Style(9):
[dragonfly.git] / sys / kern / vfs_jops.c
CommitLineData
6ddb7618
MD
1/*
2 * Copyright (c) 2004 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 *
b2f7ec6c 34 * $DragonFly: src/sys/kern/vfs_jops.c,v 1.7 2005/02/28 17:41:00 dillon Exp $
2281065e
MD
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 * The journaling protocol is intended to evolve into a two-way stream
49 * whereby transaction IDs can be acknowledged by the journaling target
50 * when the data has been committed to hard storage. Both implicit and
51 * explicit acknowledgement schemes will be supported, depending on the
52 * sophistication of the journaling stream, plus resynchronization and
53 * restart when a journaling stream is interrupted. This information will
54 * also be made available to journaling-aware filesystems to allow better
55 * management of their own physical storage synchronization mechanisms as
56 * well as to allow such filesystems to take direct advantage of the kernel's
57 * journaling layer so they don't have to roll their own.
58 *
82eaef15 59 * In addition, the worker thread will have access to much larger
2281065e
MD
60 * spooling areas then the memory buffer is able to provide by e.g.
61 * reserving swap space, in order to absorb potentially long interruptions
62 * of off-site journaling streams, and to prevent 'slow' off-site linkages
63 * from radically slowing down local filesystem operations.
64 *
65 * Because of the non-trivial algorithms the journaling system will be
66 * required to support, use of a worker thread is mandatory. Efficiencies
67 * are maintained by utilitizing the memory FIFO to batch transactions when
68 * possible, reducing the number of gratuitous thread switches and taking
69 * advantage of cpu caches through the use of shorter batched code paths
70 * rather then trying to do everything in the context of the process
82eaef15
MD
71 * originating the filesystem op. In the future the memory FIFO can be
72 * made per-cpu to remove BGL or other locking requirements.
6ddb7618 73 */
6ddb7618
MD
74#include <sys/param.h>
75#include <sys/systm.h>
76#include <sys/buf.h>
77#include <sys/conf.h>
78#include <sys/kernel.h>
82eaef15 79#include <sys/queue.h>
6ddb7618
MD
80#include <sys/lock.h>
81#include <sys/malloc.h>
82#include <sys/mount.h>
83#include <sys/unistd.h>
84#include <sys/vnode.h>
85#include <sys/poll.h>
2281065e 86#include <sys/mountctl.h>
b2f7ec6c 87#include <sys/journal.h>
2281065e 88#include <sys/file.h>
b2f7ec6c 89#include <sys/proc.h>
6ddb7618
MD
90
91#include <machine/limits.h>
92
93#include <vm/vm.h>
94#include <vm/vm_object.h>
95#include <vm/vm_page.h>
96#include <vm/vm_pager.h>
97#include <vm/vnode_pager.h>
98
2281065e
MD
99#include <sys/file2.h>
100#include <sys/thread2.h>
101
102static int journal_attach(struct mount *mp);
103static void journal_detach(struct mount *mp);
104static int journal_install_vfs_journal(struct mount *mp, struct file *fp,
105 const struct mountctl_install_journal *info);
106static int journal_remove_vfs_journal(struct mount *mp,
107 const struct mountctl_remove_journal *info);
108static int journal_resync_vfs_journal(struct mount *mp, const void *ctl);
39b13188
MD
109static int journal_status_vfs_journal(struct mount *mp,
110 const struct mountctl_status_journal *info,
111 struct mountctl_journal_ret_status *rstat,
112 int buflen, int *res);
2281065e 113static void journal_thread(void *info);
82eaef15
MD
114
115static void *journal_reserve(struct journal *jo,
116 struct journal_rawrecbeg **rawpp,
117 int16_t streamid, int bytes);
118static void *journal_extend(struct journal *jo,
119 struct journal_rawrecbeg **rawpp,
120 int truncbytes, int bytes, int *newstreamrecp);
121static void journal_abort(struct journal *jo,
122 struct journal_rawrecbeg **rawpp);
123static void journal_commit(struct journal *jo,
124 struct journal_rawrecbeg **rawpp,
125 int bytes, int closeout);
126
127static void jrecord_init(struct journal *jo,
128 struct jrecord *jrec, int16_t streamid);
129static struct journal_subrecord *jrecord_push(
130 struct jrecord *jrec, int16_t rectype);
131static void jrecord_pop(struct jrecord *jrec, struct journal_subrecord *parent);
132static struct journal_subrecord *jrecord_write(struct jrecord *jrec,
133 int16_t rectype, int bytes);
134static void jrecord_data(struct jrecord *jrec, const void *buf, int bytes);
135static void jrecord_done(struct jrecord *jrec, int abortit);
136
558b8e00
MD
137static int journal_setattr(struct vop_setattr_args *ap);
138static int journal_write(struct vop_write_args *ap);
139static int journal_fsync(struct vop_fsync_args *ap);
140static int journal_putpages(struct vop_putpages_args *ap);
141static int journal_setacl(struct vop_setacl_args *ap);
142static int journal_setextattr(struct vop_setextattr_args *ap);
143static int journal_ncreate(struct vop_ncreate_args *ap);
144static int journal_nmknod(struct vop_nmknod_args *ap);
145static int journal_nlink(struct vop_nlink_args *ap);
146static int journal_nsymlink(struct vop_nsymlink_args *ap);
147static int journal_nwhiteout(struct vop_nwhiteout_args *ap);
148static int journal_nremove(struct vop_nremove_args *ap);
2281065e 149static int journal_nmkdir(struct vop_nmkdir_args *ap);
558b8e00
MD
150static int journal_nrmdir(struct vop_nrmdir_args *ap);
151static int journal_nrename(struct vop_nrename_args *ap);
2281065e 152
6ddb7618 153static struct vnodeopv_entry_desc journal_vnodeop_entries[] = {
2281065e
MD
154 { &vop_default_desc, vop_journal_operate_ap },
155 { &vop_mountctl_desc, (void *)journal_mountctl },
558b8e00
MD
156 { &vop_setattr_desc, (void *)journal_setattr },
157 { &vop_write_desc, (void *)journal_write },
158 { &vop_fsync_desc, (void *)journal_fsync },
159 { &vop_putpages_desc, (void *)journal_putpages },
160 { &vop_setacl_desc, (void *)journal_setacl },
161 { &vop_setextattr_desc, (void *)journal_setextattr },
162 { &vop_ncreate_desc, (void *)journal_ncreate },
163 { &vop_nmknod_desc, (void *)journal_nmknod },
164 { &vop_nlink_desc, (void *)journal_nlink },
165 { &vop_nsymlink_desc, (void *)journal_nsymlink },
166 { &vop_nwhiteout_desc, (void *)journal_nwhiteout },
167 { &vop_nremove_desc, (void *)journal_nremove },
2281065e 168 { &vop_nmkdir_desc, (void *)journal_nmkdir },
558b8e00
MD
169 { &vop_nrmdir_desc, (void *)journal_nrmdir },
170 { &vop_nrename_desc, (void *)journal_nrename },
2281065e 171 { NULL, NULL }
6ddb7618
MD
172};
173
82eaef15 174static MALLOC_DEFINE(M_JOURNAL, "journal", "Journaling structures");
2281065e
MD
175static MALLOC_DEFINE(M_JFIFO, "journal-fifo", "Journal FIFO");
176
6ddb7618 177int
2281065e
MD
178journal_mountctl(struct vop_mountctl_args *ap)
179{
180 struct mount *mp;
181 int error = 0;
182
183 mp = ap->a_head.a_ops->vv_mount;
184 KKASSERT(mp);
185
186 if (mp->mnt_vn_journal_ops == NULL) {
187 switch(ap->a_op) {
188 case MOUNTCTL_INSTALL_VFS_JOURNAL:
189 error = journal_attach(mp);
190 if (error == 0 && ap->a_ctllen != sizeof(struct mountctl_install_journal))
191 error = EINVAL;
192 if (error == 0 && ap->a_fp == NULL)
193 error = EBADF;
194 if (error == 0)
195 error = journal_install_vfs_journal(mp, ap->a_fp, ap->a_ctl);
196 if (TAILQ_EMPTY(&mp->mnt_jlist))
197 journal_detach(mp);
198 break;
199 case MOUNTCTL_REMOVE_VFS_JOURNAL:
200 case MOUNTCTL_RESYNC_VFS_JOURNAL:
39b13188
MD
201 case MOUNTCTL_STATUS_VFS_JOURNAL:
202 error = ENOENT;
2281065e
MD
203 break;
204 default:
205 error = EOPNOTSUPP;
206 break;
207 }
208 } else {
209 switch(ap->a_op) {
210 case MOUNTCTL_INSTALL_VFS_JOURNAL:
211 if (ap->a_ctllen != sizeof(struct mountctl_install_journal))
212 error = EINVAL;
213 if (error == 0 && ap->a_fp == NULL)
214 error = EBADF;
215 if (error == 0)
216 error = journal_install_vfs_journal(mp, ap->a_fp, ap->a_ctl);
217 break;
218 case MOUNTCTL_REMOVE_VFS_JOURNAL:
219 if (ap->a_ctllen != sizeof(struct mountctl_remove_journal))
220 error = EINVAL;
221 if (error == 0)
222 error = journal_remove_vfs_journal(mp, ap->a_ctl);
223 if (TAILQ_EMPTY(&mp->mnt_jlist))
224 journal_detach(mp);
225 break;
226 case MOUNTCTL_RESYNC_VFS_JOURNAL:
227 if (ap->a_ctllen != 0)
228 error = EINVAL;
229 error = journal_resync_vfs_journal(mp, ap->a_ctl);
230 break;
39b13188
MD
231 case MOUNTCTL_STATUS_VFS_JOURNAL:
232 if (ap->a_ctllen != sizeof(struct mountctl_status_journal))
233 error = EINVAL;
234 if (error == 0) {
235 error = journal_status_vfs_journal(mp, ap->a_ctl,
236 ap->a_buf, ap->a_buflen, ap->a_res);
237 }
238 break;
2281065e
MD
239 default:
240 error = EOPNOTSUPP;
241 break;
242 }
243 }
244 return (error);
245}
246
247/*
248 * High level mount point setup. When a
249 */
250static int
6ddb7618
MD
251journal_attach(struct mount *mp)
252{
2281065e
MD
253 vfs_add_vnodeops(mp, &mp->mnt_vn_journal_ops, journal_vnodeop_entries);
254 return(0);
6ddb7618
MD
255}
256
2281065e 257static void
6ddb7618
MD
258journal_detach(struct mount *mp)
259{
2281065e
MD
260 if (mp->mnt_vn_journal_ops)
261 vfs_rm_vnodeops(&mp->mnt_vn_journal_ops);
262}
263
264/*
82eaef15
MD
265 * Install a journal on a mount point. Each journal has an associated worker
266 * thread which is responsible for buffering and spooling the data to the
267 * target. A mount point may have multiple journals attached to it. An
268 * initial start record is generated when the journal is associated.
