Filesystem journaling. Reorganize the journal scan for the mountpoint to
[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 *
26e603ed 34 * $DragonFly: src/sys/kern/vfs_jops.c,v 1.19 2005/08/24 20:28:31 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>
9578bde0 90#include <sys/msfbuf.h>
500b6a22
MD
91#include <sys/socket.h>
92#include <sys/socketvar.h>
6ddb7618
MD
93
94#include <machine/limits.h>
95
96#include <vm/vm.h>
97#include <vm/vm_object.h>
98#include <vm/vm_page.h>
99#include <vm/vm_pager.h>
100#include <vm/vnode_pager.h>
101
2281065e
MD
102#include <sys/file2.h>
103#include <sys/thread2.h>
104
105static int journal_attach(struct mount *mp);
106static void journal_detach(struct mount *mp);
107static int journal_install_vfs_journal(struct mount *mp, struct file *fp,
108 const struct mountctl_install_journal *info);
500b6a22
MD
109static int journal_restart_vfs_journal(struct mount *mp, struct file *fp,
110 const struct mountctl_restart_journal *info);
2281065e
MD
111static int journal_remove_vfs_journal(struct mount *mp,
112 const struct mountctl_remove_journal *info);
500b6a22
MD
113static int journal_restart(struct mount *mp, struct file *fp,
114 struct journal *jo, int flags);
432b8263 115static int journal_destroy(struct mount *mp, struct journal *jo, int flags);
2281065e 116static int journal_resync_vfs_journal(struct mount *mp, const void *ctl);
39b13188
MD
117static int journal_status_vfs_journal(struct mount *mp,
118 const struct mountctl_status_journal *info,
119 struct mountctl_journal_ret_status *rstat,
120 int buflen, int *res);
500b6a22
MD
121static void journal_create_threads(struct journal *jo);
122static void journal_destroy_threads(struct journal *jo, int flags);
432b8263
MD
123static void journal_wthread(void *info);
124static void journal_rthread(void *info);
82eaef15
MD
125
126static void *journal_reserve(struct journal *jo,
127 struct journal_rawrecbeg **rawpp,
128 int16_t streamid, int bytes);
129static void *journal_extend(struct journal *jo,
130 struct journal_rawrecbeg **rawpp,
131 int truncbytes, int bytes, int *newstreamrecp);
132static void journal_abort(struct journal *jo,
133 struct journal_rawrecbeg **rawpp);
134static void journal_commit(struct journal *jo,
135 struct journal_rawrecbeg **rawpp,
136 int bytes, int closeout);
137
138static void jrecord_init(struct journal *jo,
139 struct jrecord *jrec, int16_t streamid);
140static struct journal_subrecord *jrecord_push(
141 struct jrecord *jrec, int16_t rectype);
142static void jrecord_pop(struct jrecord *jrec, struct journal_subrecord *parent);
143static struct journal_subrecord *jrecord_write(struct jrecord *jrec,
144 int16_t rectype, int bytes);
145static void jrecord_data(struct jrecord *jrec, const void *buf, int bytes);
146static void jrecord_done(struct jrecord *jrec, int abortit);
26e603ed
MD
147static void jrecord_undo_file(struct jrecord *jrec, struct vnode *vp,
148 int jrflags, off_t off, off_t bytes);
82eaef15 149
558b8e00
MD
150static int journal_setattr(struct vop_setattr_args *ap);
151static int journal_write(struct vop_write_args *ap);
152static int journal_fsync(struct vop_fsync_args *ap);
153static int journal_putpages(struct vop_putpages_args *ap);
154static int journal_setacl(struct vop_setacl_args *ap);
155static int journal_setextattr(struct vop_setextattr_args *ap);
156static int journal_ncreate(struct vop_ncreate_args *ap);
157static int journal_nmknod(struct vop_nmknod_args *ap);
158static int journal_nlink(struct vop_nlink_args *ap);
159static int journal_nsymlink(struct vop_nsymlink_args *ap);
160static int journal_nwhiteout(struct vop_nwhiteout_args *ap);
161static int journal_nremove(struct vop_nremove_args *ap);
2281065e 162static int journal_nmkdir(struct vop_nmkdir_args *ap);
558b8e00
MD
163static int journal_nrmdir(struct vop_nrmdir_args *ap);
164static int journal_nrename(struct vop_nrename_args *ap);
2281065e 165
26e603ed
MD
166#define JRUNDO_SIZE 0x00000001
167#define JRUNDO_UID 0x00000002
168#define JRUNDO_GID 0x00000004
169#define JRUNDO_FSID 0x00000008
170#define JRUNDO_MODES 0x00000010
171#define JRUNDO_INUM 0x00000020
172#define JRUNDO_ATIME 0x00000040
173#define JRUNDO_MTIME 0x00000080
174#define JRUNDO_CTIME 0x00000100
175#define JRUNDO_GEN 0x00000200
176#define JRUNDO_FLAGS 0x00000400
177#define JRUNDO_UDEV 0x00000800
178#define JRUNDO_FILEDATA 0x00010000
179#define JRUNDO_GETVP 0x00020000
180#define JRUNDO_CONDLINK 0x00040000 /* write file data if link count 1 */
181#define JRUNDO_VATTR (JRUNDO_SIZE|JRUNDO_UID|JRUNDO_GID|JRUNDO_FSID|\
182 JRUNDO_MODES|JRUNDO_INUM|JRUNDO_ATIME|JRUNDO_MTIME|\
183 JRUNDO_CTIME|JRUNDO_GEN|JRUNDO_FLAGS|JRUNDO_UDEV)
184#define JRUNDO_ALL (JRUNDO_VATTR|JRUNDO_FILEDATA)
185
6ddb7618 186static struct vnodeopv_entry_desc journal_vnodeop_entries[] = {
2281065e
MD
187 { &vop_default_desc, vop_journal_operate_ap },
188 { &vop_mountctl_desc, (void *)journal_mountctl },
558b8e00
MD
189 { &vop_setattr_desc, (void *)journal_setattr },
190 { &vop_write_desc, (void *)journal_write },
191 { &vop_fsync_desc, (void *)journal_fsync },
192 { &vop_putpages_desc, (void *)journal_putpages },
193 { &vop_setacl_desc, (void *)journal_setacl },
194 { &vop_setextattr_desc, (void *)journal_setextattr },
195 { &vop_ncreate_desc, (void *)journal_ncreate },
196 { &vop_nmknod_desc, (void *)journal_nmknod },
197 { &vop_nlink_desc, (void *)journal_nlink },
198 { &vop_nsymlink_desc, (void *)journal_nsymlink },
199 { &vop_nwhiteout_desc, (void *)journal_nwhiteout },
200 { &vop_nremove_desc, (void *)journal_nremove },
2281065e 201 { &vop_nmkdir_desc, (void *)journal_nmkdir },
558b8e00
MD
202 { &vop_nrmdir_desc, (void *)journal_nrmdir },
203 { &vop_nrename_desc, (void *)journal_nrename },
2281065e 204 { NULL, NULL }
6ddb7618
MD
205};
206
82eaef15 207static MALLOC_DEFINE(M_JOURNAL, "journal", "Journaling structures");
2281065e
MD
208static MALLOC_DEFINE(M_JFIFO, "journal-fifo", "Journal FIFO");
209
6ddb7618 210int
2281065e
MD
211journal_mountctl(struct vop_mountctl_args *ap)
212{
213 struct mount *mp;
214 int error = 0;
215
216 mp = ap->a_head.a_ops->vv_mount;
217 KKASSERT(mp);
218
219 if (mp->mnt_vn_journal_ops == NULL) {
220 switch(ap->a_op) {
221 case MOUNTCTL_INSTALL_VFS_JOURNAL:
222 error = journal_attach(mp);
223 if (error == 0 && ap->a_ctllen != sizeof(struct mountctl_install_journal))
224 error = EINVAL;
225 if (error == 0 && ap->a_fp == NULL)
226 error = EBADF;
227 if (error == 0)
228 error = journal_install_vfs_journal(mp, ap->a_fp, ap->a_ctl);
229 if (TAILQ_EMPTY(&mp->mnt_jlist))
230 journal_detach(mp);
231 break;
500b6a22 232 case MOUNTCTL_RESTART_VFS_JOURNAL:
2281065e
MD
233 case MOUNTCTL_REMOVE_VFS_JOURNAL:
234 case MOUNTCTL_RESYNC_VFS_JOURNAL:
39b13188
MD
235 case MOUNTCTL_STATUS_VFS_JOURNAL:
236 error = ENOENT;
2281065e
MD
237 break;
238 default:
239 error = EOPNOTSUPP;
240 break;
241 }
242 } else {
243 switch(ap->a_op) {
244 case MOUNTCTL_INSTALL_VFS_JOURNAL:
245 if (ap->a_ctllen != sizeof(struct mountctl_install_journal))
246 error = EINVAL;
247 if (error == 0 && ap->a_fp == NULL)
248 error = EBADF;
249 if (error == 0)
250 error = journal_install_vfs_journal(mp, ap->a_fp, ap->a_ctl);
251 break;
500b6a22
MD
252 case MOUNTCTL_RESTART_VFS_JOURNAL:
253 if (ap->a_ctllen != sizeof(struct mountctl_restart_journal))
254 error = EINVAL;
255 if (error == 0 && ap->a_fp == NULL)
256 error = EBADF;
257 if (error == 0)
258 error = journal_restart_vfs_journal(mp, ap->a_fp, ap->a_ctl);
259 break;
2281065e
MD
260 case MOUNTCTL_REMOVE_VFS_JOURNAL:
261 if (ap->a_ctllen != sizeof(struct mountctl_remove_journal))
262 error = EINVAL;
263 if (error == 0)
264 error = journal_remove_vfs_journal(mp, ap->a_ctl);
265 if (TAILQ_EMPTY(&mp->mnt_jlist))
266 journal_detach(mp);
267 break;
268 case MOUNTCTL_RESYNC_VFS_JOURNAL:
269 if (ap->a_ctllen != 0)
270 error = EINVAL;
271 error = journal_resync_vfs_journal(mp, ap->a_ctl);
272 break;
39b13188
MD
273 case MOUNTCTL_STATUS_VFS_JOURNAL:
274 if (ap->a_ctllen != sizeof(struct mountctl_status_journal))
275 error = EINVAL;
276 if (error == 0) {
277 error = journal_status_vfs_journal(mp, ap->a_ctl,
278 ap->a_buf, ap->a_buflen, ap->a_res);
279 }
280 break;
2281065e
MD
281 default:
282 error = EOPNOTSUPP;
283 break;
284 }
285 }
286 return (error);
287}
288
289/*
290 * High level mount point setup. When a
291 */
292static int
6ddb7618
MD
293journal_attach(struct mount *mp)
294{
2281065e
MD
295 vfs_add_vnodeops(mp, &mp->mnt_vn_journal_ops, journal_vnodeop_entries);
296 return(0);
6ddb7618
MD
297}
298
2281065e 299static void
6ddb7618
MD
300journal_detach(struct mount *mp)
301{
2281065e
MD
302 if (mp->mnt_vn_journal_ops)
303 vfs_rm_vnodeops(&mp->mnt_vn_journal_ops);
304}
305
306/*
82eaef15
MD
307 * Install a journal on a mount point. Each journal has an associated worker
308 * thread which is responsible for buffering and spooling the data to the
309 * target. A mount point may have multiple journals attached to it. An
310 * initial start record is generated when the journal is associated.
