero(&ipdata->u.blockset, sizeof(ipdata->u.blockset));
1416 	}
1417 
1418 	/*
1419 	 * Resize the chain element at the old EOF.
1420 	 */
1421 	if (((int)osize & HAMMER2_PBUFMASK)) {
1422 retry:
1423 		error = 0;
1424 		parent = hammer2_chain_lookup_init(ip->chain, 0);
1425 		nradix = hammer2_getradix(nblksize);
1426 
1427 		chain = hammer2_chain_lookup(&parent,
1428 					     obase, obase,
1429 					     HAMMER2_LOOKUP_NODATA);
1430 		if (chain == NULL) {
1431 			error = hammer2_chain_create(trans, &parent, &chain,
1432 						     obase, nblksize,
1433 						     HAMMER2_BREF_TYPE_DATA,
1434 						     nblksize);
1435 			if (chain == NULL) {
1436 				hammer2_chain_lookup_done(parent);
1437 				panic("hammer2_chain_create: par=%p error=%d\n",
1438 					parent, error);
1439 				goto retry;
1440 			}
1441 			/*ip->delta_dcount += nblksize;*/
1442 		} else {
1443 			KKASSERT(chain->bref.type == HAMMER2_BREF_TYPE_DATA);
1444 			hammer2_chain_resize(trans, ip, bp,
1445 					     parent, &chain,
1446 					     nradix,
1447 					     HAMMER2_MODIFY_OPTDATA);
1448 		}
1449 		if (obase != nbase) {
1450 			if (oblksize != HAMMER2_PBUFSIZE)
1451 				allocbuf(bp, HAMMER2_PBUFSIZE);
1452 		} else {
1453 			if (oblksize != nblksize)
1454 				allocbuf(bp, nblksize);
1455 		}
1456 		hammer2_write_bp(chain, bp, 0);
1457 		hammer2_chain_unlock(chain);
1458 		hammer2_chain_lookup_done(parent);
1459 	}
1460 }
1461 
1462 static
1463 int
1464 hammer2_vop_nresolve(struct vop_nresolve_args *ap)
1465 {
1466 	hammer2_inode_t *ip;
1467 	hammer2_inode_t *dip;
1468 	hammer2_chain_t *parent;
1469 	hammer2_chain_t *chain;
1470 	hammer2_chain_t *ochain;
1471 	hammer2_trans_t trans;
1472 	struct namecache *ncp;
1473 	const uint8_t *name;
1474 	size_t name_len;
1475 	hammer2_key_t lhc;
1476 	int error = 0;
1477 	struct vnode *vp;
1478 
1479 	dip = VTOI(ap->a_dvp);
1480 	ncp = ap->a_nch->ncp;
1481 	name = ncp->nc_name;
1482 	name_len = ncp->nc_nlen;
1483 	lhc = hammer2_dirhash(name, name_len);
1484 
1485 	/*
1486 	 * Note: In DragonFly the kernel handles '.' and '..'.
1487 	 */
1488 	parent = hammer2_inode_lock_sh(dip);
1489 	chain = hammer2_chain_lookup(&parent,
1490 				     lhc, lhc + HAMMER2_DIRHASH_LOMASK,
1491 				     HAMMER2_LOOKUP_SHARED);
1492 	while (chain) {
1493 		if (chain->bref.type == HAMMER2_BREF_TYPE_INODE &&
1494 		    name_len == chain->data->ipdata.name_len &&
1495 		    bcmp(name, chain->data->ipdata.filename, name_len) == 0) {
1496 			break;
1497 		}
1498 		chain = hammer2_chain_next(&parent, chain,
1499 					   lhc, lhc + HAMMER2_DIRHASH_LOMASK,
1500 					   HAMMER2_LOOKUP_SHARED);
1501 	}
1502 	hammer2_inode_unlock_sh(dip, parent);
1503 
1504 	/*
1505 	 * If the inode represents a forwarding entry for a hardlink we have
1506 	 * to locate the actual inode.  The original ip is saved for possible
1507 	 * deconsolidation.  (ip) will only be set to non-NULL when we have
1508 	 * to locate the real file via a hardlink.  ip will be referenced but
1509 	 * not locked in that situation.  chain is passed in locked and
1510 	 * returned locked.
1511 	 *
1512 	 * XXX what kind of chain lock?
1513 	 */
1514 	ochain = NULL;
1515 	if (chain && chain->data->ipdata.type == HAMMER2_OBJTYPE_HARDLINK) {
1516 		error = hammer2_hardlink_find(dip, &chain, &ochain);
1517 		if (error) {
1518 			kprintf("hammer2: unable to find hardlink\n");
1519 			if (chain) {
1520 				hammer2_chain_unlock(chain);
1521 				chain = NULL;
1522 			}
1523 			goto failed;
1524 		}
1525 	}
1526 
1527 	/*
1528 	 * Deconsolidate any hardlink whos nlinks == 1.  Ignore errors.
1529 	 * If an error occurs chain and ip are left alone.
1530 	 *
1531 	 * XXX upgrade shared lock?