2281065e
MD
269 */
270static int
271journal_install_vfs_journal(struct mount *mp, struct file *fp,
272 const struct mountctl_install_journal *info)
273{
274 struct journal *jo;
82eaef15 275 struct jrecord jrec;
2281065e
MD
276 int error = 0;
277 int size;
278
279 jo = malloc(sizeof(struct journal), M_JOURNAL, M_WAITOK|M_ZERO);
280 bcopy(info->id, jo->id, sizeof(jo->id));
281 jo->flags = info->flags & ~(MC_JOURNAL_ACTIVE | MC_JOURNAL_STOP_REQ);
282
283 /*
284 * Memory FIFO size, round to nearest power of 2
285 */
82eaef15 286 if (info->membufsize) {
2281065e
MD
287 if (info->membufsize < 65536)
288 size = 65536;
289 else if (info->membufsize > 128 * 1024 * 1024)
290 size = 128 * 1024 * 1024;
291 else
292 size = (int)info->membufsize;
293 } else {
294 size = 1024 * 1024;
295 }
296 jo->fifo.size = 1;
297 while (jo->fifo.size < size)
298 jo->fifo.size <<= 1;
299
300 /*
301 * Other parameters. If not specified the starting transaction id
302 * will be the current date.
303 */
82eaef15 304 if (info->transid) {
2281065e
MD
305 jo->transid = info->transid;
306 } else {
307 struct timespec ts;
308 getnanotime(&ts);
309 jo->transid = ((int64_t)ts.tv_sec << 30) | ts.tv_nsec;
310 }
311
312 jo->fp = fp;
313
314 /*
315 * Allocate the memory FIFO
316 */
317 jo->fifo.mask = jo->fifo.size - 1;
318 jo->fifo.membase = malloc(jo->fifo.size, M_JFIFO, M_WAITOK|M_ZERO|M_NULLOK);
319 if (jo->fifo.membase == NULL)
320 error = ENOMEM;
321
82eaef15
MD
322 /*
323 * Create the worker thread and generate the association record.
324 */
2281065e
MD
325 if (error) {
326 free(jo, M_JOURNAL);
327 } else {
328 fhold(fp);
329 jo->flags |= MC_JOURNAL_ACTIVE;
330 lwkt_create(journal_thread, jo, NULL, &jo->thread,
331 TDF_STOPREQ, -1, "journal %.*s", JIDMAX, jo->id);
332 lwkt_setpri(&jo->thread, TDPRI_KERN_DAEMON);
333 lwkt_schedule(&jo->thread);
2281065e 334
82eaef15
MD
335 jrecord_init(jo, &jrec, JREC_STREAMID_DISCONT);
336 jrecord_write(&jrec, JTYPE_ASSOCIATE, 0);
337 jrecord_done(&jrec, 0);
2281065e
MD
338 TAILQ_INSERT_TAIL(&mp->mnt_jlist, jo, jentry);
339 }
340 return(error);
341}
342
82eaef15
MD
343/*
344 * Disassociate a journal from a mount point and terminate its worker thread.
345 * A final termination record is written out before the file pointer is
346 * dropped.
347 */
2281065e 348static int
82eaef15
MD
349journal_remove_vfs_journal(struct mount *mp,
350 const struct mountctl_remove_journal *info)
2281065e
MD
351{
352 struct journal *jo;
82eaef15 353 struct jrecord jrec;
2281065e
MD
354 int error;
355
356 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
357 if (bcmp(jo->id, info->id, sizeof(jo->id)) == 0)
358 break;
359 }
360 if (jo) {
361 error = 0;
362 TAILQ_REMOVE(&mp->mnt_jlist, jo, jentry);
82eaef15
MD
363
364 jrecord_init(jo, &jrec, JREC_STREAMID_DISCONT);
365 jrecord_write(&jrec, JTYPE_DISASSOCIATE, 0);
366 jrecord_done(&jrec, 0);
367
2281065e
MD
368 jo->flags |= MC_JOURNAL_STOP_REQ | (info->flags & MC_JOURNAL_STOP_IMM);
369 wakeup(&jo->fifo);
370 while (jo->flags & MC_JOURNAL_ACTIVE) {
371 tsleep(jo, 0, "jwait", 0);
372 }
373 lwkt_free_thread(&jo->thread); /* XXX SMP */
374 if (jo->fp)
375 fdrop(jo->fp, curthread);
376 if (jo->fifo.membase)
377 free(jo->fifo.membase, M_JFIFO);
378 free(jo, M_JOURNAL);
379 } else {
380 error = EINVAL;
381 }
382 return (error);
383}
384
385static int
386journal_resync_vfs_journal(struct mount *mp, const void *ctl)
387{
388 return(EINVAL);
389}
390
39b13188
MD
391static int
392journal_status_vfs_journal(struct mount *mp,
393 const struct mountctl_status_journal *info,
394 struct mountctl_journal_ret_status *rstat,
395 int buflen, int *res)
396{
397 struct journal *jo;
398 int error = 0;
399 int index;
400
401 index = 0;
402 *res = 0;
403 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
404 if (info->index == MC_JOURNAL_INDEX_ID) {
405 if (bcmp(jo->id, info->id, sizeof(jo->id)) != 0)
406 continue;
407 } else if (info->index >= 0) {
408 if (info->index < index)
409 continue;
410 } else if (info->index != MC_JOURNAL_INDEX_ALL) {
411 continue;
412 }
413 if (buflen < sizeof(*rstat)) {
414 if (*res)
415 rstat[-1].flags |= MC_JOURNAL_STATUS_MORETOCOME;
416 else
417 error = EINVAL;
418 break;
419 }
420 bzero(rstat, sizeof(*rstat));
421 rstat->recsize = sizeof(*rstat);
422 bcopy(jo->id, rstat->id, sizeof(jo->id));
423 rstat->index = index;
424 rstat->membufsize = jo->fifo.size;
425 rstat->membufused = jo->fifo.xindex - jo->fifo.rindex;
426 rstat->membufiopend = jo->fifo.windex - jo->fifo.rindex;
427 rstat->bytessent = jo->total_acked;
428 ++rstat;
429 ++index;
430 *res += sizeof(*rstat);
431 buflen -= sizeof(*rstat);
432 }
433 return(error);
434}
82eaef15
MD
435/*
436 * The per-journal worker thread is responsible for writing out the
437 * journal's FIFO to the target stream.
438 */
2281065e
MD
439static void
440journal_thread(void *info)
441{
442 struct journal *jo = info;
82eaef15 443 struct journal_rawrecbeg *rawp;
2281065e
MD
444 int bytes;
445 int error;
82eaef15 446 int avail;
2281065e
MD
447 int res;
448
449 for (;;) {
82eaef15
MD
450 /*
451 * Calculate the number of bytes available to write. This buffer
452 * area may contain reserved records so we can't just write it out
453 * without further checks.
454 */
455 bytes = jo->fifo.windex - jo->fifo.rindex;
456
457 /*
458 * sleep if no bytes are available or if an incomplete record is
459 * encountered (it needs to be filled in before we can write it
460 * out), and skip any pad records that we encounter.
461 */
462 if (bytes == 0) {
2281065e
MD
463 if (jo->flags & MC_JOURNAL_STOP_REQ)
464 break;
82eaef15
MD
465 tsleep(&jo->fifo, 0, "jfifo", hz);
466 continue;
467 }
468 rawp = (void *)(jo->fifo.membase + (jo->fifo.rindex & jo->fifo.mask));
469 if (rawp->begmagic == JREC_INCOMPLETEMAGIC) {
470 tsleep(&jo->fifo, 0, "jpad", hz);
471 continue;
472 }
473 if (rawp->streamid == JREC_STREAMID_PAD) {
474 jo->fifo.rindex += (rawp->recsize + 15) & ~15;
475 KKASSERT(jo->fifo.windex - jo->fifo.rindex > 0);
476 continue;
477 }
478
479 /*
480 * Figure out how much we can write out, beware the buffer wrap
481 * case.