2281065e
MD
311 */
312static int
313journal_install_vfs_journal(struct mount *mp, struct file *fp,
314 const struct mountctl_install_journal *info)
315{
316 struct journal *jo;
82eaef15 317 struct jrecord jrec;
2281065e
MD
318 int error = 0;
319 int size;
320
321 jo = malloc(sizeof(struct journal), M_JOURNAL, M_WAITOK|M_ZERO);
322 bcopy(info->id, jo->id, sizeof(jo->id));
432b8263
MD
323 jo->flags = info->flags & ~(MC_JOURNAL_WACTIVE | MC_JOURNAL_RACTIVE |
324 MC_JOURNAL_STOP_REQ);
2281065e
MD
325
326 /*
327 * Memory FIFO size, round to nearest power of 2
328 */
82eaef15 329 if (info->membufsize) {
2281065e
MD
330 if (info->membufsize < 65536)
331 size = 65536;
332 else if (info->membufsize > 128 * 1024 * 1024)
333 size = 128 * 1024 * 1024;
334 else
335 size = (int)info->membufsize;
336 } else {
337 size = 1024 * 1024;
338 }
339 jo->fifo.size = 1;
340 while (jo->fifo.size < size)
341 jo->fifo.size <<= 1;
342
343 /*
344 * Other parameters. If not specified the starting transaction id
345 * will be the current date.
346 */
82eaef15 347 if (info->transid) {
2281065e
MD
348 jo->transid = info->transid;
349 } else {
350 struct timespec ts;
351 getnanotime(&ts);
352 jo->transid = ((int64_t)ts.tv_sec << 30) | ts.tv_nsec;
353 }
354
355 jo->fp = fp;
356
357 /*
358 * Allocate the memory FIFO
359 */
360 jo->fifo.mask = jo->fifo.size - 1;
361 jo->fifo.membase = malloc(jo->fifo.size, M_JFIFO, M_WAITOK|M_ZERO|M_NULLOK);
362 if (jo->fifo.membase == NULL)
363 error = ENOMEM;
364
82eaef15 365 /*
3119bac5 366 * Create the worker threads and generate the association record.
82eaef15 367 */
2281065e
MD
368 if (error) {
369 free(jo, M_JOURNAL);
370 } else {
371 fhold(fp);
500b6a22 372 journal_create_threads(jo);
82eaef15
MD
373 jrecord_init(jo, &jrec, JREC_STREAMID_DISCONT);
374 jrecord_write(&jrec, JTYPE_ASSOCIATE, 0);
375 jrecord_done(&jrec, 0);
2281065e
MD
376 TAILQ_INSERT_TAIL(&mp->mnt_jlist, jo, jentry);
377 }
378 return(error);
379}
380
500b6a22
MD
381/*
382 * Restart a journal with a new descriptor. The existing reader and writer
383 * threads are terminated and a new descriptor is associated with the
384 * journal. The FIFO rindex is reset to xindex and the threads are then
385 * restarted.
386 */
387static int
388journal_restart_vfs_journal(struct mount *mp, struct file *fp,
389 const struct mountctl_restart_journal *info)
390{
391 struct journal *jo;
392 int error;
393
394 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
395 if (bcmp(jo->id, info->id, sizeof(jo->id)) == 0)
396 break;
397 }
398 if (jo)
399 error = journal_restart(mp, fp, jo, info->flags);
400 else
401 error = EINVAL;
402 return (error);
403}
404
405static int
406journal_restart(struct mount *mp, struct file *fp,
407 struct journal *jo, int flags)
408{
409 /*
410 * XXX lock the jo
411 */
412
413#if 0
414 /*
415 * Record the fact that we are doing a restart in the journal.
416 * XXX it isn't safe to do this if the journal is being restarted
417 * because it was locked up and the writer thread has already exited.
418 */
419 jrecord_init(jo, &jrec, JREC_STREAMID_RESTART);
420 jrecord_write(&jrec, JTYPE_DISASSOCIATE, 0);
421 jrecord_done(&jrec, 0);
422#endif
423
424 /*
425 * Stop the reader and writer threads and clean up the current
426 * descriptor.
427 */
428 printf("RESTART WITH FP %p KILLING %p\n", fp, jo->fp);
429 journal_destroy_threads(jo, flags);
430
431 if (jo->fp)
432 fdrop(jo->fp, curthread);
433
434 /*
435 * Associate the new descriptor, reset the FIFO index, and recreate
436 * the threads.
437 */
438 fhold(fp);
439 jo->fp = fp;
440 jo->fifo.rindex = jo->fifo.xindex;
441 journal_create_threads(jo);
442
443 return(0);
444}
445
82eaef15
MD
446/*
447 * Disassociate a journal from a mount point and terminate its worker thread.
448 * A final termination record is written out before the file pointer is
449 * dropped.
450 */
2281065e 451static int
82eaef15
MD
452journal_remove_vfs_journal(struct mount *mp,
453 const struct mountctl_remove_journal *info)
2281065e
MD
454{
455 struct journal *jo;
456 int error;
457
458 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
459 if (bcmp(jo->id, info->id, sizeof(jo->id)) == 0)
460 break;
461 }
432b8263
MD
462 if (jo)
463 error = journal_destroy(mp, jo, info->flags);
464 else
465 error = EINVAL;
466 return (error);
467}
82eaef15 468
432b8263
MD
469/*
470 * Remove all journals associated with a mount point. Usually called
471 * by the umount code.
472 */
473void
474journal_remove_all_journals(struct mount *mp, int flags)
475{
476 struct journal *jo;
82eaef15 477
432b8263
MD
478 while ((jo = TAILQ_FIRST(&mp->mnt_jlist)) != NULL) {
479 journal_destroy(mp, jo, flags);
2281065e 480 }
432b8263
MD
481}
482
483static int
484journal_destroy(struct mount *mp, struct journal *jo, int flags)
485{
486 struct jrecord jrec;
487
488 TAILQ_REMOVE(&mp->mnt_jlist, jo, jentry);
489
490 jrecord_init(jo, &jrec, JREC_STREAMID_DISCONT);
491 jrecord_write(&jrec, JTYPE_DISASSOCIATE, 0);
492 jrecord_done(&jrec, 0);
493
500b6a22
MD
494 journal_destroy_threads(jo, flags);
495
432b8263
MD
496 if (jo->fp)
497 fdrop(jo->fp, curthread);
498 if (jo->fifo.membase)
499 free(jo->fifo.membase, M_JFIFO);
500 free(jo, M_JOURNAL);
501 return(0);
2281065e
MD
502}
503
504static int
505journal_resync_vfs_journal(struct mount *mp, const void *ctl)
506{
507 return(EINVAL);
508}
509
39b13188
MD
510static int
511journal_status_vfs_journal(struct mount *mp,
512 const struct mountctl_status_journal *info,
513 struct mountctl_journal_ret_status *rstat,
514 int buflen, int *res)
515{
516 struct journal *jo;
517 int error = 0;
518 int index;
519
520 index = 0;
521 *res = 0;
522 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
523 if (info->index == MC_JOURNAL_INDEX_ID) {
524 if (bcmp(jo->id, info->id, sizeof(jo->id)) != 0)
525 continue;
526 } else if (info->index >= 0) {
527 if (info->index < index)
528 continue;
529 } else if (info->index != MC_JOURNAL_INDEX_ALL) {
530 continue;
531 }
532 if (buflen < sizeof(*rstat)) {
533 if (*res)
534 rstat[-1].flags |= MC_JOURNAL_STATUS_MORETOCOME;
535 else
536 error = EINVAL;
537 break;
538 }
539 bzero(rstat, sizeof(*rstat));
540 rstat->recsize = sizeof(*rstat);
541 bcopy(jo->id, rstat->id, sizeof(jo->id));
542 rstat->index = index;
543 rstat->membufsize = jo->fifo.size;
3119bac5
MD
544 rstat->membufused = jo->fifo.windex - jo->fifo.xindex;
545 rstat->membufunacked = jo->fifo.rindex - jo->fifo.xindex;
39b13188 546 rstat->bytessent = jo->total_acked;
3119bac5 547 rstat->fifostalls = jo->fifostalls;
39b13188
MD
548 ++rstat;
549 ++index;
550 *res += sizeof(*rstat);
551 buflen -= sizeof(*rstat);
552 }
553 return(error);
554}
432b8263 555
500b6a22
MD
556static void
557journal_create_threads(struct journal *jo)
558{
559 jo->flags &= ~(MC_JOURNAL_STOP_REQ | MC_JOURNAL_STOP_IMM);
560 jo->flags |= MC_JOURNAL_WACTIVE;
561 lwkt_create(journal_wthread, jo, NULL, &jo->wthread,
562 TDF_STOPREQ, -1, "journal w:%.*s", JIDMAX, jo->id);
563 lwkt_setpri(&jo->wthread, TDPRI_KERN_DAEMON);
564 lwkt_schedule(&jo->wthread);
565
566 if (jo->flags & MC_JOURNAL_WANT_FULLDUPLEX) {
567 jo->flags |= MC_JOURNAL_RACTIVE;
568 lwkt_create(journal_rthread, jo, NULL, &jo->rthread,
569 TDF_STOPREQ, -1, "journal r:%.*s", JIDMAX, jo->id);
570 lwkt_setpri(&jo->rthread, TDPRI_KERN_DAEMON);
571 lwkt_schedule(&jo->rthread);
572 }
573}
574
575static void
576journal_destroy_threads(struct journal *jo, int flags)
577{
578 int wcount;
579
580 jo->flags |= MC_JOURNAL_STOP_REQ | (flags & MC_JOURNAL_STOP_IMM);
581 wakeup(&jo->fifo);
582 wcount = 0;
583 while (jo->flags & (MC_JOURNAL_WACTIVE | MC_JOURNAL_RACTIVE)) {
584 tsleep(jo, 0, "jwait", hz);
585 if (++wcount % 10 == 0) {
586 printf("Warning: journal %s waiting for descriptors to close\n",
587 jo->id);
588 }
589 }
590
591 /*
592 * XXX SMP - threads should move to cpu requesting the restart or
593 * termination before finishing up to properly interlock.
594 */
595 tsleep(jo, 0, "jwait", hz);
596 lwkt_free_thread(&jo->wthread);
597 if (jo->flags & MC_JOURNAL_WANT_FULLDUPLEX)
598 lwkt_free_thread(&jo->rthread);
599}
600
82eaef15
MD
601/*
602 * The per-journal worker thread is responsible for writing out the
603 * journal's FIFO to the target stream.
