xref: /dragonfly/sys/kern/vfs_jops.c (revision 0ca59c34)
1 /*
2  * Copyright (c) 2004-2006 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@backplane.com>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in
15  *    the documentation and/or other materials provided with the
16  *    distribution.
17  * 3. Neither the name of The DragonFly Project nor the names of its
18  *    contributors may be used to endorse or promote products derived
19  *    from this software without specific, prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  * $DragonFly: src/sys/kern/vfs_jops.c,v 1.36 2007/08/21 17:43:52 dillon Exp $
35  */
36 /*
37  * Each mount point may have zero or more independantly configured journals
38  * attached to it.  Each journal is represented by a memory FIFO and worker
39  * thread.  Journal events are streamed through the FIFO to the thread,
40  * batched up (typically on one-second intervals), and written out by the
41  * thread.
42  *
43  * Journal vnode ops are executed instead of mnt_vn_norm_ops when one or
44  * more journals have been installed on a mount point.  It becomes the
45  * responsibility of the journal op to call the underlying normal op as
46  * appropriate.
47  */
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/buf.h>
51 #include <sys/conf.h>
52 #include <sys/kernel.h>
53 #include <sys/queue.h>
54 #include <sys/lock.h>
55 #include <sys/malloc.h>
56 #include <sys/mount.h>
57 #include <sys/unistd.h>
58 #include <sys/vnode.h>
59 #include <sys/poll.h>
60 #include <sys/mountctl.h>
61 #include <sys/journal.h>
62 #include <sys/file.h>
63 #include <sys/proc.h>
64 #include <sys/socket.h>
65 #include <sys/socketvar.h>
66 
67 #include <machine/limits.h>
68 
69 #include <vm/vm.h>
70 #include <vm/vm_object.h>
71 #include <vm/vm_page.h>
72 #include <vm/vm_pager.h>
73 #include <vm/vnode_pager.h>
74 
75 #include <sys/file2.h>
76 #include <sys/thread2.h>
77 
78 static int journal_attach(struct mount *mp);
79 static void journal_detach(struct mount *mp);
80 static int journal_install_vfs_journal(struct mount *mp, struct file *fp,
81 			    const struct mountctl_install_journal *info);
82 static int journal_restart_vfs_journal(struct mount *mp, struct file *fp,
83 			    const struct mountctl_restart_journal *info);
84 static int journal_remove_vfs_journal(struct mount *mp,
85 			    const struct mountctl_remove_journal *info);
86 static int journal_restart(struct mount *mp, struct file *fp,
87 			    struct journal *jo, int flags);
88 static int journal_destroy(struct mount *mp, struct journal *jo, int flags);
89 static int journal_resync_vfs_journal(struct mount *mp, const void *ctl);
90 static int journal_status_vfs_journal(struct mount *mp,
91 		       const struct mountctl_status_journal *info,
92 		       struct mountctl_journal_ret_status *rstat,
93 		       int buflen, int *res);
94 
95 static void jrecord_undo_file(struct jrecord *jrec, struct vnode *vp,
96 			     int jrflags, off_t off, off_t bytes);
97 
98 static int journal_setattr(struct vop_setattr_args *ap);
99 static int journal_write(struct vop_write_args *ap);
100 static int journal_fsync(struct vop_fsync_args *ap);
101 static int journal_putpages(struct vop_putpages_args *ap);
102 static int journal_setacl(struct vop_setacl_args *ap);
103 static int journal_setextattr(struct vop_setextattr_args *ap);
104 static int journal_ncreate(struct vop_ncreate_args *ap);
105 static int journal_nmknod(struct vop_nmknod_args *ap);
106 static int journal_nlink(struct vop_nlink_args *ap);
107 static int journal_nsymlink(struct vop_nsymlink_args *ap);
108 static int journal_nwhiteout(struct vop_nwhiteout_args *ap);
109 static int journal_nremove(struct vop_nremove_args *ap);
110 static int journal_nmkdir(struct vop_nmkdir_args *ap);
111 static int journal_nrmdir(struct vop_nrmdir_args *ap);
112 static int journal_nrename(struct vop_nrename_args *ap);
113 
114 #define JRUNDO_SIZE	0x00000001
115 #define JRUNDO_UID	0x00000002
116 #define JRUNDO_GID	0x00000004
117 #define JRUNDO_FSID	0x00000008
118 #define JRUNDO_MODES	0x00000010
119 #define JRUNDO_INUM	0x00000020
120 #define JRUNDO_ATIME	0x00000040
121 #define JRUNDO_MTIME	0x00000080
122 #define JRUNDO_CTIME	0x00000100
123 #define JRUNDO_GEN	0x00000200
124 #define JRUNDO_FLAGS	0x00000400
125 #define JRUNDO_UDEV	0x00000800
126 #define JRUNDO_NLINK	0x00001000
127 #define JRUNDO_FILEDATA	0x00010000
128 #define JRUNDO_GETVP	0x00020000
129 #define JRUNDO_CONDLINK	0x00040000	/* write file data if link count 1 */
130 #define JRUNDO_VATTR	(JRUNDO_SIZE|JRUNDO_UID|JRUNDO_GID|JRUNDO_FSID|\
131 			 JRUNDO_MODES|JRUNDO_INUM|JRUNDO_ATIME|JRUNDO_MTIME|\
132 			 JRUNDO_CTIME|JRUNDO_GEN|JRUNDO_FLAGS|JRUNDO_UDEV|\
133 			 JRUNDO_NLINK)
134 #define JRUNDO_ALL	(JRUNDO_VATTR|JRUNDO_FILEDATA)
135 
136 static struct vop_ops journal_vnode_vops = {
137     .vop_default =	vop_journal_operate_ap,
138     .vop_mountctl =	journal_mountctl,
139     .vop_setattr =	journal_setattr,
140     .vop_write =	journal_write,
141     .vop_fsync =	journal_fsync,
142     .vop_putpages =	journal_putpages,
143     .vop_setacl =	journal_setacl,
144     .vop_setextattr =	journal_setextattr,
145     .vop_ncreate =	journal_ncreate,
146     .vop_nmknod =	journal_nmknod,
147     .vop_nlink =	journal_nlink,
148     .vop_nsymlink =	journal_nsymlink,