482 */
483 res = 0;
484 avail = jo->fifo.size - (jo->fifo.rindex & jo->fifo.mask);
485 while (res < bytes && rawp->begmagic == JREC_BEGMAGIC) {
486 res += (rawp->recsize + 15) & ~15;
487 if (res >= avail) {
488 KKASSERT(res == avail);
489 break;
490 }
2281065e 491 }
82eaef15
MD
492
493 /*
494 * Issue the write and deal with any errors or other conditions.
495 * For now assume blocking I/O. Since we are record-aware the
496 * code cannot yet handle partial writes.
497 *
498 * XXX EWOULDBLOCK/NBIO
499 * XXX notification on failure
500 * XXX two-way acknowledgement stream in the return direction / xindex
501 */
558b8e00 502 printf("write @%d,%d\n", jo->fifo.rindex & jo->fifo.mask, bytes);
82eaef15
MD
503 bytes = res;
504 error = fp_write(jo->fp,
505 jo->fifo.membase + (jo->fifo.rindex & jo->fifo.mask),
506 bytes, &res);
2281065e
MD
507 if (error) {
508 printf("journal_thread(%s) write, error %d\n", jo->id, error);
82eaef15 509 /* XXX */
2281065e 510 } else {
82eaef15 511 KKASSERT(res == bytes);
2281065e 512 printf("journal_thread(%s) write %d\n", jo->id, res);
82eaef15
MD
513 }
514
515 /*
516 * Advance rindex. XXX for now also advance xindex, which will
517 * eventually be advanced when the target acknowledges the sequence
518 * space.
519 */
520 jo->fifo.rindex += bytes;
521 jo->fifo.xindex += bytes;
39b13188 522 jo->total_acked += bytes;
82eaef15
MD
523 if (jo->flags & MC_JOURNAL_WWAIT) {
524 jo->flags &= ~MC_JOURNAL_WWAIT; /* XXX hysteresis */
525 wakeup(&jo->fifo.windex);
2281065e
MD
526 }
527 }
528 jo->flags &= ~MC_JOURNAL_ACTIVE;
529 wakeup(jo);
530 wakeup(&jo->fifo.windex);
531}
532
82eaef15
MD
533static __inline
534void
535journal_build_pad(struct journal_rawrecbeg *rawp, int recsize)
2281065e 536{
82eaef15
MD
537 struct journal_rawrecend *rendp;
538
539 KKASSERT((recsize & 15) == 0 && recsize >= 16);
540
541 rawp->begmagic = JREC_BEGMAGIC;
542 rawp->streamid = JREC_STREAMID_PAD;
543 rawp->recsize = recsize; /* must be 16-byte aligned */
544 rawp->seqno = 0;
545 /*
546 * WARNING, rendp may overlap rawp->seqno. This is necessary to
547 * allow PAD records to fit in 16 bytes. Use cpu_mb1() to
548 * hopefully cause the compiler to not make any assumptions.
549 */
550 cpu_mb1();
551 rendp = (void *)((char *)rawp + rawp->recsize - sizeof(*rendp));
552 rendp->endmagic = JREC_ENDMAGIC;
553 rendp->check = 0;
554 rendp->recsize = rawp->recsize;
2281065e
MD
555}
556
82eaef15
MD
557/*
558 * Wake up the worker thread if the FIFO is more then half full or if
559 * someone is waiting for space to be freed up. Otherwise let the
560 * heartbeat deal with it. Being able to avoid waking up the worker
561 * is the key to the journal's cpu efficiency.
562 */
563static __inline
2281065e 564void
82eaef15 565journal_commit_wakeup(struct journal *jo)
2281065e
MD
566{
567 int avail;
568
82eaef15
MD
569 avail = jo->fifo.size - (jo->fifo.windex - jo->fifo.xindex);
570 KKASSERT(avail >= 0);
571 if ((avail < (jo->fifo.size >> 1)) || (jo->flags & MC_JOURNAL_WWAIT))
572 wakeup(&jo->fifo);
573}
574
575/*
576 * Create a new BEGIN stream record with the specified streamid and the
577 * specified amount of payload space. *rawpp will be set to point to the
578 * base of the new stream record and a pointer to the base of the payload
579 * space will be returned. *rawpp does not need to be pre-NULLd prior to
580 * making this call.
581 *
582 * A stream can be extended, aborted, or committed by other API calls
583 * below. This may result in a sequence of potentially disconnected
584 * stream records to be output to the journaling target. The first record
585 * (the one created by this function) will be marked JREC_STREAMCTL_BEGIN,
586 * while the last record on commit or abort will be marked JREC_STREAMCTL_END
587 * (and possibly also JREC_STREAMCTL_ABORTED). The last record could wind
588 * up being the same as the first, in which case the bits are all set in
589 * the first record.
590 *
591 * The stream record is created in an incomplete state by setting the begin
592 * magic to JREC_INCOMPLETEMAGIC. This prevents the worker thread from
593 * flushing the fifo past our record until we have finished populating it.
594 * Other threads can reserve and operate on their own space without stalling
595 * but the stream output will stall until we have completed operations. The
596 * memory FIFO is intended to be large enough to absorb such situations
597 * without stalling out other threads.
598 */
599static
600void *
601journal_reserve(struct journal *jo, struct journal_rawrecbeg **rawpp,
602 int16_t streamid, int bytes)
603{
604 struct journal_rawrecbeg *rawp;
605 int avail;
606 int availtoend;
607 int req;
608
609 /*
610 * Add header and trailer overheads to the passed payload. Note that
611 * the passed payload size need not be aligned in any way.
612 */
613 bytes += sizeof(struct journal_rawrecbeg);
614 bytes += sizeof(struct journal_rawrecend);
615
616 for (;;) {
617 /*
618 * First, check boundary conditions. If the request would wrap around
619 * we have to skip past the ending block and return to the beginning
620 * of the FIFO's buffer. Calculate 'req' which is the actual number
621 * of bytes being reserved, including wrap-around dead space.
622 *
623 * Note that availtoend is not truncated to avail and so cannot be
624 * used to determine whether the reservation is possible by itself.
625 * Also, since all fifo ops are 16-byte aligned, we can check
626 * the size before calculating the aligned size.
627 */
628 availtoend = jo->fifo.size - (jo->fifo.windex & jo->fifo.mask);
629 if (bytes > availtoend)
630 req = bytes + availtoend; /* add pad to end */
631 else
632 req = bytes;
633
634 /*
635 * Next calculate the total available space and see if it is
636 * sufficient. We cannot overwrite previously buffered data
637 * past xindex because otherwise we would not be able to restart
638 * a broken link at the target's last point of commit.
639 */
640 avail = jo->fifo.size - (jo->fifo.windex - jo->fifo.xindex);
641 KKASSERT(avail >= 0 && (avail & 15) == 0);
642
643 if (avail < req) {
644 /* XXX MC_JOURNAL_STOP_IMM */
2281065e
MD
645 jo->flags |= MC_JOURNAL_WWAIT;
646 tsleep(&jo->fifo.windex, 0, "jwrite", 0);
647 continue;
648 }
82eaef15
MD
649
650 /*
651 * Create a pad record for any dead space and create an incomplete
652 * record for the live space, then return a pointer to the
653 * contiguous buffer space that was requested.
654 *
655 * NOTE: The worker thread will not flush past an incomplete
656 * record, so the reserved space can be filled in at-will. The
657 * journaling code must also be aware the reserved sections occuring
658 * after this one will also not be written out even if completed
659 * until this one is completed.
660 */
661 rawp = (void *)(jo->fifo.membase + (jo->fifo.windex & jo->fifo.mask));
662 if (req != bytes) {
663 journal_build_pad(rawp, req - bytes);
664 rawp = (void *)jo->fifo.membase;
665 }
666 rawp->begmagic = JREC_INCOMPLETEMAGIC; /* updated by abort/commit */
667 rawp->recsize = bytes; /* (unaligned size) */
668 rawp->streamid = streamid | JREC_STREAMCTL_BEGIN;
669 rawp->seqno = 0; /* set by caller */
670
671 /*
672 * Issue a memory barrier to guarentee that the record data has been
673 * properly initialized before we advance the write index and return
674 * a pointer to the reserved record. Otherwise the worker thread
675 * could accidently run past us.
676 *
677 * Note that stream records are always 16-byte aligned.
678 */
679 cpu_mb1();
680 jo->fifo.windex += (req + 15) & ~15;
681 *rawpp = rawp;
682 return(rawp + 1);
683 }
684 /* not reached */
685 *rawpp = NULL;
686 return(NULL);
687}
688
689/*
690 * Extend a previous reservation by the specified number of payload bytes.
691 * If it is not possible to extend the existing reservation due to either
692 * another thread having reserved space after us or due to a boundary
693 * condition, the current reservation will be committed and possibly
694 * truncated and a new reservation with the specified payload size will
695 * be created. *rawpp is set to the new reservation in this case but the
696 * caller cannot depend on a comparison with the old rawp to determine if
697 * this case occurs because we could end up using the same memory FIFO
698 * offset for the new stream record.
699 *
700 * In either case this function will return a pointer to the base of the
701 * extended payload space.
702 *
703 * If a new stream block is created the caller needs to recalculate payload
704 * byte counts, if the same stream block is used the caller needs to extend
705 * its current notion of the payload byte count.