604 */
2281065e 605static void
432b8263 606journal_wthread(void *info)
2281065e
MD
607{
608 struct journal *jo = info;
82eaef15 609 struct journal_rawrecbeg *rawp;
2281065e
MD
610 int bytes;
611 int error;
82eaef15 612 int avail;
2281065e
MD
613 int res;
614
615 for (;;) {
82eaef15
MD
616 /*
617 * Calculate the number of bytes available to write. This buffer
618 * area may contain reserved records so we can't just write it out
619 * without further checks.
620 */
621 bytes = jo->fifo.windex - jo->fifo.rindex;
622
623 /*
624 * sleep if no bytes are available or if an incomplete record is
625 * encountered (it needs to be filled in before we can write it
626 * out), and skip any pad records that we encounter.
627 */
628 if (bytes == 0) {
2281065e
MD
629 if (jo->flags & MC_JOURNAL_STOP_REQ)
630 break;
82eaef15
MD
631 tsleep(&jo->fifo, 0, "jfifo", hz);
632 continue;
633 }
9578bde0
MD
634
635 /*
636 * Sleep if we can not go any further due to hitting an incomplete
637 * record. This case should occur rarely but may have to be better
638 * optimized XXX.
639 */
82eaef15
MD
640 rawp = (void *)(jo->fifo.membase + (jo->fifo.rindex & jo->fifo.mask));
641 if (rawp->begmagic == JREC_INCOMPLETEMAGIC) {
642 tsleep(&jo->fifo, 0, "jpad", hz);
643 continue;
644 }
9578bde0
MD
645
646 /*
647 * Skip any pad records. We do not write out pad records if we can
648 * help it.
9578bde0 649 */
82eaef15 650 if (rawp->streamid == JREC_STREAMID_PAD) {
3119bac5
MD
651 if ((jo->flags & MC_JOURNAL_WANT_FULLDUPLEX) == 0) {
652 if (jo->fifo.rindex == jo->fifo.xindex) {
653 jo->fifo.xindex += (rawp->recsize + 15) & ~15;
654 jo->total_acked += (rawp->recsize + 15) & ~15;
655 }
656 }
82eaef15 657 jo->fifo.rindex += (rawp->recsize + 15) & ~15;
9578bde0
MD
658 jo->total_acked += bytes;
659 KKASSERT(jo->fifo.windex - jo->fifo.rindex >= 0);
82eaef15
MD
660 continue;
661 }
662
663 /*
9578bde0
MD
664 * 'bytes' is the amount of data that can potentially be written out.
665 * Calculate 'res', the amount of data that can actually be written
666 * out. res is bounded either by hitting the end of the physical
667 * memory buffer or by hitting an incomplete record. Incomplete
668 * records often occur due to the way the space reservation model
669 * works.
82eaef15
MD
670 */
671 res = 0;
672 avail = jo->fifo.size - (jo->fifo.rindex & jo->fifo.mask);
673 while (res < bytes && rawp->begmagic == JREC_BEGMAGIC) {
674 res += (rawp->recsize + 15) & ~15;
675 if (res >= avail) {
676 KKASSERT(res == avail);
677 break;
678 }
9578bde0 679 rawp = (void *)((char *)rawp + ((rawp->recsize + 15) & ~15));
2281065e 680 }
82eaef15
MD
681
682 /*
683 * Issue the write and deal with any errors or other conditions.
684 * For now assume blocking I/O. Since we are record-aware the
685 * code cannot yet handle partial writes.
686 *
3119bac5
MD
687 * We bump rindex prior to issuing the write to avoid racing
688 * the acknowledgement coming back (which could prevent the ack
689 * from bumping xindex). Restarts are always based on xindex so
690 * we do not try to undo the rindex if an error occurs.
691 *
82eaef15
MD
692 * XXX EWOULDBLOCK/NBIO
693 * XXX notification on failure
9578bde0 694 * XXX permanent verses temporary failures
82eaef15
MD
695 * XXX two-way acknowledgement stream in the return direction / xindex
696 */
697 bytes = res;
3119bac5 698 jo->fifo.rindex += bytes;
82eaef15 699 error = fp_write(jo->fp,
3119bac5 700 jo->fifo.membase + ((jo->fifo.rindex - bytes) & jo->fifo.mask),
82eaef15 701 bytes, &res);
2281065e
MD
702 if (error) {
703 printf("journal_thread(%s) write, error %d\n", jo->id, error);
82eaef15 704 /* XXX */
2281065e 705 } else {
82eaef15 706 KKASSERT(res == bytes);
82eaef15
MD
707 }
708
709 /*
432b8263
MD
710 * Advance rindex. If the journal stream is not full duplex we also
711 * advance xindex, otherwise the rjournal thread is responsible for
712 * advancing xindex.
82eaef15 713 */
3119bac5 714 if ((jo->flags & MC_JOURNAL_WANT_FULLDUPLEX) == 0) {
432b8263 715 jo->fifo.xindex += bytes;
3119bac5
MD
716 jo->total_acked += bytes;
717 }
9578bde0 718 KKASSERT(jo->fifo.windex - jo->fifo.rindex >= 0);
432b8263
MD
719 if ((jo->flags & MC_JOURNAL_WANT_FULLDUPLEX) == 0) {
720 if (jo->flags & MC_JOURNAL_WWAIT) {
721 jo->flags &= ~MC_JOURNAL_WWAIT; /* XXX hysteresis */
722 wakeup(&jo->fifo.windex);
723 }
724 }
725 }
500b6a22 726 fp_shutdown(jo->fp, SHUT_WR);
432b8263
MD
727 jo->flags &= ~MC_JOURNAL_WACTIVE;
728 wakeup(jo);
729 wakeup(&jo->fifo.windex);
730}
731
732/*
733 * A second per-journal worker thread is created for two-way journaling
734 * streams to deal with the return acknowledgement stream.
735 */
736static void
737journal_rthread(void *info)
738{
739 struct journal_rawrecbeg *rawp;
740 struct journal_ackrecord ack;
741 struct journal *jo = info;
742 int64_t transid;
743 int error;
744 int count;
745 int bytes;
432b8263
MD
746
747 transid = 0;
748 error = 0;
749
750 for (;;) {
751 /*
752 * We have been asked to stop
753 */
754 if (jo->flags & MC_JOURNAL_STOP_REQ)
755 break;
756
757 /*
758 * If we have no active transaction id, get one from the return
759 * stream.
760 */
761 if (transid == 0) {
3119bac5
MD
762 error = fp_read(jo->fp, &ack, sizeof(ack), &count, 1);
763#if 0
764 printf("fp_read ack error %d count %d\n", error, count);
765#endif
766 if (error || count != sizeof(ack))
767 break;
432b8263
MD
768 if (error) {
769 printf("read error %d on receive stream\n", error);
770 break;
771 }
772 if (ack.rbeg.begmagic != JREC_BEGMAGIC ||
773 ack.rend.endmagic != JREC_ENDMAGIC
774 ) {
775 printf("bad begmagic or endmagic on receive stream\n");
776 break;
777 }
778 transid = ack.rbeg.transid;
2281065e 779 }
432b8263
MD
780
781 /*
782 * Calculate the number of unacknowledged bytes. If there are no
783 * unacknowledged bytes then unsent data was acknowledged, report,
784 * sleep a bit, and loop in that case. This should not happen
785 * normally. The ack record is thrown away.
786 */
787 bytes = jo->fifo.rindex - jo->fifo.xindex;
788
789 if (bytes == 0) {
3119bac5 790 printf("warning: unsent data acknowledged transid %08llx\n", transid);
432b8263
MD
791 tsleep(&jo->fifo.xindex, 0, "jrseq", hz);
792 transid = 0;
793 continue;
794 }
795
796 /*
3119bac5 797 * Since rindex has advanced, the record pointed to by xindex
432b8263
MD
798 * must be a valid record.
799 */
800 rawp = (void *)(jo->fifo.membase + (jo->fifo.xindex & jo->fifo.mask));
801 KKASSERT(rawp->begmagic == JREC_BEGMAGIC);
802 KKASSERT(rawp->recsize <= bytes);
803
804 /*
805 * The target can acknowledge several records at once.
806 */
807 if (rawp->transid < transid) {
3119bac5 808#if 1
432b8263 809 printf("ackskip %08llx/%08llx\n", rawp->transid, transid);
3119bac5 810#endif
432b8263 811 jo->fifo.xindex += (rawp->recsize + 15) & ~15;
3119bac5 812 jo->total_acked += (rawp->recsize + 15) & ~15;
432b8263
MD
813 if (jo->flags & MC_JOURNAL_WWAIT) {
814 jo->flags &= ~MC_JOURNAL_WWAIT; /* XXX hysteresis */
815 wakeup(&jo->fifo.windex);
816 }
817 continue;
818 }
819 if (rawp->transid == transid) {
3119bac5 820#if 1
432b8263 821 printf("ackskip %08llx/%08llx\n", rawp->transid, transid);
3119bac5 822#endif
432b8263 823 jo->fifo.xindex += (rawp->recsize + 15) & ~15;
3119bac5 824 jo->total_acked += (rawp->recsize + 15) & ~15;
432b8263
MD
825 if (jo->flags & MC_JOURNAL_WWAIT) {
826 jo->flags &= ~MC_JOURNAL_WWAIT; /* XXX hysteresis */
827 wakeup(&jo->fifo.windex);
828 }
829 transid = 0;
830 continue;
831 }
3119bac5 832 printf("warning: unsent data(2) acknowledged transid %08llx\n", transid);
432b8263 833 transid = 0;
2281065e 834 }
432b8263 835 jo->flags &= ~MC_JOURNAL_RACTIVE;
2281065e
MD
836 wakeup(jo);
837 wakeup(&jo->fifo.windex);
838}
839
9578bde0
MD
840/*
841 * This builds a pad record which the journaling thread will skip over. Pad
842 * records are required when we are unable to reserve sufficient stream space
843 * due to insufficient space at the end of the physical memory fifo.
432b8263
MD
844 *
845 * Even though the record is not transmitted, a normal transid must be
846 * assigned to it so link recovery operations after a failure work properly.
9578bde0 847 */
88c28735 848static
82eaef15 849void
432b8263 850journal_build_pad(struct journal_rawrecbeg *rawp, int recsize, int64_t transid)
2281065e 851{
82eaef15
MD
852 struct journal_rawrecend *rendp;
853
854 KKASSERT((recsize & 15) == 0 && recsize >= 16);
855
82eaef15
MD
856 rawp->streamid = JREC_STREAMID_PAD;
857 rawp->recsize = recsize; /* must be 16-byte aligned */
432b8263 858 rawp->transid = transid;
82eaef15
MD
859 /*
860 * WARNING, rendp may overlap rawp->seqno. This is necessary to
35238fa5 861 * allow PAD records to fit in 16 bytes. Use cpu_ccfence() to
82eaef15
MD
862 * hopefully cause the compiler to not make any assumptions.