1532 	 */
1533 	if (ochain && chain &&
1534 	    chain->data->ipdata.nlinks == 1 && !dip->pmp->ronly) {
1535 		kprintf("hammer2: need to unconsolidate hardlink for %s\n",
1536 			chain->data->ipdata.filename);
1537 		/* XXX retain shared lock on dip? (currently not held) */
1538 		hammer2_trans_init(&trans, dip->pmp, 0);
1539 		hammer2_hardlink_deconsolidate(&trans, dip, &chain, &ochain);
1540 		hammer2_trans_done(&trans);
1541 	}
1542 
1543 	/*
1544 	 * Acquire the related vnode
1545 	 *
1546 	 * NOTE: For error processing, only ENOENT resolves the namecache
1547 	 *	 entry to NULL, otherwise we just return the error and
1548 	 *	 leave the namecache unresolved.
1549 	 *
1550 	 * NOTE: multiple hammer2_inode structures can be aliased to the
1551 	 *	 same chain element, for example for hardlinks.  This
1552 	 *	 use case does not 'reattach' inode associations that
1553 	 *	 might already exist, but always allocates a new one.
1554 	 *
1555 	 * WARNING: inode structure is locked exclusively via inode_get
1556 	 *	    but chain was locked shared.  inode_unlock_ex()
1557 	 *	    will handle it properly.
1558 	 */
1559 	if (chain) {
1560 		ip = hammer2_inode_get(dip->pmp, dip, chain);
1561 		vp = hammer2_igetv(ip, &error);
1562 		if (error == 0) {
1563 			vn_unlock(vp);
1564 			cache_setvp(ap->a_nch, vp);
1565 		} else if (error == ENOENT) {
1566 			cache_setvp(ap->a_nch, NULL);
1567 		}
1568 		hammer2_inode_unlock_ex(ip, chain);
1569 
1570 		/*
1571 		 * The vp should not be released until after we've disposed
1572 		 * of our locks, because it might cause vop_inactive() to
1573 		 * be called.
1574 		 */
1575 		if (vp)
1576 			vrele(vp);
1577 	} else {
1578 		error = ENOENT;
1579 		cache_setvp(ap->a_nch, NULL);
1580 	}
1581 failed:
1582 	KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
1583 		("resolve error %d/%p chain %p ap %p\n",
1584 		 error, ap->a_nch->ncp->nc_vp, chain, ap));
1585 	if (ochain)
1586 		hammer2_chain_drop(ochain);
1587 	return error;
1588 }
1589 
1590 static
1591 int
1592 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
1593 {
1594 	hammer2_inode_t *dip;
1595 	hammer2_inode_t *ip;
1596 	hammer2_chain_t *parent;
1597 	int error;
1598 
1599 	dip = VTOI(ap->a_dvp);
1600 
1601 	if ((ip = dip->pip) == NULL) {
1602 		*ap->a_vpp = NULL;
1603 		return ENOENT;
1604 	}
1605 	parent = hammer2_inode_lock_ex(ip);
1606 	*ap->a_vpp = hammer2_igetv(ip, &error);
1607 	hammer2_inode_unlock_ex(ip, parent);
1608 
1609 	return error;
1610 }
1611 
1612 static
1613 int
1614 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
1615 {
1616 	hammer2_inode_t *dip;
1617 	hammer2_inode_t *nip;
1618 	hammer2_trans_t trans;
1619 	hammer2_chain_t *chain;
1620 	struct namecache *ncp;
1621 	const uint8_t *name;
1622 	size_t name_len;
1623 	int error;
1624 
1625 	dip = VTOI(ap->a_dvp);
1626 	if (dip->pmp->ronly)
1627 		return (EROFS);
1628 
1629 	ncp = ap->a_nch->ncp;
1630 	name = ncp->nc_name;
1631 	name_len = ncp->nc_nlen;
1632 
1633 	hammer2_trans_init(&trans, dip->pmp, 0);
1634 	nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1635 				   name, name_len, &chain, &error);
1636 	if (error) {
1637 		KKASSERT(nip == NULL);
1638 		*ap->a_vpp = NULL;
1639 	} else {
1640 		*ap->a_vpp = hammer2_igetv(nip, &error);
1641 		hammer2_inode_unlock_ex(nip, chain);
1642 	}
1643 	hammer2_trans_done(&trans);
1644 
1645 	if (error == 0) {
1646 		cache_setunresolved(ap->a_nch);
1647 		cache_setvp(ap->a_nch, *ap->a_vpp);
1648 	}
1649 	return error;
1650 }
1651 
1652 /*
1653  * Return the largest contiguous physical disk range for the logical
1654  * request, in bytes.
1655  *
1656  * (struct vnode *vp, off_t loffset, off_t *doffsetp, int *runp, int *runb)
1657  */
1658 static
1659 int
1660 hammer2_vop_bmap(struct vop_bmap_args *ap)
1661 {
1662 	*ap->a_doffsetp = NOOFFSET;
1663 	if (ap->a_runp)
1664 		*ap->a_runp = 0;
1665 	if (ap->a_runb)
1666 		*ap->a_runb = 0;
1667 	return (EOPNOTSUPP);
1668 #if 0
1669 	struct vnode *vp;
1670 	hammer2_inode_t *ip;
1671 	hammer2_chain_t *parent;
1672 	hammer2_chain_t *chain;
1673 	hammer2_key_t lbeg;
1674 	hammer2_key_t lend;
1675 	hammer2_off_t pbeg;
1676 	hammer2_off_t pbytes;
1677 	hammer2_off_t array[HAMMER2_BMAP_COUNT][2];
1678 	int loff;
1679 	int ai;
1680 
1681 	/*
1682 	 * Only supported on regular files
1683 	 *
1684 	 * Only supported for read operations (required for cluster_read).