706 */
707static void *
708journal_extend(struct journal *jo, struct journal_rawrecbeg **rawpp,
709 int truncbytes, int bytes, int *newstreamrecp)
710{
711 struct journal_rawrecbeg *rawp;
712 int16_t streamid;
713 int availtoend;
714 int avail;
715 int osize;
716 int nsize;
717 int wbase;
718 void *rptr;
719
720 *newstreamrecp = 0;
721 rawp = *rawpp;
722 osize = (rawp->recsize + 15) & ~15;
723 nsize = (rawp->recsize + bytes + 15) & ~15;
724 wbase = (char *)rawp - jo->fifo.membase;
725
726 /*
727 * If the aligned record size does not change we can trivially extend
728 * the record.
729 */
730 if (nsize == osize) {
731 rawp->recsize += bytes;
732 return((char *)rawp + rawp->recsize - bytes);
733 }
734
735 /*
736 * If the fifo's write index hasn't been modified since we made the
737 * reservation and we do not hit any boundary conditions, we can
738 * trivially extend the record.
739 */
740 if ((jo->fifo.windex & jo->fifo.mask) == wbase + osize) {
741 availtoend = jo->fifo.size - wbase;
742 avail = jo->fifo.size - (jo->fifo.windex - jo->fifo.xindex) + osize;
743 KKASSERT((availtoend & 15) == 0);
744 KKASSERT((avail & 15) == 0);
745 if (nsize <= avail && nsize <= availtoend) {
746 jo->fifo.windex += nsize - osize;
747 rawp->recsize += bytes;
748 return((char *)rawp + rawp->recsize - bytes);
749 }
750 }
751
752 /*
753 * It was not possible to extend the buffer. Commit the current
754 * buffer and create a new one. We manually clear the BEGIN mark that
755 * journal_reserve() creates (because this is a continuing record, not
756 * the start of a new stream).
757 */
758 streamid = rawp->streamid & JREC_STREAMID_MASK;
759 journal_commit(jo, rawpp, truncbytes, 0);
760 rptr = journal_reserve(jo, rawpp, streamid, bytes);
761 rawp = *rawpp;
762 rawp->streamid &= ~JREC_STREAMCTL_BEGIN;
763 *newstreamrecp = 1;
764 return(rptr);
765}
766
767/*
768 * Abort a journal record. If the transaction record represents a stream
769 * BEGIN and we can reverse the fifo's write index we can simply reverse
770 * index the entire record, as if it were never reserved in the first place.
771 *
772 * Otherwise we set the JREC_STREAMCTL_ABORTED bit and commit the record
773 * with the payload truncated to 0 bytes.
774 */
775static void
776journal_abort(struct journal *jo, struct journal_rawrecbeg **rawpp)
777{
778 struct journal_rawrecbeg *rawp;
779 int osize;
780
781 rawp = *rawpp;
782 osize = (rawp->recsize + 15) & ~15;
783
784 if ((rawp->streamid & JREC_STREAMCTL_BEGIN) &&
785 (jo->fifo.windex & jo->fifo.mask) ==
786 (char *)rawp - jo->fifo.membase + osize)
787 {
788 jo->fifo.windex -= osize;
789 *rawpp = NULL;
790 } else {
791 rawp->streamid |= JREC_STREAMCTL_ABORTED;
792 journal_commit(jo, rawpp, 0, 1);
793 }
794}
795
796/*
797 * Commit a journal record and potentially truncate it to the specified
798 * number of payload bytes. If you do not want to truncate the record,
799 * simply pass -1 for the bytes parameter. Do not pass rawp->recsize, that
800 * field includes header and trailer and will not be correct. Note that
801 * passing 0 will truncate the entire data payload of the record.
802 *
803 * The logical stream is terminated by this function.
804 *
805 * If truncation occurs, and it is not possible to physically optimize the
806 * memory FIFO due to other threads having reserved space after ours,
807 * the remaining reserved space will be covered by a pad record.
808 */
809static void
810journal_commit(struct journal *jo, struct journal_rawrecbeg **rawpp,
811 int bytes, int closeout)
812{
813 struct journal_rawrecbeg *rawp;
814 struct journal_rawrecend *rendp;
815 int osize;
816 int nsize;
817
818 rawp = *rawpp;
819 *rawpp = NULL;
820
821 KKASSERT((char *)rawp >= jo->fifo.membase &&
822 (char *)rawp + rawp->recsize <= jo->fifo.membase + jo->fifo.size);
823 KKASSERT(((intptr_t)rawp & 15) == 0);
824
825 /*
826 * Truncate the record if requested. If the FIFO write index as still
827 * at the end of our record we can optimally backindex it. Otherwise
828 * we have to insert a pad record.
829 *
830 * We calculate osize which is the 16-byte-aligned original recsize.
831 * We calculate nsize which is the 16-byte-aligned new recsize.
832 *
833 * Due to alignment issues or in case the passed truncation bytes is
834 * the same as the original payload, windex will be equal to nindex.
835 */
836 if (bytes >= 0) {
837 KKASSERT(bytes >= 0 && bytes <= rawp->recsize - sizeof(struct journal_rawrecbeg) - sizeof(struct journal_rawrecend));
838 osize = (rawp->recsize + 15) & ~15;
839 rawp->recsize = bytes + sizeof(struct journal_rawrecbeg) +
840 sizeof(struct journal_rawrecend);
841 nsize = (rawp->recsize + 15) & ~15;
842 if (osize == nsize) {
843 /* do nothing */
844 } else if ((jo->fifo.windex & jo->fifo.mask) == (char *)rawp - jo->fifo.membase + osize) {
845 /* we are able to backindex the fifo */
846 jo->fifo.windex -= osize - nsize;
847 } else {
848 /* we cannot backindex the fifo, emplace a pad in the dead space */
849 journal_build_pad((void *)((char *)rawp + osize), osize - nsize);
850 }
2281065e 851 }
82eaef15
MD
852
853 /*
854 * Fill in the trailer. Note that unlike pad records, the trailer will
855 * never overlap the header.
856 */
857 rendp = (void *)((char *)rawp +
858 ((rawp->recsize + 15) & ~15) - sizeof(*rendp));
859 rendp->endmagic = JREC_ENDMAGIC;
860 rendp->recsize = rawp->recsize;
861 rendp->check = 0; /* XXX check word, disabled for now */
862
863 /*
864 * Fill in begmagic last. This will allow the worker thread to proceed.
865 * Use a memory barrier to guarentee write ordering. Mark the stream
866 * as terminated if closeout is set. This is the typical case.
867 */
868 if (closeout)
869 rawp->streamid |= JREC_STREAMCTL_END;
870 cpu_mb1(); /* memory barrier */
871 rawp->begmagic = JREC_BEGMAGIC;
872
873 journal_commit_wakeup(jo);
874}
875
876/************************************************************************
877 * TRANSACTION SUPPORT ROUTINES *
878 ************************************************************************
879 *
880 * JRECORD_*() - routines to create subrecord transactions and embed them
881 * in the logical streams managed by the journal_*() routines.
882 */
883
884static int16_t sid = JREC_STREAMID_JMIN;
885
886/*
887 * Initialize the passed jrecord structure and start a new stream transaction
888 * by reserving an initial build space in the journal's memory FIFO.
889 */
890static void
891jrecord_init(struct journal *jo, struct jrecord *jrec, int16_t streamid)
892{
893 bzero(jrec, sizeof(*jrec));
894 jrec->jo = jo;
895 if (streamid < 0) {
896 streamid = sid++; /* XXX need to track stream ids! */
897 if (sid == JREC_STREAMID_JMAX)
898 sid = JREC_STREAMID_JMIN;
899 }
900 jrec->streamid = streamid;
901 jrec->stream_residual = JREC_DEFAULTSIZE;
902 jrec->stream_reserved = jrec->stream_residual;
903 jrec->stream_ptr =
904 journal_reserve(jo, &jrec->rawp, streamid, jrec->stream_reserved);
905}
906
907/*
908 * Push a recursive record type. All pushes should have matching pops.
909 * The old parent is returned and the newly pushed record becomes the
910 * new parent. Note that the old parent's pointer may already be invalid
911 * or may become invalid if jrecord_write() had to build a new stream
912 * record, so the caller should not mess with the returned pointer in
913 * any way other then to save it.
914 */
915static
916struct journal_subrecord *
917jrecord_push(struct jrecord *jrec, int16_t rectype)
918{
919 struct journal_subrecord *save;
920
921 save = jrec->parent;
922 jrec->parent = jrecord_write(jrec, rectype|JMASK_NESTED, 0);
923 jrec->last = NULL;
924 KKASSERT(jrec->parent != NULL);
925 ++jrec->pushcount;
926 ++jrec->pushptrgood; /* cleared on flush */
927 return(save);
928}
929
930/*
931 * Pop a previously pushed sub-transaction. We must set JMASK_LAST
932 * on the last record written within the subtransaction. If the last
933 * record written is not accessible or if the subtransaction is empty,
934 * we must write out a pad record with JMASK_LAST set before popping.
935 *
936 * When popping a subtransaction the parent record's recsize field
937 * will be properly set. If the parent pointer is no longer valid
938 * (which can occur if the data has already been flushed out to the
939 * stream), the protocol spec allows us to leave it 0.
940 *
941 * The saved parent pointer which we restore may or may not be valid,
942 * and if not valid may or may not be NULL, depending on the value
943 * of pushptrgood.