863 */
82eaef15
MD
864 rendp = (void *)((char *)rawp + rawp->recsize - sizeof(*rendp));
865 rendp->endmagic = JREC_ENDMAGIC;
866 rendp->check = 0;
867 rendp->recsize = rawp->recsize;
9578bde0
MD
868
869 /*
870 * Set the begin magic last. This is what will allow the journal
35238fa5
MD
871 * thread to write the record out. Use a store fence to prevent
872 * compiler and cpu reordering of the writes.
9578bde0 873 */
35238fa5 874 cpu_sfence();
9578bde0 875 rawp->begmagic = JREC_BEGMAGIC;
2281065e
MD
876}
877
82eaef15
MD
878/*
879 * Wake up the worker thread if the FIFO is more then half full or if
880 * someone is waiting for space to be freed up. Otherwise let the
881 * heartbeat deal with it. Being able to avoid waking up the worker
9578bde0 882 * is the key to the journal's cpu performance.
82eaef15
MD
883 */
884static __inline
2281065e 885void
82eaef15 886journal_commit_wakeup(struct journal *jo)
2281065e
MD
887{
888 int avail;
889
82eaef15
MD
890 avail = jo->fifo.size - (jo->fifo.windex - jo->fifo.xindex);
891 KKASSERT(avail >= 0);
892 if ((avail < (jo->fifo.size >> 1)) || (jo->flags & MC_JOURNAL_WWAIT))
893 wakeup(&jo->fifo);
894}
895
896/*
897 * Create a new BEGIN stream record with the specified streamid and the
898 * specified amount of payload space. *rawpp will be set to point to the
899 * base of the new stream record and a pointer to the base of the payload
900 * space will be returned. *rawpp does not need to be pre-NULLd prior to
432b8263 901 * making this call. The raw record header will be partially initialized.
82eaef15
MD
902 *
903 * A stream can be extended, aborted, or committed by other API calls
904 * below. This may result in a sequence of potentially disconnected
905 * stream records to be output to the journaling target. The first record
906 * (the one created by this function) will be marked JREC_STREAMCTL_BEGIN,
907 * while the last record on commit or abort will be marked JREC_STREAMCTL_END
908 * (and possibly also JREC_STREAMCTL_ABORTED). The last record could wind
909 * up being the same as the first, in which case the bits are all set in
910 * the first record.
911 *
912 * The stream record is created in an incomplete state by setting the begin
913 * magic to JREC_INCOMPLETEMAGIC. This prevents the worker thread from
914 * flushing the fifo past our record until we have finished populating it.
915 * Other threads can reserve and operate on their own space without stalling
916 * but the stream output will stall until we have completed operations. The
917 * memory FIFO is intended to be large enough to absorb such situations
918 * without stalling out other threads.
919 */
920static
921void *
922journal_reserve(struct journal *jo, struct journal_rawrecbeg **rawpp,
923 int16_t streamid, int bytes)
924{
925 struct journal_rawrecbeg *rawp;
926 int avail;
927 int availtoend;
928 int req;
929
930 /*
931 * Add header and trailer overheads to the passed payload. Note that
932 * the passed payload size need not be aligned in any way.
933 */
934 bytes += sizeof(struct journal_rawrecbeg);
935 bytes += sizeof(struct journal_rawrecend);
936
937 for (;;) {
938 /*
939 * First, check boundary conditions. If the request would wrap around
940 * we have to skip past the ending block and return to the beginning
941 * of the FIFO's buffer. Calculate 'req' which is the actual number
942 * of bytes being reserved, including wrap-around dead space.
943 *
88c28735
MD
944 * Neither 'bytes' or 'req' are aligned.
945 *
82eaef15
MD
946 * Note that availtoend is not truncated to avail and so cannot be
947 * used to determine whether the reservation is possible by itself.
948 * Also, since all fifo ops are 16-byte aligned, we can check
949 * the size before calculating the aligned size.
950 */
951 availtoend = jo->fifo.size - (jo->fifo.windex & jo->fifo.mask);
88c28735 952 KKASSERT((availtoend & 15) == 0);
82eaef15
MD
953 if (bytes > availtoend)
954 req = bytes + availtoend; /* add pad to end */
955 else
956 req = bytes;
957
958 /*
959 * Next calculate the total available space and see if it is
960 * sufficient. We cannot overwrite previously buffered data
961 * past xindex because otherwise we would not be able to restart
962 * a broken link at the target's last point of commit.
963 */
964 avail = jo->fifo.size - (jo->fifo.windex - jo->fifo.xindex);
965 KKASSERT(avail >= 0 && (avail & 15) == 0);
966
967 if (avail < req) {
968 /* XXX MC_JOURNAL_STOP_IMM */
2281065e 969 jo->flags |= MC_JOURNAL_WWAIT;
3119bac5 970 ++jo->fifostalls;
2281065e
MD
971 tsleep(&jo->fifo.windex, 0, "jwrite", 0);
972 continue;
973 }
82eaef15
MD
974
975 /*
976 * Create a pad record for any dead space and create an incomplete
977 * record for the live space, then return a pointer to the
978 * contiguous buffer space that was requested.
979 *
980 * NOTE: The worker thread will not flush past an incomplete
981 * record, so the reserved space can be filled in at-will. The
982 * journaling code must also be aware the reserved sections occuring
983 * after this one will also not be written out even if completed
984 * until this one is completed.
432b8263
MD
985 *
986 * The transaction id must accomodate real and potential pad creation.
82eaef15
MD
987 */
988 rawp = (void *)(jo->fifo.membase + (jo->fifo.windex & jo->fifo.mask));
989 if (req != bytes) {
432b8263
MD
990 journal_build_pad(rawp, availtoend, jo->transid);
991 ++jo->transid;
82eaef15
MD
992 rawp = (void *)jo->fifo.membase;
993 }
994 rawp->begmagic = JREC_INCOMPLETEMAGIC; /* updated by abort/commit */
995 rawp->recsize = bytes; /* (unaligned size) */
996 rawp->streamid = streamid | JREC_STREAMCTL_BEGIN;
432b8263
MD
997 rawp->transid = jo->transid;
998 jo->transid += 2;
82eaef15
MD
999
1000 /*
1001 * Issue a memory barrier to guarentee that the record data has been
1002 * properly initialized before we advance the write index and return
1003 * a pointer to the reserved record. Otherwise the worker thread
1004 * could accidently run past us.
1005 *
1006 * Note that stream records are always 16-byte aligned.
1007 */
35238fa5 1008 cpu_sfence();
82eaef15
MD
1009 jo->fifo.windex += (req + 15) & ~15;
1010 *rawpp = rawp;
1011 return(rawp + 1);
1012 }
1013 /* not reached */
1014 *rawpp = NULL;
1015 return(NULL);
1016}
1017
1018/*
143c4f15
MD
1019 * Attempt to extend the stream record by <bytes> worth of payload space.
1020 *
1021 * If it is possible to extend the existing stream record no truncation
1022 * occurs and the record is extended as specified. A pointer to the
1023 * truncation offset within the payload space is returned.
82eaef15 1024 *
143c4f15
MD
1025 * If it is not possible to do this the existing stream record is truncated
1026 * and committed, and a new stream record of size <bytes> is created. A
1027 * pointer to the base of the new stream record's payload space is returned.
82eaef15 1028 *
143c4f15
MD
1029 * *rawpp is set to the new reservation in the case of a new record but
1030 * the caller cannot depend on a comparison with the old rawp to determine if
1031 * this case occurs because we could end up using the same memory FIFO
1032 * offset for the new stream record. Use *newstreamrecp instead.
82eaef15
MD
1033 */
1034static void *
1035journal_extend(struct journal *jo, struct journal_rawrecbeg **rawpp,
1036 int truncbytes, int bytes, int *newstreamrecp)
1037{
1038 struct journal_rawrecbeg *rawp;
1039 int16_t streamid;
1040 int availtoend;
1041 int avail;
1042 int osize;
1043 int nsize;
1044 int wbase;
1045 void *rptr;
1046
1047 *newstreamrecp = 0;
1048 rawp = *rawpp;
1049 osize = (rawp->recsize + 15) & ~15;
1050 nsize = (rawp->recsize + bytes + 15) & ~15;
1051 wbase = (char *)rawp - jo->fifo.membase;
1052
1053 /*
143c4f15
MD
1054 * If the aligned record size does not change we can trivially adjust
1055 * the record size.
82eaef15
MD
1056 */
1057 if (nsize == osize) {
1058 rawp->recsize += bytes;
143c4f15 1059 return((char *)(rawp + 1) + truncbytes);
82eaef15
MD
1060 }
1061
1062 /*
1063 * If the fifo's write index hasn't been modified since we made the
1064 * reservation and we do not hit any boundary conditions, we can
143c4f15 1065 * trivially make the record smaller or larger.
82eaef15
MD
1066 */
1067 if ((jo->fifo.windex & jo->fifo.mask) == wbase + osize) {
1068 availtoend = jo->fifo.size - wbase;
1069 avail = jo->fifo.size - (jo->fifo.windex - jo->fifo.xindex) + osize;
1070 KKASSERT((availtoend & 15) == 0);
1071 KKASSERT((avail & 15) == 0);
1072 if (nsize <= avail && nsize <= availtoend) {
1073 jo->fifo.windex += nsize - osize;
1074 rawp->recsize += bytes;
143c4f15 1075 return((char *)(rawp + 1) + truncbytes);
82eaef15
MD
1076 }
1077 }
1078
1079 /*
1080 * It was not possible to extend the buffer. Commit the current
1081 * buffer and create a new one. We manually clear the BEGIN mark that
1082 * journal_reserve() creates (because this is a continuing record, not
1083 * the start of a new stream).
1084 */
1085 streamid = rawp->streamid & JREC_STREAMID_MASK;
1086 journal_commit(jo, rawpp, truncbytes, 0);
1087 rptr = journal_reserve(jo, rawpp, streamid, bytes);
1088 rawp = *rawpp;
1089 rawp->streamid &= ~JREC_STREAMCTL_BEGIN;
1090 *newstreamrecp = 1;
1091 return(rptr);
1092}
1093
1094/*
1095 * Abort a journal record. If the transaction record represents a stream
1096 * BEGIN and we can reverse the fifo's write index we can simply reverse
1097 * index the entire record, as if it were never reserved in the first place.
1098 *
1099 * Otherwise we set the JREC_STREAMCTL_ABORTED bit and commit the record
1100 * with the payload truncated to 0 bytes.