149     .vop_nwhiteout =	journal_nwhiteout,
150     .vop_nremove =	journal_nremove,
151     .vop_nmkdir =	journal_nmkdir,
152     .vop_nrmdir =	journal_nrmdir,
153     .vop_nrename =	journal_nrename
154 };
155 
156 int
157 journal_mountctl(struct vop_mountctl_args *ap)
158 {
159     struct mount *mp;
160     int error = 0;
161 
162     mp = ap->a_head.a_ops->head.vv_mount;
163     KKASSERT(mp);
164 
165     if (mp->mnt_vn_journal_ops == NULL) {
166 	switch(ap->a_op) {
167 	case MOUNTCTL_INSTALL_VFS_JOURNAL:
168 	    error = journal_attach(mp);
169 	    if (error == 0 && ap->a_ctllen != sizeof(struct mountctl_install_journal))
170 		error = EINVAL;
171 	    if (error == 0 && ap->a_fp == NULL)
172 		error = EBADF;
173 	    if (error == 0)
174 		error = journal_install_vfs_journal(mp, ap->a_fp, ap->a_ctl);
175 	    if (TAILQ_EMPTY(&mp->mnt_jlist))
176 		journal_detach(mp);
177 	    break;
178 	case MOUNTCTL_RESTART_VFS_JOURNAL:
179 	case MOUNTCTL_REMOVE_VFS_JOURNAL:
180 	case MOUNTCTL_RESYNC_VFS_JOURNAL:
181 	case MOUNTCTL_STATUS_VFS_JOURNAL:
182 	    error = ENOENT;
183 	    break;
184 	default:
185 	    error = EOPNOTSUPP;
186 	    break;
187 	}
188     } else {
189 	switch(ap->a_op) {
190 	case MOUNTCTL_INSTALL_VFS_JOURNAL:
191 	    if (ap->a_ctllen != sizeof(struct mountctl_install_journal))
192 		error = EINVAL;
193 	    if (error == 0 && ap->a_fp == NULL)
194 		error = EBADF;
195 	    if (error == 0)
196 		error = journal_install_vfs_journal(mp, ap->a_fp, ap->a_ctl);
197 	    break;
198 	case MOUNTCTL_RESTART_VFS_JOURNAL:
199 	    if (ap->a_ctllen != sizeof(struct mountctl_restart_journal))
200 		error = EINVAL;
201 	    if (error == 0 && ap->a_fp == NULL)
202 		error = EBADF;
203 	    if (error == 0)
204 		error = journal_restart_vfs_journal(mp, ap->a_fp, ap->a_ctl);
205 	    break;
206 	case MOUNTCTL_REMOVE_VFS_JOURNAL:
207 	    if (ap->a_ctllen != sizeof(struct mountctl_remove_journal))
208 		error = EINVAL;
209 	    if (error == 0)
210 		error = journal_remove_vfs_journal(mp, ap->a_ctl);
211 	    if (TAILQ_EMPTY(&mp->mnt_jlist))
212 		journal_detach(mp);
213 	    break;
214 	case MOUNTCTL_RESYNC_VFS_JOURNAL:
215 	    if (ap->a_ctllen != 0)
216 		error = EINVAL;
217 	    error = journal_resync_vfs_journal(mp, ap->a_ctl);
218 	    break;
219 	case MOUNTCTL_STATUS_VFS_JOURNAL:
220 	    if (ap->a_ctllen != sizeof(struct mountctl_status_journal))
221 		error = EINVAL;
222 	    if (error == 0) {
223 		error = journal_status_vfs_journal(mp, ap->a_ctl,
224 					ap->a_buf, ap->a_buflen, ap->a_res);
225 	    }
226 	    break;
227 	default:
228 	    error = EOPNOTSUPP;
229 	    break;
230 	}
231     }
232     return (error);
233 }
234 
235 /*
236  * High level mount point setup.  When a
237  */
238 static int
239 journal_attach(struct mount *mp)
240 {
241     KKASSERT(mp->mnt_jbitmap == NULL);
242     vfs_add_vnodeops(mp, &journal_vnode_vops, &mp->mnt_vn_journal_ops);
243     mp->mnt_jbitmap = kmalloc(JREC_STREAMID_JMAX/8, M_JOURNAL, M_WAITOK|M_ZERO);
244     mp->mnt_streamid = JREC_STREAMID_JMIN;
245     return(0);
246 }
247 
248 static void
249 journal_detach(struct mount *mp)
250 {
251     KKASSERT(mp->mnt_jbitmap != NULL);
252     if (mp->mnt_vn_journal_ops)
253 	vfs_rm_vnodeops(mp, &journal_vnode_vops, &mp->mnt_vn_journal_ops);
254     kfree(mp->mnt_jbitmap, M_JOURNAL);
255     mp->mnt_jbitmap = NULL;
256 }
257 
258 /*
259  * Install a journal on a mount point.  Each journal has an associated worker
260  * thread which is responsible for buffering and spooling the data to the
261  * target.  A mount point may have multiple journals attached to it.  An
262  * initial start record is generated when the journal is associated.
263  */
264 static int
265 journal_install_vfs_journal(struct mount *mp, struct file *fp,
266 			    const struct mountctl_install_journal *info)
267 {
268     struct journal *jo;
269     struct jrecord jrec;
270     int error = 0;
271     int size;
272 
273     jo = kmalloc(sizeof(struct journal), M_JOURNAL, M_WAITOK|M_ZERO);
274     bcopy(info->id, jo->id, sizeof(jo->id));
275     jo->flags = info->flags & ~(MC_JOURNAL_WACTIVE | MC_JOURNAL_RACTIVE |
276 				MC_JOURNAL_STOP_REQ);
277 
278     /*
279      * Memory FIFO size, round to nearest power of 2
280      */
281     if (info->membufsize) {
282 	if (info->membufsize < 65536)
283 	    size = 65536;
284 	else if (info->membufsize > 128 * 1024 * 1024)
285 	    size = 128 * 1024 * 1024;
286 	else
287 	    size = (int)info->membufsize;
288     } else {
289 	size = 1024 * 1024;
290     }
291     jo->fifo.size = 1;
292     while (jo->fifo.size < size)
293 	jo->fifo.size <<= 1;
294 
295     /*
296      * Other parameters.  If not specified the starting transaction id
297      * will be the current date.
298      */
299     if (info->transid) {
300 	jo->transid = info->transid;
301     } else {
302 	struct timespec ts;
303 	getnanotime(&ts);
304 	jo->transid = ((int64_t)ts.tv_sec << 30) | ts.tv_nsec;
305     }
306 
307     jo->fp = fp;
308 
309     /*
310      * Allocate the memory FIFO
311      */
312     jo->fifo.mask = jo->fifo.size - 1;
313     jo->fifo.membase = kmalloc(jo->fifo.size, M_JFIFO, M_WAITOK|M_ZERO|M_NULLOK);
314     if (jo->fifo.membase == NULL)
315 	error = ENOMEM;
316 
317     /*
318      * Create the worker threads and generate the association record.