944 */
945static void
946jrecord_pop(struct jrecord *jrec, struct journal_subrecord *save)
947{
948 struct journal_subrecord *last;
949
950 KKASSERT(jrec->pushcount > 0);
951 KKASSERT(jrec->residual == 0);
952
953 /*
954 * Set JMASK_LAST on the last record we wrote at the current
955 * level. If last is NULL we either no longer have access to the
956 * record or the subtransaction was empty and we must write out a pad
957 * record.
958 */
959 if ((last = jrec->last) == NULL) {
960 jrecord_write(jrec, JLEAF_PAD|JMASK_LAST, 0);
961 last = jrec->last; /* reload after possible flush */
962 } else {
963 last->rectype |= JMASK_LAST;
964 }
965
966 /*
967 * pushptrgood tells us how many levels of parent record pointers
968 * are valid. The jrec only stores the current parent record pointer
969 * (and it is only valid if pushptrgood != 0). The higher level parent
970 * record pointers are saved by the routines calling jrecord_push() and
971 * jrecord_pop(). These pointers may become stale and we determine
972 * that fact by tracking the count of valid parent pointers with
973 * pushptrgood. Pointers become invalid when their related stream
974 * record gets pushed out.
975 *
b2f7ec6c
MD
976 * If no pointer is available (the data has already been pushed out),
977 * then no fixup of e.g. the length field is possible for non-leaf
978 * nodes. The protocol allows for this situation by placing a larger
979 * burden on the program scanning the stream on the other end.
980 *
82eaef15
MD
981 * [parentA]
982 * [node X]
983 * [parentB]
984 * [node Y]
985 * [node Z]
986 * (pop B) see NOTE B
987 * (pop A) see NOTE A
988 *
989 * NOTE B: This pop sets LAST in node Z if the node is still accessible,
990 * else a PAD record is appended and LAST is set in that.
991 *
992 * This pop sets the record size in parentB if parentB is still
993 * accessible, else the record size is left 0 (the scanner must
994 * deal with that).
995 *
996 * This pop sets the new 'last' record to parentB, the pointer
997 * to which may or may not still be accessible.
998 *
999 * NOTE A: This pop sets LAST in parentB if the node is still accessible,
1000 * else a PAD record is appended and LAST is set in that.
1001 *
1002 * This pop sets the record size in parentA if parentA is still
1003 * accessible, else the record size is left 0 (the scanner must
1004 * deal with that).
1005 *
1006 * This pop sets the new 'last' record to parentA, the pointer
1007 * to which may or may not still be accessible.
1008 *
1009 * Also note that the last record in the stream transaction, which in
1010 * the above example is parentA, does not currently have the LAST bit
1011 * set.
1012 *
1013 * The current parent becomes the last record relative to the
1014 * saved parent passed into us. It's validity is based on
1015 * whether pushptrgood is non-zero prior to decrementing. The saved
1016 * parent becomes the new parent, and its validity is based on whether
1017 * pushptrgood is non-zero after decrementing.
1018 *
1019 * The old jrec->parent may be NULL if it is no longer accessible.
1020 * If pushptrgood is non-zero, however, it is guarenteed to not
1021 * be NULL (since no flush occured).
1022 */
1023 jrec->last = jrec->parent;
1024 --jrec->pushcount;
1025 if (jrec->pushptrgood) {
1026 KKASSERT(jrec->last != NULL && last != NULL);
1027 if (--jrec->pushptrgood == 0) {
1028 jrec->parent = NULL; /* 'save' contains garbage or NULL */
1029 } else {
1030 KKASSERT(save != NULL);
1031 jrec->parent = save; /* 'save' must not be NULL */
1032 }
1033
1034 /*
1035 * Set the record size in the old parent. 'last' still points to
1036 * the original last record in the subtransaction being popped,
1037 * jrec->last points to the old parent (which became the last
1038 * record relative to the new parent being popped into).
1039 */
1040 jrec->last->recsize = (char *)last + last->recsize - (char *)jrec->last;
1041 } else {
1042 jrec->parent = NULL;
1043 KKASSERT(jrec->last == NULL);
1044 }
1045}
1046
b2f7ec6c
MD
1047/*
1048 * Write out a leaf record, including associated data.
1049 */
1050static
1051void
1052jrecord_leaf(struct jrecord *jrec, int16_t rectype, void *ptr, int bytes)
1053{
1054 jrecord_write(jrec, rectype, bytes);
1055 jrecord_data(jrec, ptr, bytes);
1056 jrecord_done(jrec, 0);
1057}
1058
82eaef15
MD
1059/*
1060 * Write a leaf record out and return a pointer to its base. The leaf
1061 * record may contain potentially megabytes of data which is supplied
1062 * in jrecord_data() calls. The exact amount must be specified in this
1063 * call.
b2f7ec6c
MD
1064 *
1065 * THE RETURNED SUBRECORD POINTER IS ONLY VALID IMMEDIATELY AFTER THE
1066 * CALL AND MAY BECOME INVALID AT ANY TIME. ONLY THE PUSH/POP CODE SHOULD
1067 * USE THE RETURN VALUE.
82eaef15
MD
1068 */
1069static
1070struct journal_subrecord *
1071jrecord_write(struct jrecord *jrec, int16_t rectype, int bytes)
1072{
1073 struct journal_subrecord *last;
1074 int pusheditout;
1075
1076 /*
1077 * Try to catch some obvious errors. Nesting records must specify a
1078 * size of 0, and there should be no left-overs from previous operations
1079 * (such as incomplete data writeouts).
1080 */
1081 KKASSERT(bytes == 0 || (rectype & JMASK_NESTED) == 0);
1082 KKASSERT(jrec->residual == 0);
1083
1084 /*
1085 * Check to see if the current stream record has enough room for
1086 * the new subrecord header. If it doesn't we extend the current
1087 * stream record.
1088 *
1089 * This may have the side effect of pushing out the current stream record
1090 * and creating a new one. We must adjust our stream tracking fields
1091 * accordingly.
1092 */
1093 if (jrec->stream_residual < sizeof(struct journal_subrecord)) {
1094 jrec->stream_ptr = journal_extend(jrec->jo, &jrec->rawp,
1095 jrec->stream_reserved - jrec->stream_residual,
1096 JREC_DEFAULTSIZE, &pusheditout);
1097 if (pusheditout) {
1098 jrec->stream_reserved = JREC_DEFAULTSIZE;
1099 jrec->stream_residual = JREC_DEFAULTSIZE;
1100 jrec->parent = NULL; /* no longer accessible */
1101 jrec->pushptrgood = 0; /* restored parents in pops no good */
1102 } else {
1103 jrec->stream_reserved += JREC_DEFAULTSIZE;
1104 jrec->stream_residual += JREC_DEFAULTSIZE;
1105 }
1106 }
1107 last = (void *)jrec->stream_ptr;
1108 last->rectype = rectype;
1109 last->reserved = 0;
1110 last->recsize = sizeof(struct journal_subrecord) + bytes;
1111 jrec->last = last;
1112 jrec->residual = bytes; /* remaining data to be posted */
1113 jrec->residual_align = -bytes & 7; /* post-data alignment required */
1114 return(last);
1115}
1116
1117/*
1118 * Write out the data associated with a leaf record. Any number of calls
1119 * to this routine may be made as long as the byte count adds up to the
1120 * amount originally specified in jrecord_write().
1121 *
1122 * The act of writing out the leaf data may result in numerous stream records
1123 * being pushed out. Callers should be aware that even the associated
1124 * subrecord header may become inaccessible due to stream record pushouts.
1125 */
1126static void
1127jrecord_data(struct jrecord *jrec, const void *buf, int bytes)
1128{
1129 int pusheditout;
1130 int extsize;
1131
1132 KKASSERT(bytes >= 0 && bytes <= jrec->residual);
1133
1134 /*
1135 * Push out stream records as long as there is insufficient room to hold
1136 * the remaining data.
1137 */
1138 while (jrec->stream_residual < bytes) {
1139 /*
1140 * Fill in any remaining space in the current stream record.
1141 */
1142 bcopy(buf, jrec->stream_ptr, jrec->stream_residual);
1143 buf = (const char *)buf + jrec->stream_residual;
1144 bytes -= jrec->stream_residual;
1145 /*jrec->stream_ptr += jrec->stream_residual;*/
1146 jrec->stream_residual = 0;
1147 jrec->residual -= jrec->stream_residual;
1148
1149 /*
1150 * Try to extend the current stream record, but no more then 1/4
1151 * the size of the FIFO.
1152 */
1153 extsize = jrec->jo->fifo.size >> 2;
1154 if (extsize > bytes)
1155 extsize = (bytes + 15) & ~15;
1156
1157 jrec->stream_ptr = journal_extend(jrec->jo, &jrec->rawp,
1158 jrec->stream_reserved - jrec->stream_residual,
1159 extsize, &pusheditout);
1160 if (pusheditout) {
1161 jrec->stream_reserved = extsize;
1162 jrec->stream_residual = extsize;
1163 jrec->parent = NULL; /* no longer accessible */
1164 jrec->last = NULL; /* no longer accessible */
1165 jrec->pushptrgood = 0; /* restored parents in pops no good */
1166 } else {
1167 jrec->stream_reserved += extsize;
1168 jrec->stream_residual += extsize;
1169 }
1170 }
1171
1172 /*
1173 * Push out any remaining bytes into the current stream record.
1174 */
1175 if (bytes) {
1176 bcopy(buf, jrec->stream_ptr, bytes);
1177 jrec->stream_ptr += bytes;
1178 jrec->stream_residual -= bytes;
1179 jrec->residual -= bytes;
1180 }
1181
1182 /*
1183 * Handle data alignment requirements for the subrecord. Because the
1184 * stream record's data space is more strictly aligned, it must already
1185 * have sufficient space to hold any subrecord alignment slop.