1101 */
1102static void
1103journal_abort(struct journal *jo, struct journal_rawrecbeg **rawpp)
1104{
1105 struct journal_rawrecbeg *rawp;
1106 int osize;
1107
1108 rawp = *rawpp;
1109 osize = (rawp->recsize + 15) & ~15;
1110
1111 if ((rawp->streamid & JREC_STREAMCTL_BEGIN) &&
1112 (jo->fifo.windex & jo->fifo.mask) ==
1113 (char *)rawp - jo->fifo.membase + osize)
1114 {
1115 jo->fifo.windex -= osize;
1116 *rawpp = NULL;
1117 } else {
1118 rawp->streamid |= JREC_STREAMCTL_ABORTED;
1119 journal_commit(jo, rawpp, 0, 1);
1120 }
1121}
1122
1123/*
1124 * Commit a journal record and potentially truncate it to the specified
1125 * number of payload bytes. If you do not want to truncate the record,
1126 * simply pass -1 for the bytes parameter. Do not pass rawp->recsize, that
1127 * field includes header and trailer and will not be correct. Note that
1128 * passing 0 will truncate the entire data payload of the record.
1129 *
1130 * The logical stream is terminated by this function.
1131 *
1132 * If truncation occurs, and it is not possible to physically optimize the
1133 * memory FIFO due to other threads having reserved space after ours,
1134 * the remaining reserved space will be covered by a pad record.
1135 */
1136static void
1137journal_commit(struct journal *jo, struct journal_rawrecbeg **rawpp,
1138 int bytes, int closeout)
1139{
1140 struct journal_rawrecbeg *rawp;
1141 struct journal_rawrecend *rendp;
1142 int osize;
1143 int nsize;
1144
1145 rawp = *rawpp;
1146 *rawpp = NULL;
1147
1148 KKASSERT((char *)rawp >= jo->fifo.membase &&
1149 (char *)rawp + rawp->recsize <= jo->fifo.membase + jo->fifo.size);
1150 KKASSERT(((intptr_t)rawp & 15) == 0);
1151
1152 /*
88c28735 1153 * Truncate the record if necessary. If the FIFO write index as still
82eaef15 1154 * at the end of our record we can optimally backindex it. Otherwise
88c28735 1155 * we have to insert a pad record to cover the dead space.
82eaef15
MD
1156 *
1157 * We calculate osize which is the 16-byte-aligned original recsize.
1158 * We calculate nsize which is the 16-byte-aligned new recsize.
1159 *
1160 * Due to alignment issues or in case the passed truncation bytes is
88c28735
MD
1161 * the same as the original payload, nsize may be equal to osize even
1162 * if the committed bytes is less then the originally reserved bytes.
82eaef15
MD
1163 */
1164 if (bytes >= 0) {
1165 KKASSERT(bytes >= 0 && bytes <= rawp->recsize - sizeof(struct journal_rawrecbeg) - sizeof(struct journal_rawrecend));
1166 osize = (rawp->recsize + 15) & ~15;
1167 rawp->recsize = bytes + sizeof(struct journal_rawrecbeg) +
1168 sizeof(struct journal_rawrecend);
1169 nsize = (rawp->recsize + 15) & ~15;
88c28735 1170 KKASSERT(nsize <= osize);
82eaef15
MD
1171 if (osize == nsize) {
1172 /* do nothing */
1173 } else if ((jo->fifo.windex & jo->fifo.mask) == (char *)rawp - jo->fifo.membase + osize) {
1174 /* we are able to backindex the fifo */
1175 jo->fifo.windex -= osize - nsize;
1176 } else {
1177 /* we cannot backindex the fifo, emplace a pad in the dead space */
432b8263
MD
1178 journal_build_pad((void *)((char *)rawp + nsize), osize - nsize,
1179 rawp->transid + 1);
82eaef15 1180 }
2281065e 1181 }
82eaef15
MD
1182
1183 /*
1184 * Fill in the trailer. Note that unlike pad records, the trailer will
1185 * never overlap the header.
1186 */
1187 rendp = (void *)((char *)rawp +
1188 ((rawp->recsize + 15) & ~15) - sizeof(*rendp));
1189 rendp->endmagic = JREC_ENDMAGIC;
1190 rendp->recsize = rawp->recsize;
1191 rendp->check = 0; /* XXX check word, disabled for now */
1192
1193 /*
1194 * Fill in begmagic last. This will allow the worker thread to proceed.
1195 * Use a memory barrier to guarentee write ordering. Mark the stream
1196 * as terminated if closeout is set. This is the typical case.
1197 */
1198 if (closeout)
1199 rawp->streamid |= JREC_STREAMCTL_END;
35238fa5 1200 cpu_sfence(); /* memory and compiler barrier */
82eaef15
MD
1201 rawp->begmagic = JREC_BEGMAGIC;
1202
1203 journal_commit_wakeup(jo);
1204}
1205
26e603ed
MD
1206/************************************************************************
1207 * PARALLEL TRANSACTION SUPPORT ROUTINES *
1208 ************************************************************************
1209 *
1210 * JRECLIST_*() - routines which create and iterate over jrecord structures,
1211 * because a mount point may have multiple attached journals.
1212 */
1213
1214/*
1215 * Initialize the passed jrecord_list and create a jrecord for each
1216 * journal we need to write to. Unnecessary mallocs are avoided by
1217 * using the passed jrecord structure as the first jrecord in the list.
1218 * A starting transaction is pushed for each jrecord.
1219 *
1220 * Returns non-zero if any of the journals require undo records.
1221 */
1222static
1223int
1224jreclist_init(struct mount *mp, struct jrecord_list *jreclist,
1225 struct jrecord *jreccache, int16_t rectype)
1226{
1227 struct journal *jo;
1228 struct jrecord *jrec;
1229 int wantrev = 0;
1230 int count = 0;
1231
1232 TAILQ_INIT(jreclist);
1233 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
1234 if (count == 0)
1235 jrec = jreccache;
1236 else
1237 jrec = malloc(sizeof(*jrec), M_JOURNAL, M_WAITOK);
1238 jrecord_init(jo, jrec, -1);
1239 jrec->user_save = jrecord_push(jrec, rectype);
1240 TAILQ_INSERT_TAIL(jreclist, jrec, user_entry);
1241 if (jo->flags & MC_JOURNAL_WANT_REVERSABLE)
1242 wantrev = 1;
1243 ++count;
1244 }
1245 return(wantrev);
1246}
1247
1248/*
1249 * Terminate the journaled transactions started by jreclist_init(). If
1250 * an error occured, the transaction records will be aborted.
1251 */
1252static
1253void
1254jreclist_done(struct jrecord_list *jreclist, int error)
1255{
1256 struct jrecord *jrec;
1257 int count;
1258
1259 TAILQ_FOREACH(jrec, jreclist, user_entry) {
1260 jrecord_pop(jrec, jrec->user_save);
1261 jrecord_done(jrec, error);
1262 }
1263 count = 0;
1264 while ((jrec = TAILQ_FIRST(jreclist)) != NULL) {
1265 TAILQ_REMOVE(jreclist, jrec, user_entry);
1266 if (count)
1267 free(jrec, M_JOURNAL);
1268 ++count;
1269 }
1270}
1271
1272/*
1273 * This procedure writes out UNDO records for available reversable
1274 * journals.
1275 *
1276 * XXX could use improvement. There is no need to re-read the file
1277 * for each journal.
1278 */
1279static
1280void
1281jreclist_undo_file(struct jrecord_list *jreclist, struct vnode *vp,
1282 int jrflags, off_t off, off_t bytes)
1283{
1284 struct jrecord *jrec;
1285 int error;
1286
1287 error = 0;
1288 if (jrflags & JRUNDO_GETVP)
1289 error = vget(vp, LK_SHARED, curthread);
1290 if (error == 0) {
1291 TAILQ_FOREACH(jrec, jreclist, user_entry) {
1292 if (jrec->jo->flags & MC_JOURNAL_WANT_REVERSABLE) {
1293 jrecord_undo_file(jrec, vp, jrflags, off, bytes);
1294 }
1295 }
1296 }
1297 if (error == 0 && jrflags & JRUNDO_GETVP)
1298 vput(vp);
1299}
1300
82eaef15
MD
1301/************************************************************************
1302 * TRANSACTION SUPPORT ROUTINES *
1303 ************************************************************************
1304 *
1305 * JRECORD_*() - routines to create subrecord transactions and embed them
1306 * in the logical streams managed by the journal_*() routines.
1307 */
1308
1309static int16_t sid = JREC_STREAMID_JMIN;
1310
1311/*
1312 * Initialize the passed jrecord structure and start a new stream transaction
1313 * by reserving an initial build space in the journal's memory FIFO.
1314 */
1315static void
1316jrecord_init(struct journal *jo, struct jrecord *jrec, int16_t streamid)
1317{
1318 bzero(jrec, sizeof(*jrec));
1319 jrec->jo = jo;
1320 if (streamid < 0) {
1321 streamid = sid++; /* XXX need to track stream ids! */
1322 if (sid == JREC_STREAMID_JMAX)
1323 sid = JREC_STREAMID_JMIN;
1324 }
1325 jrec->streamid = streamid;
1326 jrec->stream_residual = JREC_DEFAULTSIZE;
1327 jrec->stream_reserved = jrec->stream_residual;
1328 jrec->stream_ptr =
1329 journal_reserve(jo, &jrec->rawp, streamid, jrec->stream_reserved);
1330}
1331
1332/*
1333 * Push a recursive record type. All pushes should have matching pops.
1334 * The old parent is returned and the newly pushed record becomes the
1335 * new parent. Note that the old parent's pointer may already be invalid
1336 * or may become invalid if jrecord_write() had to build a new stream
1337 * record, so the caller should not mess with the returned pointer in
1338 * any way other then to save it.
1339 */
1340static
1341struct journal_subrecord *
1342jrecord_push(struct jrecord *jrec, int16_t rectype)
1343{
1344 struct journal_subrecord *save;
1345
1346 save = jrec->parent;
1347 jrec->parent = jrecord_write(jrec, rectype|JMASK_NESTED, 0);
1348 jrec->last = NULL;
1349 KKASSERT(jrec->parent != NULL);
1350 ++jrec->pushcount;
1351 ++jrec->pushptrgood; /* cleared on flush */
1352 return(save);
1353}
1354
1355/*
1356 * Pop a previously pushed sub-transaction. We must set JMASK_LAST
1357 * on the last record written within the subtransaction. If the last
1358 * record written is not accessible or if the subtransaction is empty,
1359 * we must write out a pad record with JMASK_LAST set before popping.
1360 *
1361 * When popping a subtransaction the parent record's recsize field
1362 * will be properly set. If the parent pointer is no longer valid
1363 * (which can occur if the data has already been flushed out to the
1364 * stream), the protocol spec allows us to leave it 0.
1365 *
1366 * The saved parent pointer which we restore may or may not be valid,
1367 * and if not valid may or may not be NULL, depending on the value
1368 * of pushptrgood.
1369 */
1370static void
1371jrecord_pop(struct jrecord *jrec, struct journal_subrecord *save)
1372{
1373 struct journal_subrecord *last;
1374
1375 KKASSERT(jrec->pushcount > 0);
1376 KKASSERT(jrec->residual == 0);
1377
1378 /*
1379 * Set JMASK_LAST on the last record we wrote at the current
1380 * level. If last is NULL we either no longer have access to the
1381 * record or the subtransaction was empty and we must write out a pad
1382 * record.