1685 	 * The block allocation is delayed for write operations.
1686 	 */
1687 	vp = ap->a_vp;
1688 	if (vp->v_type != VREG)
1689 		return (EOPNOTSUPP);
1690 	if (ap->a_cmd != BUF_CMD_READ)
1691 		return (EOPNOTSUPP);
1692 
1693 	ip = VTOI(vp);
1694 	bzero(array, sizeof(array));
1695 
1696 	/*
1697 	 * Calculate logical range
1698 	 */
1699 	KKASSERT((ap->a_loffset & HAMMER2_LBUFMASK64) == 0);
1700 	lbeg = ap->a_loffset & HAMMER2_OFF_MASK_HI;
1701 	lend = lbeg + HAMMER2_BMAP_COUNT * HAMMER2_PBUFSIZE - 1;
1702 	if (lend < lbeg)
1703 		lend = lbeg;
1704 	loff = ap->a_loffset & HAMMER2_OFF_MASK_LO;
1705 
1706 	parent = hammer2_inode_lock_sh(ip);
1707 	chain = hammer2_chain_lookup(&parent,
1708 				     lbeg, lend,
1709 				     HAMMER2_LOOKUP_NODATA |
1710 				     HAMMER2_LOOKUP_SHARED);
1711 	if (chain == NULL) {
1712 		*ap->a_doffsetp = ZFOFFSET;
1713 		hammer2_inode_unlock_sh(ip, parent);
1714 		return (0);
1715 	}
1716 
1717 	while (chain) {
1718 		if (chain->bref.type == HAMMER2_BREF_TYPE_DATA) {
1719 			ai = (chain->bref.key - lbeg) / HAMMER2_PBUFSIZE;
1720 			KKASSERT(ai >= 0 && ai < HAMMER2_BMAP_COUNT);
1721 			array[ai][0] = chain->bref.data_off & HAMMER2_OFF_MASK;
1722 			array[ai][1] = chain->bytes;
1723 		}
1724 		chain = hammer2_chain_next(&parent, chain,
1725 					   lbeg, lend,
1726 					   HAMMER2_LOOKUP_NODATA |
1727 					   HAMMER2_LOOKUP_SHARED);
1728 	}
1729 	hammer2_inode_unlock_sh(ip, parent);
1730 
1731 	/*
1732 	 * If the requested loffset is not mappable physically we can't
1733 	 * bmap.  The caller will have to access the file data via a
1734 	 * device buffer.
1735 	 */
1736 	if (array[0][0] == 0 || array[0][1] < loff + HAMMER2_MINIOSIZE) {
1737 		*ap->a_doffsetp = NOOFFSET;
1738 		return (0);
1739 	}
1740 
1741 	/*
1742 	 * Calculate the physical disk offset range for array[0]
1743 	 */
1744 	pbeg = array[0][0] + loff;
1745 	pbytes = array[0][1] - loff;
1746 
1747 	for (ai = 1; ai < HAMMER2_BMAP_COUNT; ++ai) {
1748 		if (array[ai][0] != pbeg + pbytes)
1749 			break;
1750 		pbytes += array[ai][1];
1751 	}
1752 
1753 	*ap->a_doffsetp = pbeg;
1754 	if (ap->a_runp)
1755 		*ap->a_runp = pbytes;
1756 	return (0);
1757 #endif
1758 }
1759 
1760 static
1761 int
1762 hammer2_vop_open(struct vop_open_args *ap)
1763 {
1764 	return vop_stdopen(ap);
1765 }
1766 
1767 /*
1768  * hammer2_vop_advlock { vp, id, op, fl, flags }
1769  */
1770 static
1771 int
1772 hammer2_vop_advlock(struct vop_advlock_args *ap)
1773 {
1774 	hammer2_inode_t *ip = VTOI(ap->a_vp);
1775 	hammer2_chain_t *parent;
1776 	hammer2_off_t size;
1777 
1778 	parent = hammer2_inode_lock_sh(ip);
1779 	size = parent->data->ipdata.size;
1780 	hammer2_inode_unlock_sh(ip, parent);
1781 	return (lf_advlock(ap, &ip->advlock, size));
1782 }
1783 
1784 
1785 static
1786 int
1787 hammer2_vop_close(struct vop_close_args *ap)
1788 {
1789 	return vop_stdclose(ap);
1790 }
1791 
1792 /*
1793  * hammer2_vop_nlink { nch, dvp, vp, cred }
1794  *
1795  * Create a hardlink from (vp) to {dvp, nch}.