319      */
320     if (error) {
321 	kfree(jo, M_JOURNAL);
322     } else {
323 	fhold(fp);
324 	journal_create_threads(jo);
325 	jrecord_init(jo, &jrec, JREC_STREAMID_DISCONT);
326 	jrecord_write(&jrec, JTYPE_ASSOCIATE, 0);
327 	jrecord_done(&jrec, 0);
328 	TAILQ_INSERT_TAIL(&mp->mnt_jlist, jo, jentry);
329     }
330     return(error);
331 }
332 
333 /*
334  * Restart a journal with a new descriptor.   The existing reader and writer
335  * threads are terminated and a new descriptor is associated with the
336  * journal.  The FIFO rindex is reset to xindex and the threads are then
337  * restarted.
338  */
339 static int
340 journal_restart_vfs_journal(struct mount *mp, struct file *fp,
341 			   const struct mountctl_restart_journal *info)
342 {
343     struct journal *jo;
344     int error;
345 
346     TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
347 	if (bcmp(jo->id, info->id, sizeof(jo->id)) == 0)
348 	    break;
349     }
350     if (jo)
351 	error = journal_restart(mp, fp, jo, info->flags);
352     else
353 	error = EINVAL;
354     return (error);
355 }
356 
357 static int
358 journal_restart(struct mount *mp, struct file *fp,
359 		struct journal *jo, int flags)
360 {
361     /*
362      * XXX lock the jo
363      */
364 
365 #if 0
366     /*
367      * Record the fact that we are doing a restart in the journal.
368      * XXX it isn't safe to do this if the journal is being restarted
369      * because it was locked up and the writer thread has already exited.
370      */
371     jrecord_init(jo, &jrec, JREC_STREAMID_RESTART);
372     jrecord_write(&jrec, JTYPE_DISASSOCIATE, 0);
373     jrecord_done(&jrec, 0);
374 #endif
375 
376     /*
377      * Stop the reader and writer threads and clean up the current
378      * descriptor.
379      */
380     kprintf("RESTART WITH FP %p KILLING %p\n", fp, jo->fp);
381     journal_destroy_threads(jo, flags);
382 
383     if (jo->fp)
384 	fdrop(jo->fp);
385 
386     /*
387      * Associate the new descriptor, reset the FIFO index, and recreate
388      * the threads.
389      */
390     fhold(fp);
391     jo->fp = fp;
392     jo->fifo.rindex = jo->fifo.xindex;
393     journal_create_threads(jo);
394 
395     return(0);
396 }
397 
398 /*
399  * Disassociate a journal from a mount point and terminate its worker thread.
400  * A final termination record is written out before the file pointer is
401  * dropped.
402  */
403 static int
404 journal_remove_vfs_journal(struct mount *mp,
405 			   const struct mountctl_remove_journal *info)
406 {
407     struct journal *jo;
408     int error;
409 
410     TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
411 	if (bcmp(jo->id, info->id, sizeof(jo->id)) == 0)
412 	    break;
413     }
414     if (jo)
415 	error = journal_destroy(mp, jo, info->flags);
416     else
417 	error = EINVAL;
418     return (error);
419 }
420 
421 /*
422  * Remove all journals associated with a mount point.  Usually called
423  * by the umount code.
424  */
425 void
426 journal_remove_all_journals(struct mount *mp, int flags)
427 {
428     struct journal *jo;
429 
430     while ((jo = TAILQ_FIRST(&mp->mnt_jlist)) != NULL) {
431 	journal_destroy(mp, jo, flags);
432     }
433 }
434 
435 static int
436 journal_destroy(struct mount *mp, struct journal *jo, int flags)
437 {
438     struct jrecord jrec;
439 
440     TAILQ_REMOVE(&mp->mnt_jlist, jo, jentry);
441 
442     jrecord_init(jo, &jrec, JREC_STREAMID_DISCONT);
443     jrecord_write(&jrec, JTYPE_DISASSOCIATE, 0);
444     jrecord_done(&jrec, 0);
445 
446     journal_destroy_threads(jo, flags);
447 
448     if (jo->fp)
449 	fdrop(jo->fp);
450     if (jo->fifo.membase)
451 	kfree(jo->fifo.membase, M_JFIFO);
452     kfree(jo, M_JOURNAL);
453 
454     return(0);
455 }
456 
457 static int
458 journal_resync_vfs_journal(struct mount *mp, const void *ctl)
459 {
460     return(EINVAL);
461 }
462 
463 static int
464 journal_status_vfs_journal(struct mount *mp,
465 		       const struct mountctl_status_journal *info,
466 		       struct mountctl_journal_ret_status *rstat,
467 		       int buflen, int *res)
468 {
469     struct journal *jo;
470     int error = 0;
471     int index;
472 
473     index = 0;
474     *res = 0;
475     TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
476 	if (info->index == MC_JOURNAL_INDEX_ID) {
477 	    if (bcmp(jo->id, info->id, sizeof(jo->id)) != 0)
478 		continue;
479 	} else if (info->index >= 0) {
480 	    if (info->index < index)
481 		continue;
482 	} else if (info->index != MC_JOURNAL_INDEX_ALL) {
483 	    continue;
484 	}
485 	if (buflen < sizeof(*rstat)) {
486 	    if (*res)
487 		rstat[-1].flags |= MC_JOURNAL_STATUS_MORETOCOME;
488 	    else
489 		error = EINVAL;
490 	    break;
491 	}
492 	bzero(rstat, sizeof(*rstat));
493 	rstat->recsize = sizeof(*rstat);
494 	bcopy(jo->id, rstat->id, sizeof(jo->id));
495 	rstat->index = index;
496 	rstat->membufsize = jo->fifo.size;
497 	rstat->membufused = jo->fifo.windex - jo->fifo.xindex;
498 	rstat->membufunacked = jo->fifo.rindex - jo->fifo.xindex;
499 	rstat->bytessent = jo->total_acked;
500 	rstat->fifostalls = jo->fifostalls;
501 	++rstat;
502 	++index;
503 	*res += sizeof(*rstat);
504 	buflen -= sizeof(*rstat);
505     }
506     return(error);
507 }
508 
509 /************************************************************************
510  *			PARALLEL TRANSACTION SUPPORT ROUTINES		*
511  ************************************************************************
512  *
513  * JRECLIST_*() - routines which create and iterate over jrecord structures,
514  *		  because a mount point may have multiple attached journals.
515  */
516 
517 /*
518  * Initialize the passed jrecord_list and create a jrecord for each
519  * journal we need to write to.  Unnecessary mallocs are avoided by
520  * using the passed jrecord structure as the first jrecord in the list.
521  * A starting transaction is pushed for each jrecord.
522  *
523  * Returns non-zero if any of the journals require undo records.