1383 */
1384 if ((last = jrec->last) == NULL) {
1385 jrecord_write(jrec, JLEAF_PAD|JMASK_LAST, 0);
1386 last = jrec->last; /* reload after possible flush */
1387 } else {
1388 last->rectype |= JMASK_LAST;
1389 }
1390
1391 /*
1392 * pushptrgood tells us how many levels of parent record pointers
1393 * are valid. The jrec only stores the current parent record pointer
1394 * (and it is only valid if pushptrgood != 0). The higher level parent
1395 * record pointers are saved by the routines calling jrecord_push() and
1396 * jrecord_pop(). These pointers may become stale and we determine
1397 * that fact by tracking the count of valid parent pointers with
1398 * pushptrgood. Pointers become invalid when their related stream
1399 * record gets pushed out.
1400 *
b2f7ec6c
MD
1401 * If no pointer is available (the data has already been pushed out),
1402 * then no fixup of e.g. the length field is possible for non-leaf
1403 * nodes. The protocol allows for this situation by placing a larger
1404 * burden on the program scanning the stream on the other end.
1405 *
82eaef15
MD
1406 * [parentA]
1407 * [node X]
1408 * [parentB]
1409 * [node Y]
1410 * [node Z]
1411 * (pop B) see NOTE B
1412 * (pop A) see NOTE A
1413 *
1414 * NOTE B: This pop sets LAST in node Z if the node is still accessible,
1415 * else a PAD record is appended and LAST is set in that.
1416 *
1417 * This pop sets the record size in parentB if parentB is still
1418 * accessible, else the record size is left 0 (the scanner must
1419 * deal with that).
1420 *
1421 * This pop sets the new 'last' record to parentB, the pointer
1422 * to which may or may not still be accessible.
1423 *
1424 * NOTE A: This pop sets LAST in parentB if the node is still accessible,
1425 * else a PAD record is appended and LAST is set in that.
1426 *
1427 * This pop sets the record size in parentA if parentA is still
1428 * accessible, else the record size is left 0 (the scanner must
1429 * deal with that).
1430 *
1431 * This pop sets the new 'last' record to parentA, the pointer
1432 * to which may or may not still be accessible.
1433 *
1434 * Also note that the last record in the stream transaction, which in
1435 * the above example is parentA, does not currently have the LAST bit
1436 * set.
1437 *
1438 * The current parent becomes the last record relative to the
1439 * saved parent passed into us. It's validity is based on
1440 * whether pushptrgood is non-zero prior to decrementing. The saved
1441 * parent becomes the new parent, and its validity is based on whether
1442 * pushptrgood is non-zero after decrementing.
1443 *
1444 * The old jrec->parent may be NULL if it is no longer accessible.
1445 * If pushptrgood is non-zero, however, it is guarenteed to not
1446 * be NULL (since no flush occured).
1447 */
1448 jrec->last = jrec->parent;
1449 --jrec->pushcount;
1450 if (jrec->pushptrgood) {
1451 KKASSERT(jrec->last != NULL && last != NULL);
1452 if (--jrec->pushptrgood == 0) {
1453 jrec->parent = NULL; /* 'save' contains garbage or NULL */
1454 } else {
1455 KKASSERT(save != NULL);
1456 jrec->parent = save; /* 'save' must not be NULL */
1457 }
1458
1459 /*
1460 * Set the record size in the old parent. 'last' still points to
1461 * the original last record in the subtransaction being popped,
1462 * jrec->last points to the old parent (which became the last
1463 * record relative to the new parent being popped into).
1464 */
1465 jrec->last->recsize = (char *)last + last->recsize - (char *)jrec->last;
1466 } else {
1467 jrec->parent = NULL;
1468 KKASSERT(jrec->last == NULL);
1469 }
1470}
1471
b2f7ec6c
MD
1472/*
1473 * Write out a leaf record, including associated data.
1474 */
1475static
1476void
1477jrecord_leaf(struct jrecord *jrec, int16_t rectype, void *ptr, int bytes)
1478{
1479 jrecord_write(jrec, rectype, bytes);
1480 jrecord_data(jrec, ptr, bytes);
b2f7ec6c
MD
1481}
1482
82eaef15
MD
1483/*
1484 * Write a leaf record out and return a pointer to its base. The leaf
1485 * record may contain potentially megabytes of data which is supplied
1486 * in jrecord_data() calls. The exact amount must be specified in this
1487 * call.
b2f7ec6c
MD
1488 *
1489 * THE RETURNED SUBRECORD POINTER IS ONLY VALID IMMEDIATELY AFTER THE
1490 * CALL AND MAY BECOME INVALID AT ANY TIME. ONLY THE PUSH/POP CODE SHOULD
1491 * USE THE RETURN VALUE.
82eaef15
MD
1492 */
1493static
1494struct journal_subrecord *
1495jrecord_write(struct jrecord *jrec, int16_t rectype, int bytes)
1496{
1497 struct journal_subrecord *last;
1498 int pusheditout;
1499
1500 /*
1501 * Try to catch some obvious errors. Nesting records must specify a
1502 * size of 0, and there should be no left-overs from previous operations
1503 * (such as incomplete data writeouts).
1504 */
1505 KKASSERT(bytes == 0 || (rectype & JMASK_NESTED) == 0);
1506 KKASSERT(jrec->residual == 0);
1507
1508 /*
1509 * Check to see if the current stream record has enough room for
1510 * the new subrecord header. If it doesn't we extend the current
1511 * stream record.
1512 *
1513 * This may have the side effect of pushing out the current stream record
1514 * and creating a new one. We must adjust our stream tracking fields
1515 * accordingly.
1516 */
1517 if (jrec->stream_residual < sizeof(struct journal_subrecord)) {
1518 jrec->stream_ptr = journal_extend(jrec->jo, &jrec->rawp,
1519 jrec->stream_reserved - jrec->stream_residual,
1520 JREC_DEFAULTSIZE, &pusheditout);
1521 if (pusheditout) {
143c4f15
MD
1522 /*
1523 * If a pushout occured, the pushed out stream record was
1524 * truncated as specified and the new record is exactly the
1525 * extension size specified.
1526 */
82eaef15
MD
1527 jrec->stream_reserved = JREC_DEFAULTSIZE;
1528 jrec->stream_residual = JREC_DEFAULTSIZE;
1529 jrec->parent = NULL; /* no longer accessible */
1530 jrec->pushptrgood = 0; /* restored parents in pops no good */
1531 } else {
143c4f15
MD
1532 /*
1533 * If no pushout occured the stream record is NOT truncated and
1534 * IS extended.
1535 */
82eaef15
MD
1536 jrec->stream_reserved += JREC_DEFAULTSIZE;
1537 jrec->stream_residual += JREC_DEFAULTSIZE;
1538 }
1539 }
1540 last = (void *)jrec->stream_ptr;
1541 last->rectype = rectype;
1542 last->reserved = 0;
b7ef558f
MD
1543
1544 /*
1545 * We may not know the record size for recursive records and the
1546 * header may become unavailable due to limited FIFO space. Write
1547 * -1 to indicate this special case.
1548 */
1549 if ((rectype & JMASK_NESTED) && bytes == 0)
1550 last->recsize = -1;
1551 else
1552 last->recsize = sizeof(struct journal_subrecord) + bytes;
82eaef15
MD
1553 jrec->last = last;
1554 jrec->residual = bytes; /* remaining data to be posted */
1555 jrec->residual_align = -bytes & 7; /* post-data alignment required */
143c4f15
MD
1556 jrec->stream_ptr += sizeof(*last); /* current write pointer */
1557 jrec->stream_residual -= sizeof(*last); /* space remaining in stream */
82eaef15
MD
1558 return(last);
1559}
1560
1561/*
1562 * Write out the data associated with a leaf record. Any number of calls
1563 * to this routine may be made as long as the byte count adds up to the
1564 * amount originally specified in jrecord_write().
1565 *
1566 * The act of writing out the leaf data may result in numerous stream records
1567 * being pushed out. Callers should be aware that even the associated
1568 * subrecord header may become inaccessible due to stream record pushouts.
1569 */
1570static void
1571jrecord_data(struct jrecord *jrec, const void *buf, int bytes)
1572{
1573 int pusheditout;
1574 int extsize;
1575
1576 KKASSERT(bytes >= 0 && bytes <= jrec->residual);
1577
1578 /*
1579 * Push out stream records as long as there is insufficient room to hold
1580 * the remaining data.
1581 */
1582 while (jrec->stream_residual < bytes) {
1583 /*
1584 * Fill in any remaining space in the current stream record.
1585 */
1586 bcopy(buf, jrec->stream_ptr, jrec->stream_residual);
1587 buf = (const char *)buf + jrec->stream_residual;
1588 bytes -= jrec->stream_residual;
1589 /*jrec->stream_ptr += jrec->stream_residual;*/
82eaef15 1590 jrec->residual -= jrec->stream_residual;
9578bde0 1591 jrec->stream_residual = 0;
82eaef15
MD
1592
1593 /*
1594 * Try to extend the current stream record, but no more then 1/4
1595 * the size of the FIFO.
1596 */
1597 extsize = jrec->jo->fifo.size >> 2;
1598 if (extsize > bytes)
1599 extsize = (bytes + 15) & ~15;
1600
1601 jrec->stream_ptr = journal_extend(jrec->jo, &jrec->rawp,
1602 jrec->stream_reserved - jrec->stream_residual,
1603 extsize, &pusheditout);
1604 if (pusheditout) {
1605 jrec->stream_reserved = extsize;
1606 jrec->stream_residual = extsize;
1607 jrec->parent = NULL; /* no longer accessible */
1608 jrec->last = NULL; /* no longer accessible */
1609 jrec->pushptrgood = 0; /* restored parents in pops no good */
1610 } else {
1611 jrec->stream_reserved += extsize;
1612 jrec->stream_residual += extsize;
1613 }
1614 }
1615
1616 /*
1617 * Push out any remaining bytes into the current stream record.
1618 */
1619 if (bytes) {
1620 bcopy(buf, jrec->stream_ptr, bytes);
1621 jrec->stream_ptr += bytes;
1622 jrec->stream_residual -= bytes;
1623 jrec->residual -= bytes;
1624 }
1625
1626 /*
1627 * Handle data alignment requirements for the subrecord. Because the
1628 * stream record's data space is more strictly aligned, it must already
1629 * have sufficient space to hold any subrecord alignment slop.
1630 */
1631 if (jrec->residual == 0 && jrec->residual_align) {
1632 KKASSERT(jrec->residual_align <= jrec->stream_residual);
1633 bzero(jrec->stream_ptr, jrec->residual_align);
1634 jrec->stream_ptr += jrec->residual_align;
1635 jrec->stream_residual -= jrec->residual_align;
1636 jrec->residual_align = 0;
1637 }
1638}
1639
1640/*
9578bde0
MD
1641 * We are finished with the transaction. This closes the transaction created
1642 * by jrecord_init().