1796  */
1797 static
1798 int
1799 hammer2_vop_nlink(struct vop_nlink_args *ap)
1800 {
1801 	hammer2_inode_t *dip;	/* target directory to create link in */
1802 	hammer2_inode_t *ip;	/* inode we are hardlinking to */
1803 	hammer2_chain_t *chain;
1804 	hammer2_trans_t trans;
1805 	struct namecache *ncp;
1806 	const uint8_t *name;
1807 	size_t name_len;
1808 	int error;
1809 
1810 	dip = VTOI(ap->a_dvp);
1811 	if (dip->pmp->ronly)
1812 		return (EROFS);
1813 
1814 	ncp = ap->a_nch->ncp;
1815 	name = ncp->nc_name;
1816 	name_len = ncp->nc_nlen;
1817 
1818 	/*
1819 	 * ip represents the file being hardlinked.  The file could be a
1820 	 * normal file or a hardlink target if it has already been hardlinked.
1821 	 * If ip is a hardlinked target then ip->pip represents the location
1822 	 * of the hardlinked target, NOT the location of the hardlink pointer.
1823 	 *
1824 	 * Bump nlinks and potentially also create or move the hardlink
1825 	 * target in the parent directory common to (ip) and (dip).  The
1826 	 * consolidation code can modify ip->chain and ip->pip.  The
1827 	 * returned chain is locked.
1828 	 */
1829 	ip = VTOI(ap->a_vp);
1830 	hammer2_trans_init(&trans, ip->pmp, 0);
1831 
1832 	chain = hammer2_inode_lock_ex(ip);
1833 	error = hammer2_hardlink_consolidate(&trans, ip, &chain, dip, 1);
1834 	if (error)
1835 		goto done;
1836 
1837 	/*
1838 	 * Create a directory entry connected to the specified chain.
1839 	 * The hardlink consolidation code has already adjusted ip->pip
1840 	 * to the common parent directory containing the actual hardlink
1841 	 *
1842 	 * (which may be different from dip where we created our hardlink
1843 	 * entry. ip->chain always represents the actual hardlink and not
1844 	 * any of the pointers to the actual hardlink).
1845 	 */
1846 	error = hammer2_inode_connect(&trans, 1,
1847 				      dip, &chain,
1848 				      name, name_len);
1849 	if (error == 0) {
1850 		cache_setunresolved(ap->a_nch);
1851 		cache_setvp(ap->a_nch, ap->a_vp);
1852 	}
1853 done:
1854 	hammer2_inode_unlock_ex(ip, chain);
1855 	hammer2_trans_done(&trans);
1856 
1857 	return error;
1858 }
1859 
1860 /*
1861  * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1862  *
1863  * The operating system has already ensured that the directory entry
1864  * does not exist and done all appropriate namespace locking.
1865  */
1866 static
1867 int
1868 hammer2_vop_ncreate(struct vop_ncreate_args *ap)
1869 {
1870 	hammer2_inode_t *dip;
1871 	hammer2_inode_t *nip;
1872 	hammer2_trans_t trans;
1873 	hammer2_chain_t *nchain;
1874 	struct namecache *ncp;
1875 	const uint8_t *name;
1876 	size_t name_len;
1877 	int error;
1878 
1879 	dip = VTOI(ap->a_dvp);
1880 	if (dip->pmp->ronly)
1881 		return (EROFS);
1882 
1883 	ncp = ap->a_nch->ncp;
1884 	name = ncp->nc_name;
1885 	name_len = ncp->nc_nlen;
1886 	hammer2_trans_init(&trans, dip->pmp, 0);
1887 
1888 	nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1889 				   name, name_len, &nchain, &error);
1890 	if (error) {
1891 		KKASSERT(nip == NULL);
1892 		*ap->a_vpp = NULL;
1893 	} else {
1894 		*ap->a_vpp = hammer2_igetv(nip, &error);
1895 		hammer2_inode_unlock_ex(nip, nchain);
1896 	}
1897 	hammer2_trans_done(&trans);
1898 
1899 	if (error == 0) {
1900 		cache_setunresolved(ap->a_nch);
1901 		cache_setvp(ap->a_nch, *ap->a_vpp);
1902 	}
1903 	return error;
1904 }
1905 
1906 /*
1907  * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1908  */
1909 static
1910 int
1911 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
1912 {
1913 	hammer2_inode_t *dip;
1914 	hammer2_inode_t *nip;
1915 	hammer2_chain_t *nparent;
1916 	hammer2_trans_t trans;
1917 	struct namecache *ncp;
1918 	const uint8_t *name;
1919 	size_t name_len;
1920 	int error;
1921 
1922 	dip = VTOI(ap->a_dvp);
1923 	if (dip->pmp->ronly)
1924 		return (EROFS);
1925 
1926 	ncp = ap->a_nch->ncp;
1927 	name = ncp->nc_name;
1928 	name_len = ncp->nc_nlen;
1929 	hammer2_trans_init(&trans, dip->pmp, 0);
1930 
1931 	ap->a_vap->va_type = VLNK;	/* enforce type */
1932 
1933 	nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1934 				   name, name_len, &nparent, &error);
1935 	if (error) {
1936 		KKASSERT(nip == NULL);
1937 		*ap->a_vpp = NULL;
1938 		hammer2_trans_done(&trans);
1939 		return error;
1940 	}
1941 	*ap->a_vpp = hammer2_igetv(nip, &error);
1942 
1943 	/*
1944 	 * Build the softlink (~like file data) and finalize the namecache.