524  */
525 static
526 int
527 jreclist_init(struct mount *mp, struct jrecord_list *jreclist,
528 	      struct jrecord *jreccache, int16_t rectype)
529 {
530     struct journal *jo;
531     struct jrecord *jrec;
532     int wantrev;
533     int count;
534     int16_t streamid;
535 
536     TAILQ_INIT(&jreclist->list);
537 
538     /*
539      * Select the stream ID to use for the transaction.  We must select
540      * a stream ID that is not currently in use by some other parallel
541      * transaction.
542      *
543      * Don't bother calculating the next streamid when reassigning
544      * mnt_streamid, since parallel transactions are fairly rare.  This
545      * also allows someone observing the raw records to clearly see
546      * when parallel transactions occur.
547      */
548     streamid = mp->mnt_streamid;
549     count = 0;
550     while (mp->mnt_jbitmap[streamid >> 3] & (1 << (streamid & 7))) {
551 	if (++streamid == JREC_STREAMID_JMAX)
552 		streamid = JREC_STREAMID_JMIN;
553 	if (++count == JREC_STREAMID_JMAX - JREC_STREAMID_JMIN) {
554 		kprintf("jreclist_init: all streamid's in use! sleeping\n");
555 		tsleep(jreclist, 0, "jsidfl", hz * 10);
556 		count = 0;
557 	}
558     }
559     mp->mnt_jbitmap[streamid >> 3] |= 1 << (streamid & 7);
560     mp->mnt_streamid = streamid;
561     jreclist->streamid = streamid;
562 
563     /*
564      * Now initialize a stream on each journal.
565      */
566     count = 0;
567     wantrev = 0;
568     TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
569 	if (count == 0)
570 	    jrec = jreccache;
571 	else
572 	    jrec = kmalloc(sizeof(*jrec), M_JOURNAL, M_WAITOK);
573 	jrecord_init(jo, jrec, streamid);
574 	jrec->user_save = jrecord_push(jrec, rectype);
575 	TAILQ_INSERT_TAIL(&jreclist->list, jrec, user_entry);
576 	if (jo->flags & MC_JOURNAL_WANT_REVERSABLE)
577 	    wantrev = 1;
578 	++count;
579     }
580     return(wantrev);
581 }
582 
583 /*
584  * Terminate the journaled transactions started by jreclist_init().  If
585  * an error occured, the transaction records will be aborted.
586  */
587 static
588 void
589 jreclist_done(struct mount *mp, struct jrecord_list *jreclist, int error)
590 {
591     struct jrecord *jrec;
592     int count;
593 
594     /*
595      * Cleanup the jrecord state on each journal.
596      */
597     TAILQ_FOREACH(jrec, &jreclist->list, user_entry) {
598 	jrecord_pop(jrec, jrec->user_save);
599 	jrecord_done(jrec, error);
600     }
601 
602     /*
603      * Free allocated jrec's (the first is always supplied)
604      */
605     count = 0;
606     while ((jrec = TAILQ_FIRST(&jreclist->list)) != NULL) {
607 	TAILQ_REMOVE(&jreclist->list, jrec, user_entry);
608 	if (count)
609 	    kfree(jrec, M_JOURNAL);
610 	++count;
611     }
612 
613     /*
614      * Clear the streamid so it can be reused.
615      */
616     mp->mnt_jbitmap[jreclist->streamid >> 3] &= ~(1 << (jreclist->streamid & 7));
617 }
618 
619 /*
620  * This procedure writes out UNDO records for available reversable
621  * journals.
622  *
623  * XXX could use improvement.  There is no need to re-read the file
624  * for each journal.
625  */
626 static
627 void
628 jreclist_undo_file(struct jrecord_list *jreclist, struct vnode *vp,
629 		   int jrflags, off_t off, off_t bytes)
630 {
631     struct jrecord *jrec;
632     int error;
633 
634     error = 0;
635     if (jrflags & JRUNDO_GETVP)
636 	error = vget(vp, LK_SHARED);
637     if (error == 0) {
638 	TAILQ_FOREACH(jrec, &jreclist->list, user_entry) {
639 	    if (jrec->jo->flags & MC_JOURNAL_WANT_REVERSABLE) {
640 		jrecord_undo_file(jrec, vp, jrflags, off, bytes);
641 	    }
642 	}
643     }
644     if (error == 0 && jrflags & JRUNDO_GETVP)
645 	vput(vp);
646 }
647 
648 /************************************************************************
649  *			LOW LEVEL UNDO SUPPORT ROUTINE			*
650  ************************************************************************
651  *
652  * This function is used to support UNDO records.  It will generate an
653  * appropriate record with the requested portion of the file data.  Note
654  * that file data is only recorded if JRUNDO_FILEDATA is passed.  If bytes
655  * is -1, it will be set to the size of the file.
656  */
657 static void
658 jrecord_undo_file(struct jrecord *jrec, struct vnode *vp, int jrflags,
659 		  off_t off, off_t bytes)
660 {
661     struct vattr attr;
662     void *save1; /* warning, save pointers do not always remain valid */
663     void *save2;
664     int error;
665 
666     /*
667      * Setup.  Start the UNDO record, obtain a shared lock on the vnode,
668      * and retrieve attribute info.
669      */
670     save1 = jrecord_push(jrec, JTYPE_UNDO);
671     error = VOP_GETATTR(vp, &attr);
672     if (error)
673 	goto done;
674 
675     /*
676      * Generate UNDO records as requested.