1643 *
1644 * NOTE: If abortit is not set then we must be at the top level with no
1645 * residual subrecord data left to output.
1646 *
1647 * If abortit is set then we can be in any state, all pushes will be
1648 * popped and it is ok for there to be residual data. This works
1649 * because the virtual stream itself is truncated. Scanners must deal
1650 * with this situation.
82eaef15
MD
1651 *
1652 * The stream record will be committed or aborted as specified and jrecord
1653 * resources will be cleaned up.
1654 */
1655static void
1656jrecord_done(struct jrecord *jrec, int abortit)
1657{
1658 KKASSERT(jrec->rawp != NULL);
1659
1660 if (abortit) {
1661 journal_abort(jrec->jo, &jrec->rawp);
1662 } else {
1663 KKASSERT(jrec->pushcount == 0 && jrec->residual == 0);
1664 journal_commit(jrec->jo, &jrec->rawp,
1665 jrec->stream_reserved - jrec->stream_residual, 1);
1666 }
1667
1668 /*
1669 * jrec should not be used beyond this point without another init,
1670 * but clean up some fields to ensure that we panic if it is.
1671 *
1672 * Note that jrec->rawp is NULLd out by journal_abort/journal_commit.
1673 */
1674 jrec->jo = NULL;
1675 jrec->stream_ptr = NULL;
1676}
1677
1678/************************************************************************
b2f7ec6c 1679 * LOW LEVEL RECORD SUPPORT ROUTINES *
82eaef15
MD
1680 ************************************************************************
1681 *
b2f7ec6c
MD
1682 * These routine create low level recursive and leaf subrecords representing
1683 * common filesystem structures.
82eaef15
MD
1684 */
1685
b2f7ec6c
MD
1686/*
1687 * Write out a filename path relative to the base of the mount point.
1688 * rectype is typically JLEAF_PATH{1,2,3,4}.
1689 */
82eaef15
MD
1690static void
1691jrecord_write_path(struct jrecord *jrec, int16_t rectype, struct namecache *ncp)
1692{
b2f7ec6c
MD
1693 char buf[64]; /* local buffer if it fits, else malloced */
1694 char *base;
1695 int pathlen;
1696 int index;
1697 struct namecache *scan;
1698
1699 /*
1700 * Pass 1 - figure out the number of bytes required. Include terminating
1701 * \0 on last element and '/' separator on other elements.
1702 */
1703again:
1704 pathlen = 0;
1705 for (scan = ncp;
1706 scan && (scan->nc_flag & NCF_MOUNTPT) == 0;
1707 scan = scan->nc_parent
1708 ) {
1709 pathlen += scan->nc_nlen + 1;
1710 }
1711
1712 if (pathlen <= sizeof(buf))
1713 base = buf;
1714 else
1715 base = malloc(pathlen, M_TEMP, M_INTWAIT);
1716
1717 /*
1718 * Pass 2 - generate the path buffer
1719 */
1720 index = pathlen;
1721 for (scan = ncp;
1722 scan && (scan->nc_flag & NCF_MOUNTPT) == 0;
1723 scan = scan->nc_parent
1724 ) {
1725 if (scan->nc_nlen >= index) {
1726 if (base != buf)
1727 free(base, M_TEMP);
1728 goto again;
1729 }
1730 if (index == pathlen)
1731 base[--index] = 0;
1732 else
1733 base[--index] = '/';
1734 index -= scan->nc_nlen;
1735 bcopy(scan->nc_name, base + index, scan->nc_nlen);
1736 }
1737 jrecord_leaf(jrec, rectype, base + index, pathlen - index);
1738 if (base != buf)
1739 free(base, M_TEMP);
82eaef15
MD
1740}
1741
b2f7ec6c
MD
1742/*
1743 * Write out a file attribute structure. While somewhat inefficient, using
1744 * a recursive data structure is the most portable and extensible way.
1745 */
82eaef15
MD
1746static void
1747jrecord_write_vattr(struct jrecord *jrec, struct vattr *vat)
1748{
b2f7ec6c
MD
1749 void *save;
1750
1751 save = jrecord_push(jrec, JTYPE_VATTR);
1752 if (vat->va_type != VNON)
432b8263 1753 jrecord_leaf(jrec, JLEAF_VTYPE, &vat->va_type, sizeof(vat->va_type));
d0887c34 1754 if (vat->va_mode != (mode_t)VNOVAL)
432b8263 1755 jrecord_leaf(jrec, JLEAF_MODES, &vat->va_mode, sizeof(vat->va_mode));
b2f7ec6c
MD
1756 if (vat->va_nlink != VNOVAL)
1757 jrecord_leaf(jrec, JLEAF_NLINK, &vat->va_nlink, sizeof(vat->va_nlink));
1758 if (vat->va_uid != VNOVAL)
1759 jrecord_leaf(jrec, JLEAF_UID, &vat->va_uid, sizeof(vat->va_uid));
1760 if (vat->va_gid != VNOVAL)
1761 jrecord_leaf(jrec, JLEAF_GID, &vat->va_gid, sizeof(vat->va_gid));
1762 if (vat->va_fsid != VNOVAL)
1763 jrecord_leaf(jrec, JLEAF_FSID, &vat->va_fsid, sizeof(vat->va_fsid));
1764 if (vat->va_fileid != VNOVAL)
1765 jrecord_leaf(jrec, JLEAF_INUM, &vat->va_fileid, sizeof(vat->va_fileid));
1766 if (vat->va_size != VNOVAL)
1767 jrecord_leaf(jrec, JLEAF_SIZE, &vat->va_size, sizeof(vat->va_size));
1768 if (vat->va_atime.tv_sec != VNOVAL)
1769 jrecord_leaf(jrec, JLEAF_ATIME, &vat->va_atime, sizeof(vat->va_atime));
1770 if (vat->va_mtime.tv_sec != VNOVAL)
1771 jrecord_leaf(jrec, JLEAF_MTIME, &vat->va_mtime, sizeof(vat->va_mtime));
1772 if (vat->va_ctime.tv_sec != VNOVAL)
1773 jrecord_leaf(jrec, JLEAF_CTIME, &vat->va_ctime, sizeof(vat->va_ctime));
1774 if (vat->va_gen != VNOVAL)
1775 jrecord_leaf(jrec, JLEAF_GEN, &vat->va_gen, sizeof(vat->va_gen));
1776 if (vat->va_flags != VNOVAL)
1777 jrecord_leaf(jrec, JLEAF_FLAGS, &vat->va_flags, sizeof(vat->va_flags));
1778 if (vat->va_rdev != VNOVAL)
1779 jrecord_leaf(jrec, JLEAF_UDEV, &vat->va_rdev, sizeof(vat->va_rdev));
1780#if 0
1781 if (vat->va_filerev != VNOVAL)
1782 jrecord_leaf(jrec, JLEAF_FILEREV, &vat->va_filerev, sizeof(vat->va_filerev));
1783#endif
1784 jrecord_pop(jrec, save);
b2f7ec6c
MD
1785}
1786
1787/*
1788 * Write out the creds used to issue a file operation. If a process is
1789 * available write out additional tracking information related to the
1790 * process.
1791 *
1792 * XXX additional tracking info
1793 * XXX tty line info
1794 */
1795static void
1796jrecord_write_cred(struct jrecord *jrec, struct thread *td, struct ucred *cred)
1797{
1798 void *save;
1799 struct proc *p;
1800
1801 save = jrecord_push(jrec, JTYPE_CRED);
1802 jrecord_leaf(jrec, JLEAF_UID, &cred->cr_uid, sizeof(cred->cr_uid));
1803 jrecord_leaf(jrec, JLEAF_GID, &cred->cr_gid, sizeof(cred->cr_gid));
1804 if (td && (p = td->td_proc) != NULL) {
1805 jrecord_leaf(jrec, JLEAF_PID, &p->p_pid, sizeof(p->p_pid));
1806 jrecord_leaf(jrec, JLEAF_COMM, p->p_comm, sizeof(p->p_comm));
1807 }
1808 jrecord_pop(jrec, save);
b2f7ec6c
MD
1809}
1810
1811/*
1812 * Write out information required to identify a vnode
143c4f15
MD
1813 *
1814 * XXX this needs work. We should write out the inode number as well,
1815 * and in fact avoid writing out the file path for seqential writes
1816 * occuring within e.g. a certain period of time.
b2f7ec6c
MD
1817 */
1818static void
1819jrecord_write_vnode_ref(struct jrecord *jrec, struct vnode *vp)
1820{
143c4f15
MD
1821 struct namecache *ncp;
1822
1823 TAILQ_FOREACH(ncp, &vp->v_namecache, nc_vnode) {
1824 if ((ncp->nc_flag & (NCF_UNRESOLVED|NCF_DESTROYED)) == 0)
1825 break;
1826 }
1827 if (ncp)
1828 jrecord_write_path(jrec, JLEAF_PATH_REF, ncp);
1829}
1830
f4659a6c
MD
1831static void
1832jrecord_write_vnode_link(struct jrecord *jrec, struct vnode *vp,
1833 struct namecache *notncp)
1834{
1835 struct namecache *ncp;
1836
1837 TAILQ_FOREACH(ncp, &vp->v_namecache, nc_vnode) {
1838 if (ncp == notncp)
1839 continue;
1840 if ((ncp->nc_flag & (NCF_UNRESOLVED|NCF_DESTROYED)) == 0)
1841 break;
1842 }
1843 if (ncp)
1844 jrecord_write_path(jrec, JLEAF_PATH_REF, ncp);
1845}
1846
143c4f15
MD
1847#if 0
1848/*
1849 * Write out the current contents of the file within the specified
1850 * range. This is typically called from within an UNDO section. A
1851 * locked vnode must be passed.
1852 */
1853static int
1854jrecord_write_filearea(struct jrecord *jrec, struct vnode *vp,
1855 off_t begoff, off_t endoff)
1856{
1857}
1858#endif
1859
1860/*
1861 * Write out the data represented by a pagelist
1862 */
1863static void
1864jrecord_write_pagelist(struct jrecord *jrec, int16_t rectype,
1865 struct vm_page **pglist, int *rtvals, int pgcount,
1866 off_t offset)
1867{
1868 struct msf_buf *msf;
1869 int error;
1870 int b;
1871 int i;
1872
1873 i = 0;
1874 while (i < pgcount) {
1875 /*
1876 * Find the next valid section. Skip any invalid elements
1877 */
1878 if (rtvals[i] != VM_PAGER_OK) {
1879 ++i;
1880 offset += PAGE_SIZE;
1881 continue;
1882 }
1883
1884 /*
1885 * Figure out how big the valid section is, capping I/O at what the
1886 * MSFBUF can represent.
1887 */
1888 b = i;
1889 while (i < pgcount && i - b != XIO_INTERNAL_PAGES &&
1890 rtvals[i] == VM_PAGER_OK
1891 ) {
1892 ++i;
1893 }
1894
1895 /*
1896 * And write it out.