1186 */
1187 if (jrec->residual == 0 && jrec->residual_align) {
1188 KKASSERT(jrec->residual_align <= jrec->stream_residual);
1189 bzero(jrec->stream_ptr, jrec->residual_align);
1190 jrec->stream_ptr += jrec->residual_align;
1191 jrec->stream_residual -= jrec->residual_align;
1192 jrec->residual_align = 0;
1193 }
1194}
1195
1196/*
1197 * We are finished with a transaction. If abortit is not set then we must
1198 * be at the top level with no residual subrecord data left to output.
1199 * If abortit is set then we can be in any state.
1200 *
1201 * The stream record will be committed or aborted as specified and jrecord
1202 * resources will be cleaned up.
1203 */
1204static void
1205jrecord_done(struct jrecord *jrec, int abortit)
1206{
1207 KKASSERT(jrec->rawp != NULL);
1208
1209 if (abortit) {
1210 journal_abort(jrec->jo, &jrec->rawp);
1211 } else {
1212 KKASSERT(jrec->pushcount == 0 && jrec->residual == 0);
1213 journal_commit(jrec->jo, &jrec->rawp,
1214 jrec->stream_reserved - jrec->stream_residual, 1);
1215 }
1216
1217 /*
1218 * jrec should not be used beyond this point without another init,
1219 * but clean up some fields to ensure that we panic if it is.
1220 *
1221 * Note that jrec->rawp is NULLd out by journal_abort/journal_commit.
1222 */
1223 jrec->jo = NULL;
1224 jrec->stream_ptr = NULL;
1225}
1226
1227/************************************************************************
b2f7ec6c 1228 * LOW LEVEL RECORD SUPPORT ROUTINES *
82eaef15
MD
1229 ************************************************************************
1230 *
b2f7ec6c
MD
1231 * These routine create low level recursive and leaf subrecords representing
1232 * common filesystem structures.
82eaef15
MD
1233 */
1234
b2f7ec6c
MD
1235/*
1236 * Write out a filename path relative to the base of the mount point.
1237 * rectype is typically JLEAF_PATH{1,2,3,4}.
1238 */
82eaef15
MD
1239static void
1240jrecord_write_path(struct jrecord *jrec, int16_t rectype, struct namecache *ncp)
1241{
b2f7ec6c
MD
1242 char buf[64]; /* local buffer if it fits, else malloced */
1243 char *base;
1244 int pathlen;
1245 int index;
1246 struct namecache *scan;
1247
1248 /*
1249 * Pass 1 - figure out the number of bytes required. Include terminating
1250 * \0 on last element and '/' separator on other elements.
1251 */
1252again:
1253 pathlen = 0;
1254 for (scan = ncp;
1255 scan && (scan->nc_flag & NCF_MOUNTPT) == 0;
1256 scan = scan->nc_parent
1257 ) {
1258 pathlen += scan->nc_nlen + 1;
1259 }
1260
1261 if (pathlen <= sizeof(buf))
1262 base = buf;
1263 else
1264 base = malloc(pathlen, M_TEMP, M_INTWAIT);
1265
1266 /*
1267 * Pass 2 - generate the path buffer
1268 */
1269 index = pathlen;
1270 for (scan = ncp;
1271 scan && (scan->nc_flag & NCF_MOUNTPT) == 0;
1272 scan = scan->nc_parent
1273 ) {
1274 if (scan->nc_nlen >= index) {
1275 if (base != buf)
1276 free(base, M_TEMP);
1277 goto again;
1278 }
1279 if (index == pathlen)
1280 base[--index] = 0;
1281 else
1282 base[--index] = '/';
1283 index -= scan->nc_nlen;
1284 bcopy(scan->nc_name, base + index, scan->nc_nlen);
1285 }
1286 jrecord_leaf(jrec, rectype, base + index, pathlen - index);
1287 if (base != buf)
1288 free(base, M_TEMP);
82eaef15
MD
1289}
1290
b2f7ec6c
MD
1291/*
1292 * Write out a file attribute structure. While somewhat inefficient, using
1293 * a recursive data structure is the most portable and extensible way.
1294 */
82eaef15
MD
1295static void
1296jrecord_write_vattr(struct jrecord *jrec, struct vattr *vat)
1297{
b2f7ec6c
MD
1298 void *save;
1299
1300 save = jrecord_push(jrec, JTYPE_VATTR);
1301 if (vat->va_type != VNON)
1302 jrecord_leaf(jrec, JLEAF_UID, &vat->va_type, sizeof(vat->va_type));
1303 if (vat->va_uid != VNOVAL)
1304 jrecord_leaf(jrec, JLEAF_UID, &vat->va_mode, sizeof(vat->va_mode));
1305 if (vat->va_nlink != VNOVAL)
1306 jrecord_leaf(jrec, JLEAF_NLINK, &vat->va_nlink, sizeof(vat->va_nlink));
1307 if (vat->va_uid != VNOVAL)
1308 jrecord_leaf(jrec, JLEAF_UID, &vat->va_uid, sizeof(vat->va_uid));
1309 if (vat->va_gid != VNOVAL)
1310 jrecord_leaf(jrec, JLEAF_GID, &vat->va_gid, sizeof(vat->va_gid));
1311 if (vat->va_fsid != VNOVAL)
1312 jrecord_leaf(jrec, JLEAF_FSID, &vat->va_fsid, sizeof(vat->va_fsid));
1313 if (vat->va_fileid != VNOVAL)
1314 jrecord_leaf(jrec, JLEAF_INUM, &vat->va_fileid, sizeof(vat->va_fileid));
1315 if (vat->va_size != VNOVAL)
1316 jrecord_leaf(jrec, JLEAF_SIZE, &vat->va_size, sizeof(vat->va_size));
1317 if (vat->va_atime.tv_sec != VNOVAL)
1318 jrecord_leaf(jrec, JLEAF_ATIME, &vat->va_atime, sizeof(vat->va_atime));
1319 if (vat->va_mtime.tv_sec != VNOVAL)
1320 jrecord_leaf(jrec, JLEAF_MTIME, &vat->va_mtime, sizeof(vat->va_mtime));
1321 if (vat->va_ctime.tv_sec != VNOVAL)
1322 jrecord_leaf(jrec, JLEAF_CTIME, &vat->va_ctime, sizeof(vat->va_ctime));
1323 if (vat->va_gen != VNOVAL)
1324 jrecord_leaf(jrec, JLEAF_GEN, &vat->va_gen, sizeof(vat->va_gen));
1325 if (vat->va_flags != VNOVAL)
1326 jrecord_leaf(jrec, JLEAF_FLAGS, &vat->va_flags, sizeof(vat->va_flags));
1327 if (vat->va_rdev != VNOVAL)
1328 jrecord_leaf(jrec, JLEAF_UDEV, &vat->va_rdev, sizeof(vat->va_rdev));
1329#if 0
1330 if (vat->va_filerev != VNOVAL)
1331 jrecord_leaf(jrec, JLEAF_FILEREV, &vat->va_filerev, sizeof(vat->va_filerev));
1332#endif
1333 jrecord_pop(jrec, save);
1334 jrecord_done(jrec, 0);
1335}
1336
1337/*
1338 * Write out the creds used to issue a file operation. If a process is
1339 * available write out additional tracking information related to the
1340 * process.
1341 *
1342 * XXX additional tracking info
1343 * XXX tty line info
1344 */
1345static void
1346jrecord_write_cred(struct jrecord *jrec, struct thread *td, struct ucred *cred)
1347{
1348 void *save;
1349 struct proc *p;
1350
1351 save = jrecord_push(jrec, JTYPE_CRED);
1352 jrecord_leaf(jrec, JLEAF_UID, &cred->cr_uid, sizeof(cred->cr_uid));
1353 jrecord_leaf(jrec, JLEAF_GID, &cred->cr_gid, sizeof(cred->cr_gid));
1354 if (td && (p = td->td_proc) != NULL) {
1355 jrecord_leaf(jrec, JLEAF_PID, &p->p_pid, sizeof(p->p_pid));
1356 jrecord_leaf(jrec, JLEAF_COMM, p->p_comm, sizeof(p->p_comm));
1357 }
1358 jrecord_pop(jrec, save);
1359 jrecord_done(jrec, 0);
1360}
1361
1362/*
1363 * Write out information required to identify a vnode
1364 */
1365static void
1366jrecord_write_vnode_ref(struct jrecord *jrec, struct vnode *vp)
1367{
1368 /* XXX */
1369}
1370
1371/*
1372 * Write out the data associated with a UIO
1373 */
1374static void
1375jrecord_write_uio(struct jrecord *jrec, int16_t rectype, struct uio *uio)
1376{
1377 /* XXX */
2281065e
MD
1378}
1379
1380/************************************************************************
1381 * JOURNAL VNOPS *
558b8e00
MD
1382 ************************************************************************
1383 *
1384 * These are function shims replacing the normal filesystem ops. We become
1385 * responsible for calling the underlying filesystem ops. We have the choice
1386 * of executing the underlying op first and then generating the journal entry,
1387 * or starting the journal entry, executing the underlying op, and then
1388 * either completing or aborting it.
1389 *
1390 * The journal is supposed to be a high-level entity, which generally means
1391 * identifying files by name rather then by inode. Supplying both allows
1392 * the journal to be used both for inode-number-compatible 'mirrors' and
1393 * for simple filesystem replication.