677      */
678     if (jrflags & JRUNDO_VATTR) {
679 	save2 = jrecord_push(jrec, JTYPE_VATTR);
680 	jrecord_leaf(jrec, JLEAF_VTYPE, &attr.va_type, sizeof(attr.va_type));
681 	if ((jrflags & JRUNDO_NLINK) && attr.va_nlink != VNOVAL)
682 	    jrecord_leaf(jrec, JLEAF_NLINK, &attr.va_nlink, sizeof(attr.va_nlink));
683 	if ((jrflags & JRUNDO_SIZE) && attr.va_size != VNOVAL)
684 	    jrecord_leaf(jrec, JLEAF_SIZE, &attr.va_size, sizeof(attr.va_size));
685 	if ((jrflags & JRUNDO_UID) && attr.va_uid != VNOVAL)
686 	    jrecord_leaf(jrec, JLEAF_UID, &attr.va_uid, sizeof(attr.va_uid));
687 	if ((jrflags & JRUNDO_GID) && attr.va_gid != VNOVAL)
688 	    jrecord_leaf(jrec, JLEAF_GID, &attr.va_gid, sizeof(attr.va_gid));
689 	if ((jrflags & JRUNDO_FSID) && attr.va_fsid != VNOVAL)
690 	    jrecord_leaf(jrec, JLEAF_FSID, &attr.va_fsid, sizeof(attr.va_fsid));
691 	if ((jrflags & JRUNDO_MODES) && attr.va_mode != (mode_t)VNOVAL)
692 	    jrecord_leaf(jrec, JLEAF_MODES, &attr.va_mode, sizeof(attr.va_mode));
693 	if ((jrflags & JRUNDO_INUM) && attr.va_fileid != VNOVAL)
694 	    jrecord_leaf(jrec, JLEAF_INUM, &attr.va_fileid, sizeof(attr.va_fileid));
695 	if ((jrflags & JRUNDO_ATIME) && attr.va_atime.tv_sec != VNOVAL)
696 	    jrecord_leaf(jrec, JLEAF_ATIME, &attr.va_atime, sizeof(attr.va_atime));
697 	if ((jrflags & JRUNDO_MTIME) && attr.va_mtime.tv_sec != VNOVAL)
698 	    jrecord_leaf(jrec, JLEAF_MTIME, &attr.va_mtime, sizeof(attr.va_mtime));
699 	if ((jrflags & JRUNDO_CTIME) && attr.va_ctime.tv_sec != VNOVAL)
700 	    jrecord_leaf(jrec, JLEAF_CTIME, &attr.va_ctime, sizeof(attr.va_ctime));
701 	if ((jrflags & JRUNDO_GEN) && attr.va_gen != VNOVAL)
702 	    jrecord_leaf(jrec, JLEAF_GEN, &attr.va_gen, sizeof(attr.va_gen));
703 	if ((jrflags & JRUNDO_FLAGS) && attr.va_flags != VNOVAL)
704 	    jrecord_leaf(jrec, JLEAF_FLAGS, &attr.va_flags, sizeof(attr.va_flags));
705 	if ((jrflags & JRUNDO_UDEV) && attr.va_rmajor != VNOVAL) {
706 	    udev_t rdev = makeudev(attr.va_rmajor, attr.va_rminor);
707 	    jrecord_leaf(jrec, JLEAF_UDEV, &rdev, sizeof(rdev));
708 	    jrecord_leaf(jrec, JLEAF_UMAJOR, &attr.va_rmajor, sizeof(attr.va_rmajor));
709 	    jrecord_leaf(jrec, JLEAF_UMINOR, &attr.va_rminor, sizeof(attr.va_rminor));
710 	}
711 	jrecord_pop(jrec, save2);
712     }
713 
714     /*
715      * Output the file data being overwritten by reading the file and
716      * writing it out to the journal prior to the write operation.  We
717      * do not need to write out data past the current file EOF.
718      *
719      * XXX support JRUNDO_CONDLINK - do not write out file data for files
720      * with a link count > 1.  The undo code needs to locate the inode and
721      * regenerate the hardlink.
722      */
723     if ((jrflags & JRUNDO_FILEDATA) && attr.va_type == VREG) {
724 	if (attr.va_size != VNOVAL) {
725 	    if (bytes == -1)
726 		bytes = attr.va_size - off;
727 	    if (off + bytes > attr.va_size)
728 		bytes = attr.va_size - off;
729 	    if (bytes > 0)
730 		jrecord_file_data(jrec, vp, off, bytes);
731 	} else {
732 	    error = EINVAL;
733 	}
734     }
735     if ((jrflags & JRUNDO_FILEDATA) && attr.va_type == VLNK) {
736 	struct iovec aiov;
737 	struct uio auio;
738 	char *buf;
739 
740 	buf = kmalloc(PATH_MAX, M_JOURNAL, M_WAITOK);
741 	aiov.iov_base = buf;
742 	aiov.iov_len = PATH_MAX;
743 	auio.uio_iov = &aiov;
744 	auio.uio_iovcnt = 1;
745 	auio.uio_offset = 0;
746 	auio.uio_rw = UIO_READ;
747 	auio.uio_segflg = UIO_SYSSPACE;
748 	auio.uio_td = curthread;
749 	auio.uio_resid = PATH_MAX;
750 	error = VOP_READLINK(vp, &auio, proc0.p_ucred);
751 	if (error == 0) {
752 		jrecord_leaf(jrec, JLEAF_SYMLINKDATA, buf,
753 				PATH_MAX - auio.uio_resid);
754 	}
755 	kfree(buf, M_JOURNAL);
756     }
757 done:
758     if (error)
759 	jrecord_leaf(jrec, JLEAF_ERROR, &error, sizeof(error));
760     jrecord_pop(jrec, save1);
761 }
762 
763 /************************************************************************
764  *			JOURNAL VNOPS					*
765  ************************************************************************
766  *
767  * These are function shims replacing the normal filesystem ops.  We become
768  * responsible for calling the underlying filesystem ops.  We have the choice
769  * of executing the underlying op first and then generating the journal entry,
770  * or starting the journal entry, executing the underlying op, and then
771  * either completing or aborting it.
772  *
773  * The journal is supposed to be a high-level entity, which generally means
774  * identifying files by name rather then by inode.  Supplying both allows
775  * the journal to be used both for inode-number-compatible 'mirrors' and
776  * for simple filesystem replication.
777  *
778  * Writes are particularly difficult to deal with because a single write may
779  * represent a hundred megabyte buffer or more, and both writes and truncations
780  * require the 'old' data to be written out as well as the new data if the
781  * log is reversable.  Other issues:
782  *
783  * - How to deal with operations on unlinked files (no path available),
784  *   but which may still be filesystem visible due to hard links.
785  *
786  * - How to deal with modifications made via a memory map.
787  *
788  * - Future cache coherency support will require cache coherency API calls
789  *   both prior to and after the call to the underlying VFS.
790  *
791  * ALSO NOTE: We do not have to shim compatibility VOPs like MKDIR which have
792  * new VFS equivalents (NMKDIR).
793  */
794 
795 /*
796  * Journal vop_setattr { a_vp, a_vap, a_cred }
797  */
798 static
799 int
800 journal_setattr(struct vop_setattr_args *ap)
801 {
802     struct jrecord_list jreclist;
803     struct jrecord jreccache;
804     struct jrecord *jrec;
805     struct mount *mp;
806     void *save;
807     int error;
808 
809     mp = ap->a_head.a_ops->head.vv_mount;
810     if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_SETATTR)) {
811 	jreclist_undo_file(&jreclist, ap->a_vp, JRUNDO_VATTR, 0, 0);
812     }
813     error = vop_journal_operate_ap(&ap->a_head);
814     if (error == 0) {
815 	TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
816 	    jrecord_write_cred(jrec, curthread, ap->a_cred);
817 	    jrecord_write_vnode_ref(jrec, ap->a_vp);
818 	    save = jrecord_push(jrec, JTYPE_REDO);
819 	    jrecord_write_vattr(jrec, ap->a_vap);
820 	    jrecord_pop(jrec, save);
821 	}
822     }
823     jreclist_done(mp, &jreclist, error);
824     return (error);
825 }
826 
827 /*
828  * Journal vop_write { a_vp, a_uio, a_ioflag, a_cred }
829  */
830 static
831 int
832 journal_write(struct vop_write_args *ap)
833 {
834     struct jrecord_list jreclist;
835     struct jrecord jreccache;
836     struct jrecord *jrec;
837     struct mount *mp;
838     struct uio uio_copy;
839     struct iovec uio_one_iovec;
840     void *save;
841     int error;
842 
843     /*
844      * Special synchronizing writes for VM backing store do not supply any
845      * real data
846      */
847     if (ap->a_uio->uio_segflg == UIO_NOCOPY) {
848 	    error = vop_journal_operate_ap(&ap->a_head);
849 	    return (error);
850     }
851 
852     /*
853      * This is really nasty.  UIO's don't retain sufficient information to
854      * be reusable once they've gone through the VOP chain.  The iovecs get
855      * cleared, so we have to copy the UIO.