1945 	 */
1946 	if (error == 0) {
1947 		size_t bytes;
1948 		struct uio auio;
1949 		struct iovec aiov;
1950 		hammer2_inode_data_t *nipdata;
1951 
1952 		nipdata = &nip->chain->data->ipdata;
1953 		bytes = strlen(ap->a_target);
1954 
1955 		if (bytes <= HAMMER2_EMBEDDED_BYTES) {
1956 			KKASSERT(nipdata->op_flags &
1957 				 HAMMER2_OPFLAG_DIRECTDATA);
1958 			bcopy(ap->a_target, nipdata->u.data, bytes);
1959 			nipdata->size = bytes;
1960 		} else {
1961 			bzero(&auio, sizeof(auio));
1962 			bzero(&aiov, sizeof(aiov));
1963 			auio.uio_iov = &aiov;
1964 			auio.uio_segflg = UIO_SYSSPACE;
1965 			auio.uio_rw = UIO_WRITE;
1966 			auio.uio_resid = bytes;
1967 			auio.uio_iovcnt = 1;
1968 			auio.uio_td = curthread;
1969 			aiov.iov_base = ap->a_target;
1970 			aiov.iov_len = bytes;
1971 			error = hammer2_write_file(&trans, nip, &nparent,
1972 						   &auio, IO_APPEND, 0);
1973 			nipdata = &nip->chain->data->ipdata; /* RELOAD */
1974 			/* XXX handle error */
1975 			error = 0;
1976 		}
1977 	}
1978 	hammer2_inode_unlock_ex(nip, nparent);
1979 	hammer2_trans_done(&trans);
1980 
1981 	/*
1982 	 * Finalize namecache
1983 	 */
1984 	if (error == 0) {
1985 		cache_setunresolved(ap->a_nch);
1986 		cache_setvp(ap->a_nch, *ap->a_vpp);
1987 		/* hammer2_knote(ap->a_dvp, NOTE_WRITE); */
1988 	}
1989 	return error;
1990 }
1991 
1992 /*
1993  * hammer2_vop_nremove { nch, dvp, cred }
1994  */
1995 static
1996 int
1997 hammer2_vop_nremove(struct vop_nremove_args *ap)
1998 {
1999 	hammer2_inode_t *dip;
2000 	hammer2_trans_t trans;
2001 	struct namecache *ncp;
2002 	const uint8_t *name;
2003 	size_t name_len;
2004 	int error;
2005 
2006 	dip = VTOI(ap->a_dvp);
2007 	if (dip->pmp->ronly)
2008 		return(EROFS);
2009 
2010 	ncp = ap->a_nch->ncp;
2011 	name = ncp->nc_name;
2012 	name_len = ncp->nc_nlen;
2013 	hammer2_trans_init(&trans, dip->pmp, 0);
2014 	error = hammer2_unlink_file(&trans, dip, name, name_len, 0, NULL);
2015 	hammer2_trans_done(&trans);
2016 	if (error == 0) {
2017 		cache_unlink(ap->a_nch);
2018 	}
2019 	return (error);
2020 }
2021 
2022 /*
2023  * hammer2_vop_nrmdir { nch, dvp, cred }
2024  */
2025 static
2026 int
2027 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
2028 {
2029 	hammer2_inode_t *dip;
2030 	hammer2_trans_t trans;
2031 	struct namecache *ncp;
2032 	const uint8_t *name;
2033 	size_t name_len;
2034 	int error;
2035 
2036 	dip = VTOI(ap->a_dvp);
2037 	if (dip->pmp->ronly)
2038 		return(EROFS);
2039 
2040 	ncp = ap->a_nch->ncp;
2041 	name = ncp->nc_name;
2042 	name_len = ncp->nc_nlen;
2043 
2044 	hammer2_trans_init(&trans, dip->pmp, 0);
2045 	error = hammer2_unlink_file(&trans, dip, name, name_len, 1, NULL);
2046 	hammer2_trans_done(&trans);
2047 	if (error == 0) {
2048 		cache_unlink(ap->a_nch);
2049 	}
2050 	return (error);
2051 }
2052 
2053 /*
2054  * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
2055  */
2056 static
2057 int
2058 hammer2_vop_nrename(struct vop_nrename_args *ap)
2059 {
2060 	struct namecache *fncp;
2061 	struct namecache *tncp;
2062 	hammer2_inode_t *fdip;
2063 	hammer2_inode_t *tdip;
2064 	hammer2_inode_t *ip;
2065 	hammer2_chain_t *chain;
2066 	hammer2_trans_t trans;
2067 	const uint8_t *fname;
2068 	size_t fname_len;
2069 	const uint8_t *tname;
2070 	size_t tname_len;
2071 	int error;
2072 	int hlink;
2073 
2074 	if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
2075 		return(EXDEV);
2076 	if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
2077 		return(EXDEV);
2078 
2079 	fdip = VTOI(ap->a_fdvp);	/* source directory */
2080 	tdip = VTOI(ap->a_tdvp);	/* target directory */
2081 
2082 	if (fdip->pmp->ronly)
2083 		return(EROFS);
2084 
2085 	fncp = ap->a_fnch->ncp;		/* entry name in source */
2086 	fname = fncp->nc_name;
2087 	fname_len = fncp->nc_nlen;
2088 
2089 	tncp = ap->a_tnch->ncp;		/* entry name in target */
2090 	tname = tncp->nc_name;
2091 	tname_len = tncp->nc_nlen;
2092 
2093 	hammer2_trans_init(&trans, tdip->pmp, 0);
2094 
2095 	/*
2096 	 * ip is the inode being renamed.  If this is a hardlink then
2097 	 * ip represents the actual file and not the hardlink marker.