1394 *
1395 * Writes are particularly difficult to deal with because a single write may
1396 * represent a hundred megabyte buffer or more, and both writes and truncations
1397 * require the 'old' data to be written out as well as the new data if the
1398 * log is reversable. Other issues:
1399 *
1400 * - How to deal with operations on unlinked files (no path available),
1401 * but which may still be filesystem visible due to hard links.
1402 *
1403 * - How to deal with modifications made via a memory map.
1404 *
1405 * - Future cache coherency support will require cache coherency API calls
1406 * both prior to and after the call to the underlying VFS.
1407 *
1408 * ALSO NOTE: We do not have to shim compatibility VOPs like MKDIR which have
1409 * new VFS equivalents (NMKDIR).
1410 */
1411
b2f7ec6c
MD
1412/*
1413 * Journal vop_settattr { a_vp, a_vap, a_cred, a_td }
1414 */
558b8e00
MD
1415static
1416int
1417journal_setattr(struct vop_setattr_args *ap)
1418{
1419 struct mount *mp;
1420 struct journal *jo;
1421 struct jrecord jrec;
1422 void *save; /* warning, save pointers do not always remain valid */
1423 int error;
1424
1425 error = vop_journal_operate_ap(&ap->a_head);
1426 mp = ap->a_head.a_ops->vv_mount;
1427 if (error == 0) {
1428 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
1429 jrecord_init(jo, &jrec, -1);
1430 save = jrecord_push(&jrec, JTYPE_SETATTR);
b2f7ec6c
MD
1431 jrecord_write_cred(&jrec, ap->a_td, ap->a_cred);
1432 jrecord_write_vnode_ref(&jrec, ap->a_vp);
1433 jrecord_write_vattr(&jrec, ap->a_vap);
558b8e00
MD
1434 jrecord_pop(&jrec, save);
1435 jrecord_done(&jrec, 0);
1436 }
1437 }
1438 return (error);
1439}
1440
b2f7ec6c
MD
1441/*
1442 * Journal vop_write { a_vp, a_uio, a_ioflag, a_cred }
1443 */
558b8e00
MD
1444static
1445int
1446journal_write(struct vop_write_args *ap)
1447{
1448 struct mount *mp;
1449 struct journal *jo;
1450 struct jrecord jrec;
1451 void *save; /* warning, save pointers do not always remain valid */
1452 int error;
1453
1454 error = vop_journal_operate_ap(&ap->a_head);
1455 mp = ap->a_head.a_ops->vv_mount;
1456 if (error == 0) {
1457 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
1458 jrecord_init(jo, &jrec, -1);
1459 save = jrecord_push(&jrec, JTYPE_WRITE);
b2f7ec6c
MD
1460 jrecord_write_cred(&jrec, NULL, ap->a_cred);
1461 jrecord_write_vnode_ref(&jrec, ap->a_vp);
1462 jrecord_write_uio(&jrec, JLEAF_FILEDATA, ap->a_uio);
558b8e00
MD
1463 jrecord_pop(&jrec, save);
1464 jrecord_done(&jrec, 0);
1465 }
1466 }
1467 return (error);
1468}
1469
b2f7ec6c
MD
1470/*
1471 * Journal vop_fsync { a_vp, a_waitfor, a_td }
1472 */
558b8e00
MD
1473static
1474int
1475journal_fsync(struct vop_fsync_args *ap)
1476{
1477 struct mount *mp;
1478 struct journal *jo;
1479 int error;
1480
1481 error = vop_journal_operate_ap(&ap->a_head);
1482 mp = ap->a_head.a_ops->vv_mount;
1483 if (error == 0) {
1484 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
1485 /* XXX synchronize pending journal records */
1486 }
1487 }
1488 return (error);
1489}
1490
b2f7ec6c
MD
1491/*
1492 * Journal vop_putpages { a_vp, a_m, a_count, a_sync, a_rtvals, a_offset }
1493 */
558b8e00
MD
1494static
1495int
1496journal_putpages(struct vop_putpages_args *ap)
1497{
1498 struct mount *mp;
1499 struct journal *jo;
1500 struct jrecord jrec;
1501 void *save; /* warning, save pointers do not always remain valid */
1502 int error;
1503
1504 error = vop_journal_operate_ap(&ap->a_head);
1505 mp = ap->a_head.a_ops->vv_mount;
1506 if (error == 0) {
1507 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
1508 jrecord_init(jo, &jrec, -1);
1509 save = jrecord_push(&jrec, JTYPE_PUTPAGES);
b2f7ec6c
MD
1510 jrecord_write_vnode_ref(&jrec, ap->a_vp);
1511 /* XXX pagelist */
558b8e00
MD
1512 jrecord_pop(&jrec, save);
1513 jrecord_done(&jrec, 0);
1514 }
1515 }
1516 return (error);
1517}
1518
b2f7ec6c
MD
1519/*
1520 * Journal vop_setacl { a_vp, a_type, a_aclp, a_cred, a_td }
1521 */
558b8e00
MD
1522static
1523int
1524journal_setacl(struct vop_setacl_args *ap)
1525{
1526 struct mount *mp;
1527 struct journal *jo;
1528 struct jrecord jrec;
1529 void *save; /* warning, save pointers do not always remain valid */
1530 int error;
1531
1532 error = vop_journal_operate_ap(&ap->a_head);
1533 mp = ap->a_head.a_ops->vv_mount;
1534 if (error == 0) {
1535 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
1536 jrecord_init(jo, &jrec, -1);
1537 save = jrecord_push(&jrec, JTYPE_SETACL);
b2f7ec6c
MD
1538 jrecord_write_cred(&jrec, ap->a_td, ap->a_cred);
1539 jrecord_write_vnode_ref(&jrec, ap->a_vp);
1540 /* XXX type, aclp */
558b8e00
MD
1541 jrecord_pop(&jrec, save);
1542 jrecord_done(&jrec, 0);
1543 }
1544 }
1545 return (error);
1546}
1547
b2f7ec6c
MD
1548/*
1549 * Journal vop_setextattr { a_vp, a_name, a_uio, a_cred, a_td }
1550 */
558b8e00
MD
1551static
1552int
1553journal_setextattr(struct vop_setextattr_args *ap)
1554{
1555 struct mount *mp;
1556 struct journal *jo;
1557 struct jrecord jrec;
1558 void *save; /* warning, save pointers do not always remain valid */
1559 int error;
1560
1561 error = vop_journal_operate_ap(&ap->a_head);
1562 mp = ap->a_head.a_ops->vv_mount;
1563 if (error == 0) {
1564 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
1565 jrecord_init(jo, &jrec, -1);
1566 save = jrecord_push(&jrec, JTYPE_SETEXTATTR);
b2f7ec6c
MD
1567 jrecord_write_cred(&jrec, ap->a_td, ap->a_cred);
1568 jrecord_write_vnode_ref(&jrec, ap->a_vp);
1569 jrecord_leaf(&jrec, JLEAF_ATTRNAME, ap->a_name, strlen(ap->a_name));
1570 jrecord_write_uio(&jrec, JLEAF_FILEDATA, ap->a_uio);
558b8e00
MD
1571 jrecord_pop(&jrec, save);
1572 jrecord_done(&jrec, 0);
1573 }
1574 }
1575 return (error);
1576}
1577
b2f7ec6c
MD
1578/*
1579 * Journal vop_ncreate { a_ncp, a_vpp, a_cred, a_vap }
1580 */
558b8e00
MD
1581static
1582int
1583journal_ncreate(struct vop_ncreate_args *ap)
1584{
1585 struct mount *mp;
1586 struct journal *jo;
1587 struct jrecord jrec;
1588 void *save; /* warning, save pointers do not always remain valid */
1589 int error;
1590
1591 error = vop_journal_operate_ap(&ap->a_head);
1592 mp = ap->a_head.a_ops->vv_mount;
1593 if (error == 0) {
1594 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
1595 jrecord_init(jo, &jrec, -1);
1596 save = jrecord_push(&jrec, JTYPE_CREATE);
b2f7ec6c
MD
1597 jrecord_write_cred(&jrec, NULL, ap->a_cred);
1598 jrecord_write_path(&jrec, JLEAF_PATH1, ap->a_ncp);
1599 if (*ap->a_vpp)
1600 jrecord_write_vnode_ref(&jrec, *ap->a_vpp);
558b8e00
MD
1601 jrecord_pop(&jrec, save);
1602 jrecord_done(&jrec, 0);
1603 }
1604 }
1605 return (error);
1606}
1607
b2f7ec6c
MD
1608/*
1609 * Journal vop_nmknod { a_ncp, a_vpp, a_cred, a_vap }
1610 */
558b8e00
MD
1611static
1612int
1613journal_nmknod(struct vop_nmknod_args *ap)
1614{
1615 struct mount *mp;
1616 struct journal *jo;
1617 struct jrecord jrec;
1618 void *save; /* warning, save pointers do not always remain valid */
1619 int error;
1620
1621 error = vop_journal_operate_ap(&ap->a_head);
1622 mp = ap->a_head.a_ops->vv_mount;
1623 if (error == 0) {