856      *
857      * XXX fix the UIO code to not destroy iov's during a scan so we can
858      *     reuse the uio over and over again.
859      *
860      * XXX UNDO code needs to journal the old data prior to the write.
861      */
862     uio_copy = *ap->a_uio;
863     if (uio_copy.uio_iovcnt == 1) {
864 	uio_one_iovec = ap->a_uio->uio_iov[0];
865 	uio_copy.uio_iov = &uio_one_iovec;
866     } else {
867 	uio_copy.uio_iov = kmalloc(uio_copy.uio_iovcnt * sizeof(struct iovec),
868 				    M_JOURNAL, M_WAITOK);
869 	bcopy(ap->a_uio->uio_iov, uio_copy.uio_iov,
870 		uio_copy.uio_iovcnt * sizeof(struct iovec));
871     }
872 
873     /*
874      * Write out undo data.  Note that uio_offset is incorrect if
875      * IO_APPEND is set, but fortunately we have no undo file data to
876      * write out in that case.
877      */
878     mp = ap->a_head.a_ops->head.vv_mount;
879     if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_WRITE)) {
880 	if (ap->a_ioflag & IO_APPEND) {
881 	    jreclist_undo_file(&jreclist, ap->a_vp, JRUNDO_SIZE|JRUNDO_MTIME, 0, 0);
882 	} else {
883 	    jreclist_undo_file(&jreclist, ap->a_vp,
884 			       JRUNDO_FILEDATA|JRUNDO_SIZE|JRUNDO_MTIME,
885 			       uio_copy.uio_offset, uio_copy.uio_resid);
886 	}
887     }
888     error = vop_journal_operate_ap(&ap->a_head);
889 
890     /*
891      * XXX bad hack to figure out the offset for O_APPEND writes (note:
892      * uio field state after the VFS operation).
893      */
894     uio_copy.uio_offset = ap->a_uio->uio_offset -
895 			  (uio_copy.uio_resid - ap->a_uio->uio_resid);
896 
897     /*
898      * Output the write data to the journal.
899      */
900     if (error == 0) {
901 	TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
902 	    jrecord_write_cred(jrec, NULL, ap->a_cred);
903 	    jrecord_write_vnode_ref(jrec, ap->a_vp);
904 	    save = jrecord_push(jrec, JTYPE_REDO);
905 	    jrecord_write_uio(jrec, JLEAF_FILEDATA, &uio_copy);
906 	    jrecord_pop(jrec, save);
907 	}
908     }
909     jreclist_done(mp, &jreclist, error);
910 
911     if (uio_copy.uio_iov != &uio_one_iovec)
912 	kfree(uio_copy.uio_iov, M_JOURNAL);
913     return (error);
914 }
915 
916 /*
917  * Journal vop_fsync { a_vp, a_waitfor }
918  */
919 static
920 int
921 journal_fsync(struct vop_fsync_args *ap)
922 {
923 #if 0
924     struct mount *mp;
925     struct journal *jo;
926 #endif
927     int error;
928 
929     error = vop_journal_operate_ap(&ap->a_head);
930 #if 0
931     mp = ap->a_head.a_ops->head.vv_mount;
932     if (error == 0) {
933 	TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
934 	    /* XXX synchronize pending journal records */
935 	}
936     }
937 #endif
938     return (error);
939 }
940 
941 /*
942  * Journal vop_putpages { a_vp, a_m, a_count, a_sync, a_rtvals, a_offset }
943  *
944  * note: a_count is in bytes.
945  */
946 static
947 int
948 journal_putpages(struct vop_putpages_args *ap)
949 {
950     struct jrecord_list jreclist;
951     struct jrecord jreccache;
952     struct jrecord *jrec;
953     struct mount *mp;
954     void *save;
955     int error;
956 
957     mp = ap->a_head.a_ops->head.vv_mount;
958     if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_PUTPAGES) &&
959 	ap->a_count > 0
960     ) {
961 	jreclist_undo_file(&jreclist, ap->a_vp,
962 			   JRUNDO_FILEDATA|JRUNDO_SIZE|JRUNDO_MTIME,
963 			   ap->a_offset, btoc(ap->a_count));
964     }
965     error = vop_journal_operate_ap(&ap->a_head);
966     if (error == 0 && ap->a_count > 0) {
967 	TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
968 	    jrecord_write_vnode_ref(jrec, ap->a_vp);
969 	    save = jrecord_push(jrec, JTYPE_REDO);
970 	    jrecord_write_pagelist(jrec, JLEAF_FILEDATA, ap->a_m, ap->a_rtvals,
971 				   btoc(ap->a_count), ap->a_offset);
972 	    jrecord_pop(jrec, save);
973 	}
974     }
975     jreclist_done(mp, &jreclist, error);
976     return (error);
977 }
978 
979 /*
980  * Journal vop_setacl { a_vp, a_type, a_aclp, a_cred }
981  */
982 static
983 int
984 journal_setacl(struct vop_setacl_args *ap)
985 {
986     struct jrecord_list jreclist;
987     struct jrecord jreccache;
988     struct jrecord *jrec;
989     struct mount *mp;
990     int error;
991 
992     mp = ap->a_head.a_ops->head.vv_mount;
993     jreclist_init(mp, &jreclist, &jreccache, JTYPE_SETACL);
994     error = vop_journal_operate_ap(&ap->a_head);
995     if (error == 0) {
996 	TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
997 #if 0
998 	    if ((jo->flags & MC_JOURNAL_WANT_REVERSABLE))
999 		jrecord_undo_file(jrec, ap->a_vp, JRUNDO_XXX, 0, 0);
1000 #endif
1001 	    jrecord_write_cred(jrec, curthread, ap->a_cred);
1002 	    jrecord_write_vnode_ref(jrec, ap->a_vp);
1003 #if 0
1004 	    save = jrecord_push(jrec, JTYPE_REDO);
1005 	    /* XXX type, aclp */
1006 	    jrecord_pop(jrec, save);
1007 #endif
1008 	}
1009     }
1010     jreclist_done(mp, &jreclist, error);
1011     return (error);
1012 }
1013 
1014 /*
1015  * Journal vop_setextattr { a_vp, a_name, a_uio, a_cred }
1016  */
1017 static