2098 	 */
2099 	ip = VTOI(fncp->nc_vp);
2100 	chain = NULL;
2101 
2102 	/*
2103 	 * Keep a tight grip on the inode so the temporary unlinking from
2104 	 * the source location prior to linking to the target location
2105 	 * does not cause the chain to be destroyed.
2106 	 *
2107 	 * NOTE: To avoid deadlocks we cannot lock (ip) while we are
2108 	 *	 unlinking elements from their directories.  Locking
2109 	 *	 the nlinks field does not lock the whole inode.
2110 	 */
2111 	hammer2_inode_ref(ip);
2112 
2113 	/*
2114 	 * Remove target if it exists
2115 	 */
2116 	error = hammer2_unlink_file(&trans, tdip, tname, tname_len, -1, NULL);
2117 	if (error && error != ENOENT)
2118 		goto done;
2119 	cache_setunresolved(ap->a_tnch);
2120 
2121 	/*
2122 	 * When renaming a hardlinked file we may have to re-consolidate
2123 	 * the location of the hardlink target.  Since the element is simply
2124 	 * being moved, nlinks is not modified in this case.
2125 	 *
2126 	 * If ip represents a regular file the consolidation code essentially
2127 	 * does nothing other than return the same locked chain that was
2128 	 * passed in.
2129 	 *
2130 	 * The returned chain will be locked.
2131 	 *
2132 	 * WARNING!  We do not currently have a local copy of ipdata but
2133 	 *	     we do use one later remember that it must be reloaded
2134 	 *	     on any modification to the inode, including connects.
2135 	 */
2136 	chain = hammer2_inode_lock_ex(ip);
2137 	error = hammer2_hardlink_consolidate(&trans, ip, &chain, tdip, 0);
2138 	if (error)
2139 		goto done;
2140 
2141 	/*
2142 	 * Disconnect (fdip, fname) from the source directory.  This will
2143 	 * disconnect (ip) if it represents a direct file.  If (ip) represents
2144 	 * a hardlink the HARDLINK pointer object will be removed but the
2145 	 * hardlink will stay intact.
2146 	 *
2147 	 * The target chain may be marked DELETED but will not be destroyed
2148 	 * since we retain our hold on ip and chain.
2149 	 */
2150 	error = hammer2_unlink_file(&trans, fdip, fname, fname_len, -1, &hlink);
2151 	KKASSERT(error != EAGAIN);
2152 	if (error)
2153 		goto done;
2154 
2155 	/*
2156 	 * Reconnect ip to target directory using chain.  Chains cannot
2157 	 * actually be moved, so this will duplicate the chain in the new
2158 	 * spot and assign it to the ip, replacing the old chain.
2159 	 *
2160 	 * WARNING: chain locks can lock buffer cache buffers, to avoid
2161 	 *	    deadlocks we want to unlock before issuing a cache_*()
2162 	 *	    op (that might have to lock a vnode).
2163 	 */
2164 	error = hammer2_inode_connect(&trans, hlink,
2165 				      tdip, &chain,
2166 				      tname, tname_len);
2167 	if (error == 0) {
2168 		KKASSERT(chain != NULL);
2169 		hammer2_inode_repoint(ip, (hlink ? ip->pip : tdip), chain);
2170 		cache_rename(ap->a_fnch, ap->a_tnch);
2171 	}
2172 done:
2173 	hammer2_inode_unlock_ex(ip, chain);
2174 	hammer2_inode_drop(ip);
2175 	hammer2_trans_done(&trans);
2176 
2177 	return (error);
2178 }
2179 
2180 /*
2181  * Strategy code
2182  *
2183  * WARNING: The strategy code cannot safely use hammer2 transactions
2184  *	    as this can deadlock against vfs_sync's vfsync() call
2185  *	    if multiple flushes are queued.