1624 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
1625 jrecord_init(jo, &jrec, -1);
1626 save = jrecord_push(&jrec, JTYPE_MKNOD);
b2f7ec6c
MD
1627 jrecord_write_cred(&jrec, NULL, ap->a_cred);
1628 jrecord_write_path(&jrec, JLEAF_PATH1, ap->a_ncp);
1629 jrecord_write_vattr(&jrec, ap->a_vap);
1630 if (*ap->a_vpp)
1631 jrecord_write_vnode_ref(&jrec, *ap->a_vpp);
558b8e00
MD
1632 jrecord_pop(&jrec, save);
1633 jrecord_done(&jrec, 0);
1634 }
1635 }
1636 return (error);
1637}
1638
b2f7ec6c
MD
1639/*
1640 * Journal vop_nlink { a_ncp, a_vp, a_cred }
1641 */
558b8e00
MD
1642static
1643int
1644journal_nlink(struct vop_nlink_args *ap)
1645{
1646 struct mount *mp;
1647 struct journal *jo;
1648 struct jrecord jrec;
1649 void *save; /* warning, save pointers do not always remain valid */
1650 int error;
1651
1652 error = vop_journal_operate_ap(&ap->a_head);
1653 mp = ap->a_head.a_ops->vv_mount;
1654 if (error == 0) {
1655 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
1656 jrecord_init(jo, &jrec, -1);
1657 save = jrecord_push(&jrec, JTYPE_LINK);
b2f7ec6c
MD
1658 jrecord_write_cred(&jrec, NULL, ap->a_cred);
1659 jrecord_write_path(&jrec, JLEAF_PATH1, ap->a_ncp);
1660 jrecord_write_vnode_ref(&jrec, ap->a_vp);
1661 /* XXX PATH to VP and inode number */
558b8e00
MD
1662 jrecord_pop(&jrec, save);
1663 jrecord_done(&jrec, 0);
1664 }
1665 }
1666 return (error);
1667}
1668
b2f7ec6c
MD
1669/*
1670 * Journal vop_symlink { a_ncp, a_vpp, a_cred, a_vap, a_target }
1671 */
558b8e00
MD
1672static
1673int
1674journal_nsymlink(struct vop_nsymlink_args *ap)
1675{
1676 struct mount *mp;
1677 struct journal *jo;
1678 struct jrecord jrec;
1679 void *save; /* warning, save pointers do not always remain valid */
1680 int error;
1681
1682 error = vop_journal_operate_ap(&ap->a_head);
1683 mp = ap->a_head.a_ops->vv_mount;
1684 if (error == 0) {
1685 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
1686 jrecord_init(jo, &jrec, -1);
1687 save = jrecord_push(&jrec, JTYPE_SYMLINK);
b2f7ec6c
MD
1688 jrecord_write_cred(&jrec, NULL, ap->a_cred);
1689 jrecord_write_path(&jrec, JLEAF_PATH1, ap->a_ncp);
1690 jrecord_leaf(&jrec, JLEAF_SYMLINKDATA,
1691 ap->a_target, strlen(ap->a_target));
1692 if (*ap->a_vpp)
1693 jrecord_write_vnode_ref(&jrec, *ap->a_vpp);
558b8e00
MD
1694 jrecord_pop(&jrec, save);
1695 jrecord_done(&jrec, 0);
1696 }
1697 }
1698 return (error);
1699}
1700
b2f7ec6c
MD
1701/*
1702 * Journal vop_nwhiteout { a_ncp, a_cred, a_flags }
1703 */
558b8e00
MD
1704static
1705int
1706journal_nwhiteout(struct vop_nwhiteout_args *ap)
1707{
1708 struct mount *mp;
1709 struct journal *jo;
1710 struct jrecord jrec;
1711 void *save; /* warning, save pointers do not always remain valid */
1712 int error;
1713
1714 error = vop_journal_operate_ap(&ap->a_head);
1715 mp = ap->a_head.a_ops->vv_mount;
1716 if (error == 0) {
1717 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
1718 jrecord_init(jo, &jrec, -1);
1719 save = jrecord_push(&jrec, JTYPE_WHITEOUT);
b2f7ec6c
MD
1720 jrecord_write_cred(&jrec, NULL, ap->a_cred);
1721 jrecord_write_path(&jrec, JLEAF_PATH1, ap->a_ncp);
558b8e00
MD
1722 jrecord_pop(&jrec, save);
1723 jrecord_done(&jrec, 0);
1724 }
1725 }
1726 return (error);
1727}
1728
b2f7ec6c
MD
1729/*
1730 * Journal vop_nremove { a_ncp, a_cred }
1731 */
558b8e00
MD
1732static
1733int
1734journal_nremove(struct vop_nremove_args *ap)
1735{
1736 struct mount *mp;
1737 struct journal *jo;
1738 struct jrecord jrec;
1739 void *save; /* warning, save pointers do not always remain valid */
1740 int error;
1741
1742 error = vop_journal_operate_ap(&ap->a_head);
1743 mp = ap->a_head.a_ops->vv_mount;
1744 if (error == 0) {
1745 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
1746 jrecord_init(jo, &jrec, -1);
1747 save = jrecord_push(&jrec, JTYPE_REMOVE);
b2f7ec6c
MD
1748 jrecord_write_cred(&jrec, NULL, ap->a_cred);
1749 jrecord_write_path(&jrec, JLEAF_PATH1, ap->a_ncp);
558b8e00
MD
1750 jrecord_pop(&jrec, save);
1751 jrecord_done(&jrec, 0);
1752 }
1753 }
1754 return (error);
1755}
2281065e 1756
b2f7ec6c
MD
1757/*
1758 * Journal vop_nmkdir { a_ncp, a_vpp, a_cred, a_vap }
1759 */
2281065e
MD
1760static
1761int
1762journal_nmkdir(struct vop_nmkdir_args *ap)
1763{
82eaef15
MD
1764 struct mount *mp;
1765 struct journal *jo;
1766 struct jrecord jrec;
1767 void *save; /* warning, save pointers do not always remain valid */
2281065e
MD
1768 int error;
1769
2281065e 1770 error = vop_journal_operate_ap(&ap->a_head);
82eaef15
MD
1771 mp = ap->a_head.a_ops->vv_mount;
1772 if (error == 0) {
1773 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
1774 jrecord_init(jo, &jrec, -1);
1775 if (jo->flags & MC_JOURNAL_WANT_REVERSABLE) {
1776 save = jrecord_push(&jrec, JTYPE_UNDO);
1777 /* XXX undo operations */
1778 jrecord_pop(&jrec, save);
1779 }
1780#if 0
1781 if (jo->flags & MC_JOURNAL_WANT_AUDIT) {
1782 jrecord_write_audit(&jrec);
1783 }
1784#endif
1785 save = jrecord_push(&jrec, JTYPE_MKDIR);
1786 jrecord_write_path(&jrec, JLEAF_PATH1, ap->a_ncp);
b2f7ec6c 1787 jrecord_write_cred(&jrec, NULL, ap->a_cred);
82eaef15 1788 jrecord_write_vattr(&jrec, ap->a_vap);
b2f7ec6c
MD
1789 jrecord_write_path(&jrec, JLEAF_PATH1, ap->a_ncp);
1790 if (*ap->a_vpp)
1791 jrecord_write_vnode_ref(&jrec, *ap->a_vpp);
82eaef15
MD
1792 jrecord_pop(&jrec, save);
1793 jrecord_done(&jrec, 0);
1794 }
1795 }
2281065e 1796 return (error);
6ddb7618
MD
1797}
1798
b2f7ec6c
MD
1799/*
1800 * Journal vop_nrmdir { a_ncp, a_cred }
1801 */
558b8e00
MD
1802static
1803int
1804journal_nrmdir(struct vop_nrmdir_args *ap)
1805{
1806 struct mount *mp;
1807 struct journal *jo;
1808 struct jrecord jrec;
1809 void *save; /* warning, save pointers do not always remain valid */
1810 int error;
1811
1812 error = vop_journal_operate_ap(&ap->a_head);
1813 mp = ap->a_head.a_ops->vv_mount;
1814 if (error == 0) {
1815 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
1816 jrecord_init(jo, &jrec, -1);
1817 save = jrecord_push(&jrec, JTYPE_RMDIR);
b2f7ec6c
MD
1818 jrecord_write_cred(&jrec, NULL, ap->a_cred);
1819 jrecord_write_path(&jrec, JLEAF_PATH1, ap->a_ncp);
558b8e00
MD
1820 jrecord_pop(&jrec, save);
1821 jrecord_done(&jrec, 0);
1822 }
1823 }
1824 return (error);
1825}
1826
b2f7ec6c
MD
1827/*
1828 * Journal vop_nrename { a_fncp, a_tncp, a_cred }
1829 */
558b8e00
MD
1830static
1831int
1832journal_nrename(struct vop_nrename_args *ap)
1833{
1834 struct mount *mp;
1835 struct journal *jo;
1836 struct jrecord jrec;
1837 void *save; /* warning, save pointers do not always remain valid */
1838 int error;
1839
1840 error = vop_journal_operate_ap(&ap->a_head);
1841 mp = ap->a_head.a_ops->vv_mount;
1842 if (error == 0) {
1843 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
1844 jrecord_init(jo, &jrec, -1);
1845 save = jrecord_push(&jrec, JTYPE_RENAME);
b2f7ec6c
MD
1846 jrecord_write_cred(&jrec, NULL, ap->a_cred);
1847 jrecord_write_path(&jrec, JLEAF_PATH1, ap->a_fncp);
1848 jrecord_write_path(&jrec, JLEAF_PATH2, ap->a_tncp);
558b8e00
MD
1849 jrecord_pop(&jrec, save);
1850 jrecord_done(&jrec, 0);
1851 }
1852 }
1853 return (error);
1854}
1855