1018 int
1019 journal_setextattr(struct vop_setextattr_args *ap)
1020 {
1021     struct jrecord_list jreclist;
1022     struct jrecord jreccache;
1023     struct jrecord *jrec;
1024     struct mount *mp;
1025     void *save;
1026     int error;
1027 
1028     mp = ap->a_head.a_ops->head.vv_mount;
1029     jreclist_init(mp, &jreclist, &jreccache, JTYPE_SETEXTATTR);
1030     error = vop_journal_operate_ap(&ap->a_head);
1031     if (error == 0) {
1032 	TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
1033 #if 0
1034 	    if ((jo->flags & MC_JOURNAL_WANT_REVERSABLE))
1035 		jrecord_undo_file(jrec, ap->a_vp, JRUNDO_XXX, 0, 0);
1036 #endif
1037 	    jrecord_write_cred(jrec, curthread, ap->a_cred);
1038 	    jrecord_write_vnode_ref(jrec, ap->a_vp);
1039 	    jrecord_leaf(jrec, JLEAF_ATTRNAME, ap->a_attrname,
1040 			strlen(ap->a_attrname));
1041 	    save = jrecord_push(jrec, JTYPE_REDO);
1042 	    jrecord_write_uio(jrec, JLEAF_FILEDATA, ap->a_uio);
1043 	    jrecord_pop(jrec, save);
1044 	}
1045     }
1046     jreclist_done(mp, &jreclist, error);
1047     return (error);
1048 }
1049 
1050 /*
1051  * Journal vop_ncreate { a_nch, a_vpp, a_cred, a_vap }
1052  */
1053 static
1054 int
1055 journal_ncreate(struct vop_ncreate_args *ap)
1056 {
1057     struct jrecord_list jreclist;
1058     struct jrecord jreccache;
1059     struct jrecord *jrec;
1060     struct mount *mp;
1061     void *save;
1062     int error;
1063 
1064     mp = ap->a_head.a_ops->head.vv_mount;
1065     jreclist_init(mp, &jreclist, &jreccache, JTYPE_CREATE);
1066     error = vop_journal_operate_ap(&ap->a_head);
1067     if (error == 0) {
1068 	TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
1069 	    jrecord_write_cred(jrec, NULL, ap->a_cred);
1070 	    jrecord_write_path(jrec, JLEAF_PATH1, ap->a_nch->ncp);
1071 	    if (*ap->a_vpp)
1072 		jrecord_write_vnode_ref(jrec, *ap->a_vpp);
1073 	    save = jrecord_push(jrec, JTYPE_REDO);
1074 	    jrecord_write_vattr(jrec, ap->a_vap);
1075 	    jrecord_pop(jrec, save);
1076 	}
1077     }
1078     jreclist_done(mp, &jreclist, error);
1079     return (error);
1080 }
1081 
1082 /*
1083  * Journal vop_nmknod { a_nch, a_vpp, a_cred, a_vap }
1084  */
1085 static
1086 int
1087 journal_nmknod(struct vop_nmknod_args *ap)
1088 {
1089     struct jrecord_list jreclist;
1090     struct jrecord jreccache;
1091     struct jrecord *jrec;
1092     struct mount *mp;
1093     void *save;
1094     int error;
1095 
1096     mp = ap->a_head.a_ops->head.vv_mount;
1097     jreclist_init(mp, &jreclist, &jreccache, JTYPE_MKNOD);
1098     error = vop_journal_operate_ap(&ap->a_head);
1099     if (error == 0) {
1100 	TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
1101 	    jrecord_write_cred(jrec, NULL, ap->a_cred);
1102 	    jrecord_write_path(jrec, JLEAF_PATH1, ap->a_nch->ncp);
1103 	    save = jrecord_push(jrec, JTYPE_REDO);
1104 	    jrecord_write_vattr(jrec, ap->a_vap);
1105 	    jrecord_pop(jrec, save);
1106 	    if (*ap->a_vpp)
1107 		jrecord_write_vnode_ref(jrec, *ap->a_vpp);
1108 	}
1109     }
1110     jreclist_done(mp, &jreclist, error);
1111     return (error);
1112 }
1113 
1114 /*
1115  * Journal vop_nlink { a_nch, a_vp, a_cred }
1116  */
1117 static
1118 int
1119 journal_nlink(struct vop_nlink_args *ap)
1120 {
1121     struct jrecord_list jreclist;
1122     struct jrecord jreccache;
1123     struct jrecord *jrec;
1124     struct mount *mp;
1125     void *save;
1126     int error;
1127 
1128     mp = ap->a_head.a_ops->head.vv_mount;
1129     jreclist_init(mp, &jreclist, &jreccache, JTYPE_LINK);
1130     error = vop_journal_operate_ap(&ap->a_head);
1131     if (error == 0) {
1132 	TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
1133 	    jrecord_write_cred(jrec, NULL, ap->a_cred);
1134 	    jrecord_write_path(jrec, JLEAF_PATH1, ap->a_nch->ncp);
1135 	    /* XXX PATH to VP and inode number */
1136 	    /* XXX this call may not record the correct path when
1137 	     * multiple paths are available */
1138 	    save = jrecord_push(jrec, JTYPE_REDO);
1139 	    jrecord_write_vnode_link(jrec, ap->a_vp, ap->a_nch->ncp);
1140 	    jrecord_pop(jrec, save);
1141 	}
1142     }
1143     jreclist_done(mp, &jreclist, error);
1144     return (error);
1145 }
1146 
1147 /*
1148  * Journal vop_symlink { a_nch, a_vpp, a_cred, a_vap, a_target }
1149  */
1150 static
1151 int
1152 journal_nsymlink(struct vop_nsymlink_args *ap)
1153 {
1154     struct jrecord_list jreclist;
1155     struct jrecord jreccache;
1156     struct jrecord *jrec;
1157     struct mount *mp;
1158     void *save;
1159     int error;
1160 
1161     mp = ap->a_head.a_ops->head.vv_mount;
1162     jreclist_init(mp, &jreclist, &jreccache, JTYPE_SYMLINK);
1163     error = vop_journal_operate_ap(&ap->a_head);
1164     if (error == 0) {
1165 	TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
1166 	    jrecord_write_cred(jrec, NULL, ap->a_cred);
1167 	    jrecord_write_path(jrec, JLEAF_PATH1, ap->a_nch->ncp);
1168 	    save = jrecord_push(jrec, JTYPE_REDO);
1169 	    jrecord_leaf(jrec, JLEAF_SYMLINKDATA,