2186  */
2187 static int hammer2_strategy_read(struct vop_strategy_args *ap);
2188 static int hammer2_strategy_write(struct vop_strategy_args *ap);
2189 static void hammer2_strategy_read_callback(hammer2_chain_t *chain,
2190 				struct buf *dbp, char *data, void *arg);
2191 
2192 static
2193 int
2194 hammer2_vop_strategy(struct vop_strategy_args *ap)
2195 {
2196 	struct bio *biop;
2197 	struct buf *bp;
2198 	int error;
2199 
2200 	biop = ap->a_bio;
2201 	bp = biop->bio_buf;
2202 
2203 	switch(bp->b_cmd) {
2204 	case BUF_CMD_READ:
2205 		error = hammer2_strategy_read(ap);
2206 		++hammer2_iod_file_read;
2207 		break;
2208 	case BUF_CMD_WRITE:
2209 		error = hammer2_strategy_write(ap);
2210 		++hammer2_iod_file_write;
2211 		break;
2212 	default:
2213 		bp->b_error = error = EINVAL;
2214 		bp->b_flags |= B_ERROR;
2215 		biodone(biop);
2216 		break;
2217 	}
2218 
2219 	return (error);
2220 }
2221 
2222 static
2223 int
2224 hammer2_strategy_read(struct vop_strategy_args *ap)
2225 {
2226 	struct buf *bp;
2227 	struct bio *bio;
2228 	struct bio *nbio;
2229 	hammer2_inode_t *ip;
2230 	hammer2_chain_t *parent;
2231 	hammer2_chain_t *chain;
2232 	hammer2_key_t lbase;
2233 
2234 	bio = ap->a_bio;
2235 	bp = bio->bio_buf;
2236 	ip = VTOI(ap->a_vp);
2237 	nbio = push_bio(bio);
2238 
2239 	lbase = bio->bio_offset;
2240 	chain = NULL;
2241 	KKASSERT(((int)lbase & HAMMER2_PBUFMASK) == 0);
2242 
2243 #if 0
2244 	kprintf("read lbase %jd cached %016jx\n",
2245 		lbase, nbio->bio_offset);
2246 #endif
2247 
2248 	parent = hammer2_inode_lock_sh(ip);
2249 	chain = hammer2_chain_lookup(&parent, lbase, lbase,
2250 				     HAMMER2_LOOKUP_NODATA |
2251 				     HAMMER2_LOOKUP_SHARED);
2252 
2253 	if (chain == NULL) {
2254 		/*
2255 		 * Data is zero-fill
2256 		 */
2257 		bp->b_resid = 0;
2258 		bp->b_error = 0;
2259 		bzero(bp->b_data, bp->b_bcount);
2260 		biodone(nbio);
2261 	} else if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) {
2262 		/*
2263 		 * Data is embedded in the inode (copy from inode).
2264 		 */
2265 		hammer2_chain_load_async(chain, hammer2_strategy_read_callback,
2266 					 nbio);
2267 	} else if (chain->bref.type == HAMMER2_BREF_TYPE_DATA) {
2268 		/*
2269 		 * Data is on-media, issue device I/O and copy.
2270 		 *
2271 		 * XXX direct-IO shortcut could go here XXX.
2272 		 */
2273 		hammer2_chain_load_async(chain, hammer2_strategy_read_callback,
2274 					 nbio);
2275 	} else {
2276 		panic("hammer2_strategy_read: unknown bref type");
2277 		chain = NULL;
2278 	}
2279 	hammer2_inode_unlock_sh(ip, parent);
2280 	return (0);
2281 }
2282 
2283 static
2284 void
2285 hammer2_strategy_read_callback(hammer2_chain_t *chain, struct buf *dbp,
2286 			       char *data, void *arg)
2287 {
2288 	struct bio *nbio = arg;
2289 	struct buf *bp = nbio->bio_buf;
2290 
2291 	if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) {
2292 		/*
2293 		 * Data is embedded in the inode (copy from inode).
2294 		 */
2295 		bcopy(((hammer2_inode_data_t *)data)->u.data,
2296 		      bp->b_data, HAMMER2_EMBEDDED_BYTES);
2297 		bzero(bp->b_data + HAMMER2_EMBEDDED_BYTES,
2298 		      bp->b_bcount - HAMMER2_EMBEDDED_BYTES);
2299 		bp->b_resid = 0;
2300 		bp->b_error = 0;
2301 		hammer2_chain_unlock(chain);
2302 		biodone(nbio);
2303 	} else if (chain->bref.type == HAMMER2_BREF_TYPE_DATA) {
2304 		/*
2305 		 * Data is on-media, issue device I/O and copy.
2306 		 *
2307 		 * XXX direct-IO shortcut could go here XXX.
2308 		 */
2309 		bcopy(data, bp->b_data, bp->b_bcount);
2310 		bp->b_flags |= B_NOTMETA;
2311 		bp->b_resid = 0;
2312 		bp->b_error = 0;
2313 		hammer2_chain_unlock(chain);
2314 		biodone(nbio);
2315 	} else {
2316 		if (dbp)
2317 			bqrelse(dbp);
2318 		panic("hammer2_strategy_read: unknown bref type");
2319 		chain = NULL;
2320 	}
2321 }
2322 
2323 static
2324 int
2325 hammer2_strategy_write(struct vop_strategy_args *ap)
2326 {
2327 	KKASSERT(0);
2328 #if 0
2329 	struct buf *bp;
2330 	struct bio *bio;
2331 	struct bio *nbio;
2332 	hammer2_chain_t *chain;
2333 	hammer2_mount_t *hmp;
2334 	hammer2_inode_t *ip;
2335 
2336 	bio = ap->a_bio;
2337 	bp = bio->bio_buf;
2338 	ip = VTOI(ap->a_vp);
2339 	nbio = push_bio(bio);
2340 
2341 	KKASSERT((bio->bio_offset & HAMMER2_PBUFMASK64) == 0);
2342 	KKASSERT(nbio->bio_offset != 0 && nbio->bio_offset != ZFOFFSET);
2343 
2344 	if (nbio->bio_offset == NOOFFSET) {
2345 		/*
2346 		 * The data is embedded in the inode.  Note that strategy
2347 		 * calls for embedded data are synchronous in order to
2348 		 * ensure that ip->chain is stable.  Chain modification
2349 		 * status is handled by the caller.