1170 			ap->a_target, strlen(ap->a_target));
1171 	    jrecord_pop(jrec, save);
1172 	    if (*ap->a_vpp)
1173 		jrecord_write_vnode_ref(jrec, *ap->a_vpp);
1174 	}
1175     }
1176     jreclist_done(mp, &jreclist, error);
1177     return (error);
1178 }
1179 
1180 /*
1181  * Journal vop_nwhiteout { a_nch, a_cred, a_flags }
1182  */
1183 static
1184 int
1185 journal_nwhiteout(struct vop_nwhiteout_args *ap)
1186 {
1187     struct jrecord_list jreclist;
1188     struct jrecord jreccache;
1189     struct jrecord *jrec;
1190     struct mount *mp;
1191     int error;
1192 
1193     mp = ap->a_head.a_ops->head.vv_mount;
1194     jreclist_init(mp, &jreclist, &jreccache, JTYPE_WHITEOUT);
1195     error = vop_journal_operate_ap(&ap->a_head);
1196     if (error == 0) {
1197 	TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
1198 	    jrecord_write_cred(jrec, NULL, ap->a_cred);
1199 	    jrecord_write_path(jrec, JLEAF_PATH1, ap->a_nch->ncp);
1200 	}
1201     }
1202     jreclist_done(mp, &jreclist, error);
1203     return (error);
1204 }
1205 
1206 /*
1207  * Journal vop_nremove { a_nch, a_cred }
1208  */
1209 static
1210 int
1211 journal_nremove(struct vop_nremove_args *ap)
1212 {
1213     struct jrecord_list jreclist;
1214     struct jrecord jreccache;
1215     struct jrecord *jrec;
1216     struct mount *mp;
1217     int error;
1218 
1219     mp = ap->a_head.a_ops->head.vv_mount;
1220     if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_REMOVE) &&
1221 	ap->a_nch->ncp->nc_vp
1222     ) {
1223 	jreclist_undo_file(&jreclist, ap->a_nch->ncp->nc_vp,
1224 			   JRUNDO_ALL|JRUNDO_GETVP|JRUNDO_CONDLINK, 0, -1);
1225     }
1226     error = vop_journal_operate_ap(&ap->a_head);
1227     if (error == 0) {
1228 	TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
1229 	    jrecord_write_cred(jrec, NULL, ap->a_cred);
1230 	    jrecord_write_path(jrec, JLEAF_PATH1, ap->a_nch->ncp);
1231 	}
1232     }
1233     jreclist_done(mp, &jreclist, error);
1234     return (error);
1235 }
1236 
1237 /*
1238  * Journal vop_nmkdir { a_nch, a_vpp, a_cred, a_vap }
1239  */
1240 static
1241 int
1242 journal_nmkdir(struct vop_nmkdir_args *ap)
1243 {
1244     struct jrecord_list jreclist;
1245     struct jrecord jreccache;
1246     struct jrecord *jrec;
1247     struct mount *mp;
1248     int error;
1249 
1250     mp = ap->a_head.a_ops->head.vv_mount;
1251     jreclist_init(mp, &jreclist, &jreccache, JTYPE_MKDIR);
1252     error = vop_journal_operate_ap(&ap->a_head);
1253     if (error == 0) {
1254 	TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
1255 #if 0
1256 	    if (jo->flags & MC_JOURNAL_WANT_AUDIT) {
1257 		jrecord_write_audit(jrec);
1258 	    }
1259 #endif
1260 	    jrecord_write_path(jrec, JLEAF_PATH1, ap->a_nch->ncp);
1261 	    jrecord_write_cred(jrec, NULL, ap->a_cred);
1262 	    jrecord_write_vattr(jrec, ap->a_vap);
1263 	    jrecord_write_path(jrec, JLEAF_PATH1, ap->a_nch->ncp);
1264 	    if (*ap->a_vpp)
1265 		jrecord_write_vnode_ref(jrec, *ap->a_vpp);
1266 	}
1267     }
1268     jreclist_done(mp, &jreclist, error);
1269     return (error);
1270 }
1271 
1272 /*
1273  * Journal vop_nrmdir { a_nch, a_cred }
1274  */
1275 static
1276 int
1277 journal_nrmdir(struct vop_nrmdir_args *ap)
1278 {
1279     struct jrecord_list jreclist;
1280     struct jrecord jreccache;
1281     struct jrecord *jrec;
1282     struct mount *mp;
1283     int error;
1284 
1285     mp = ap->a_head.a_ops->head.vv_mount;
1286     if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_RMDIR)) {
1287 	jreclist_undo_file(&jreclist, ap->a_nch->ncp->nc_vp,
1288 			   JRUNDO_VATTR|JRUNDO_GETVP, 0, 0);
1289     }
1290     error = vop_journal_operate_ap(&ap->a_head);
1291     if (error == 0) {
1292 	TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
1293 	    jrecord_write_cred(jrec, NULL, ap->a_cred);
1294 	    jrecord_write_path(jrec, JLEAF_PATH1, ap->a_nch->ncp);
1295 	}
1296     }
1297     jreclist_done(mp, &jreclist, error);
1298     return (error);
1299 }
1300 
1301 /*
1302  * Journal vop_nrename { a_fnch, a_tnch, a_cred }
1303  */
1304 static
1305 int
1306 journal_nrename(struct vop_nrename_args *ap)
1307 {
1308     struct jrecord_list jreclist;
1309     struct jrecord jreccache;
1310     struct jrecord *jrec;
1311     struct mount *mp;
1312     int error;
1313 
1314     mp = ap->a_head.a_ops->head.vv_mount;
1315     if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_RENAME) &&
1316 	ap->a_tnch->ncp->nc_vp
1317     ) {
1318 	jreclist_undo_file(&jreclist, ap->a_tnch->ncp->nc_vp,
1319 			   JRUNDO_ALL|JRUNDO_GETVP|JRUNDO_CONDLINK, 0, -1);
1320     }
1321     error = vop_journal_operate_ap(&ap->a_head);
1322     if (error == 0) {
1323 	TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
1324 	    jrecord_write_cred(jrec, NULL, ap->a_cred);
1325 	    jrecord_write_path(jrec, JLEAF_PATH1, ap->a_fnch->ncp);
1326 	    jrecord_write_path(jrec, JLEAF_PATH2, ap->a_tnch->ncp);
1327 	}
1328     }
1329     jreclist_done(mp, &jreclist, error);
1330     return (error);
1331 }
1332 
1333