2350 		 */
2351 		KKASSERT(ip->chain->flags & HAMMER2_CHAIN_MODIFIED);
2352 		KKASSERT(bio->bio_offset == 0);
2353 		KKASSERT(ip->chain && ip->chain->data);
2354 		chain = ip->chain;
2355 		bcopy(bp->b_data, chain->data->ipdata.u.data,
2356 		      HAMMER2_EMBEDDED_BYTES);
2357 		bp->b_resid = 0;
2358 		bp->b_error = 0;
2359 		biodone(nbio);
2360 	} else {
2361 		/*
2362 		 * Forward direct IO to the device
2363 		 */
2364 		hmp = nbio->bio_caller_info1.ptr;
2365 		KKASSERT(hmp);
2366 		vn_strategy(hmp->devvp, nbio);
2367 	}
2368 	return (0);
2369 #endif
2370 }
2371 
2372 /*
2373  * hammer2_vop_ioctl { vp, command, data, fflag, cred }
2374  */
2375 static
2376 int
2377 hammer2_vop_ioctl(struct vop_ioctl_args *ap)
2378 {
2379 	hammer2_inode_t *ip;
2380 	int error;
2381 
2382 	ip = VTOI(ap->a_vp);
2383 
2384 	error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
2385 			      ap->a_fflag, ap->a_cred);
2386 	return (error);
2387 }
2388 
2389 static
2390 int
2391 hammer2_vop_mountctl(struct vop_mountctl_args *ap)
2392 {
2393 	struct mount *mp;
2394 	hammer2_pfsmount_t *pmp;
2395 	int rc;
2396 
2397 	switch (ap->a_op) {
2398 	case (MOUNTCTL_SET_EXPORT):
2399 		mp = ap->a_head.a_ops->head.vv_mount;
2400 		pmp = MPTOPMP(mp);
2401 
2402 		if (ap->a_ctllen != sizeof(struct export_args))
2403 			rc = (EINVAL);
2404 		else
2405 			rc = vfs_export(mp, &pmp->export,
2406 					(const struct export_args *)ap->a_ctl);
2407 		break;
2408 	default:
2409 		rc = vop_stdmountctl(ap);
2410 		break;
2411 	}
2412 	return (rc);
2413 }
2414 
2415 struct vop_ops hammer2_vnode_vops = {
2416 	.vop_default	= vop_defaultop,
2417 	.vop_fsync	= hammer2_vop_fsync,
2418 	.vop_getpages	= vop_stdgetpages,
2419 	.vop_putpages	= vop_stdputpages,
2420 	.vop_access	= hammer2_vop_access,
2421 	.vop_advlock	= hammer2_vop_advlock,
2422 	.vop_close	= hammer2_vop_close,
2423 	.vop_nlink	= hammer2_vop_nlink,
2424 	.vop_ncreate	= hammer2_vop_ncreate,
2425 	.vop_nsymlink	= hammer2_vop_nsymlink,
2426 	.vop_nremove	= hammer2_vop_nremove,
2427 	.vop_nrmdir	= hammer2_vop_nrmdir,
2428 	.vop_nrename	= hammer2_vop_nrename,
2429 	.vop_getattr	= hammer2_vop_getattr,
2430 	.vop_setattr	= hammer2_vop_setattr,
2431 	.vop_readdir	= hammer2_vop_readdir,
2432 	.vop_readlink	= hammer2_vop_readlink,
2433 	.vop_getpages	= vop_stdgetpages,
2434 	.vop_putpages	= vop_stdputpages,
2435 	.vop_read	= hammer2_vop_read,
2436 	.vop_write	= hammer2_vop_write,
2437 	.vop_open	= hammer2_vop_open,
2438 	.vop_inactive	= hammer2_vop_inactive,
2439 	.vop_reclaim 	= hammer2_vop_reclaim,
2440 	.vop_nresolve	= hammer2_vop_nresolve,
2441 	.vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
2442 	.vop_nmkdir 	= hammer2_vop_nmkdir,
2443 	.vop_ioctl	= hammer2_vop_ioctl,
2444 	.vop_mountctl	= hammer2_vop_mountctl,
2445 	.vop_bmap	= hammer2_vop_bmap,
2446 	.vop_strategy	= hammer2_vop_strategy,
2447 };
2448 
2449 struct vop_ops hammer2_spec_vops = {
2450 
2451 };
2452 
2453 struct vop_ops hammer2_fifo_vops = {
2454 
2455 };
2456