1897 */
1898 if (i - b) {
1899 error = msf_map_pagelist(&msf, pglist + b, i - b, 0);
1900 if (error == 0) {
1901 printf("RECORD PUTPAGES %d\n", msf_buf_bytes(msf));
1902 jrecord_leaf(jrec, JLEAF_SEEKPOS, &offset, sizeof(offset));
1903 jrecord_leaf(jrec, rectype,
1904 msf_buf_kva(msf), msf_buf_bytes(msf));
1905 msf_buf_free(msf);
1906 } else {
1907 printf("jrecord_write_pagelist: mapping failure\n");
1908 }
1909 offset += (off_t)(i - b) << PAGE_SHIFT;
1910 }
1911 }
b2f7ec6c
MD
1912}
1913
1914/*
9578bde0 1915 * Write out the data represented by a UIO.
b2f7ec6c 1916 */
9578bde0
MD
1917struct jwuio_info {
1918 struct jrecord *jrec;
1919 int16_t rectype;
1920};
1921
1922static int jrecord_write_uio_callback(void *info, char *buf, int bytes);
1923
b2f7ec6c
MD
1924static void
1925jrecord_write_uio(struct jrecord *jrec, int16_t rectype, struct uio *uio)
1926{
9578bde0
MD
1927 struct jwuio_info info = { jrec, rectype };
1928 int error;
1929
143c4f15
MD
1930 if (uio->uio_segflg != UIO_NOCOPY) {
1931 jrecord_leaf(jrec, JLEAF_SEEKPOS, &uio->uio_offset,
1932 sizeof(uio->uio_offset));
1933 error = msf_uio_iterate(uio, jrecord_write_uio_callback, &info);
1934 if (error)
1935 printf("XXX warning uio iterate failed %d\n", error);
1936 }
9578bde0
MD
1937}
1938
1939static int
1940jrecord_write_uio_callback(void *info_arg, char *buf, int bytes)
1941{
1942 struct jwuio_info *info = info_arg;
1943
9578bde0
MD
1944 jrecord_leaf(info->jrec, info->rectype, buf, bytes);
1945 return(0);
2281065e
MD
1946}
1947
26e603ed
MD
1948static void
1949jrecord_file_data(struct jrecord *jrec, struct vnode *vp,
1950 off_t off, off_t bytes)
1951{
1952 const int bufsize = 8192;
1953 char *buf;
1954 int error;
1955 int n;
1956
1957 buf = malloc(bufsize, M_JOURNAL, M_WAITOK);
1958 jrecord_leaf(jrec, JLEAF_SEEKPOS, &off, sizeof(off));
1959 while (bytes) {
1960 n = (bytes > bufsize) ? bufsize : (int)bytes;
1961 error = vn_rdwr(UIO_READ, vp, buf, n, off, UIO_SYSSPACE, IO_NODELOCKED,
1962 proc0.p_ucred, NULL, curthread);
1963 if (error) {
1964 jrecord_leaf(jrec, JLEAF_ERROR, &error, sizeof(error));
1965 break;
1966 }
1967 jrecord_leaf(jrec, JLEAF_FILEDATA, buf, n);
1968 bytes -= n;
1969 off += n;
1970 }
1971 free(buf, M_JOURNAL);
1972}
1973
1974/************************************************************************
1975 * LOW LEVEL UNDO SUPPORT ROUTINE *
1976 ************************************************************************
1977 *
1978 * This function is used to support UNDO records. It will generate an
1979 * appropriate record with the requested portion of the file data. Note
1980 * that file data is only recorded if JRUNDO_FILEDATA is passed. If bytes
1981 * is -1, it will be set to the size of the file.
1982 */
1983static void
1984jrecord_undo_file(struct jrecord *jrec, struct vnode *vp, int jrflags,
1985 off_t off, off_t bytes)
1986{
1987 struct vattr attr;
1988 void *save1; /* warning, save pointers do not always remain valid */
1989 void *save2;
1990 int error;
1991
1992 /*
1993 * Setup. Start the UNDO record, obtain a shared lock on the vnode,
1994 * and retrieve attribute info.
1995 */
1996 save1 = jrecord_push(jrec, JTYPE_UNDO);
1997 error = VOP_GETATTR(vp, &attr, curthread);
1998 if (error)
1999 goto done;
2000
2001 /*
2002 * Generate UNDO records as requested.
2003 */
2004 if (jrflags & JRUNDO_VATTR) {
2005 save2 = jrecord_push(jrec, JTYPE_VATTR);
2006 jrecord_leaf(jrec, JLEAF_VTYPE, &attr.va_type, sizeof(attr.va_type));
2007 if ((jrflags & JRUNDO_SIZE) && attr.va_size != VNOVAL)
2008 jrecord_leaf(jrec, JLEAF_SIZE, &attr.va_size, sizeof(attr.va_size));
2009 if ((jrflags & JRUNDO_UID) && attr.va_uid != VNOVAL)
2010 jrecord_leaf(jrec, JLEAF_UID, &attr.va_uid, sizeof(attr.va_uid));
2011 if ((jrflags & JRUNDO_GID) && attr.va_gid != VNOVAL)
2012 jrecord_leaf(jrec, JLEAF_GID, &attr.va_gid, sizeof(attr.va_gid));
2013 if ((jrflags & JRUNDO_FSID) && attr.va_fsid != VNOVAL)
2014 jrecord_leaf(jrec, JLEAF_FSID, &attr.va_fsid, sizeof(attr.va_fsid));
2015 if ((jrflags & JRUNDO_MODES) && attr.va_mode != (mode_t)VNOVAL)
2016 jrecord_leaf(jrec, JLEAF_MODES, &attr.va_mode, sizeof(attr.va_mode));
2017 if ((jrflags & JRUNDO_INUM) && attr.va_fileid != VNOVAL)
2018 jrecord_leaf(jrec, JLEAF_INUM, &attr.va_fileid, sizeof(attr.va_fileid));
2019 if ((jrflags & JRUNDO_ATIME) && attr.va_atime.tv_sec != VNOVAL)
2020 jrecord_leaf(jrec, JLEAF_ATIME, &attr.va_atime, sizeof(attr.va_atime));
2021 if ((jrflags & JRUNDO_MTIME) && attr.va_mtime.tv_sec != VNOVAL)
2022 jrecord_leaf(jrec, JLEAF_MTIME, &attr.va_mtime, sizeof(attr.va_mtime));
2023 if ((jrflags & JRUNDO_CTIME) && attr.va_ctime.tv_sec != VNOVAL)
2024 jrecord_leaf(jrec, JLEAF_CTIME, &attr.va_ctime, sizeof(attr.va_ctime));
2025 if ((jrflags & JRUNDO_GEN) && attr.va_gen != VNOVAL)
2026 jrecord_leaf(jrec, JLEAF_GEN, &attr.va_gen, sizeof(attr.va_gen));
2027 if ((jrflags & JRUNDO_FLAGS) && attr.va_flags != VNOVAL)
2028 jrecord_leaf(jrec, JLEAF_FLAGS, &attr.va_flags, sizeof(attr.va_flags));
2029 if ((jrflags & JRUNDO_UDEV) && attr.va_rdev != VNOVAL)
2030 jrecord_leaf(jrec, JLEAF_UDEV, &attr.va_rdev, sizeof(attr.va_rdev));
2031 jrecord_pop(jrec, save2);
2032 }
2033
2034 /*
2035 * Output the file data being overwritten by reading the file and
2036 * writing it out to the journal prior to the write operation. We
2037 * do not need to write out data past the current file EOF.
2038 *
2039 * XXX support JRUNDO_CONDLINK - do not write out file data for files
2040 * with a link count > 1. The undo code needs to locate the inode and
2041 * regenerate the hardlink.
2042 */
2043 if (jrflags & JRUNDO_FILEDATA) {
2044 if (attr.va_size != VNOVAL) {
2045 if (bytes == -1)
2046 bytes = attr.va_size - off;
2047 if (off + bytes > attr.va_size)
2048 bytes = attr.va_size - off;
2049 if (bytes > 0)
2050 jrecord_file_data(jrec, vp, off, bytes);
2051 } else {
2052 error = EINVAL;
2053 }
2054 }
2055done:
2056 if (error)
2057 jrecord_leaf(jrec, JLEAF_ERROR, &error, sizeof(error));
2058 jrecord_pop(jrec, save1);
2059}
2060
2281065e
MD
2061/************************************************************************
2062 * JOURNAL VNOPS *
558b8e00
MD
2063 ************************************************************************
2064 *
2065 * These are function shims replacing the normal filesystem ops. We become
2066 * responsible for calling the underlying filesystem ops. We have the choice
2067 * of executing the underlying op first and then generating the journal entry,
2068 * or starting the journal entry, executing the underlying op, and then
2069 * either completing or aborting it.
2070 *
2071 * The journal is supposed to be a high-level entity, which generally means
2072 * identifying files by name rather then by inode. Supplying both allows
2073 * the journal to be used both for inode-number-compatible 'mirrors' and
2074 * for simple filesystem replication.
2075 *
2076 * Writes are particularly difficult to deal with because a single write may
2077 * represent a hundred megabyte buffer or more, and both writes and truncations
2078 * require the 'old' data to be written out as well as the new data if the
2079 * log is reversable. Other issues:
2080 *
2081 * - How to deal with operations on unlinked files (no path available),
2082 * but which may still be filesystem visible due to hard links.
2083 *
2084 * - How to deal with modifications made via a memory map.
2085 *
2086 * - Future cache coherency support will require cache coherency API calls
2087 * both prior to and after the call to the underlying VFS.
2088 *
2089 * ALSO NOTE: We do not have to shim compatibility VOPs like MKDIR which have
2090 * new VFS equivalents (NMKDIR).
2091 */
2092
b2f7ec6c
MD
2093/*
2094 * Journal vop_settattr { a_vp, a_vap, a_cred, a_td }
2095 */
558b8e00
MD
2096static
2097int
2098journal_setattr(struct vop_setattr_args *ap)
2099{
26e603ed
MD
2100 struct jrecord_list jreclist;
2101 struct jrecord jreccache;
2102 struct jrecord *jrec;
558b8e00 2103 struct mount *mp;
558b8e00
MD
2104 int error;
2105
558b8e00 2106 mp = ap->a_head.a_ops->vv_mount;
26e603ed
MD
2107 if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_SETATTR)) {
2108 jreclist_undo_file(&jreclist, ap->a_vp, JRUNDO_VATTR, 0, 0);
2109 }
2110 error = vop_journal_operate_ap(&ap->a_head);
558b8e00 2111 if (error == 0) {
26e603ed
MD
2112 TAILQ_FOREACH(jrec, &jreclist, user_entry) {
2113 jrecord_write_cred(jrec, ap->a_td, ap->a_cred);
2114 jrecord_write_vnode_ref(jrec, ap->a_vp);
2115 jrecord_write_vattr(jrec, ap->a_vap);
558b8e00
MD
2116 }
2117 }
26e603ed 2118 jreclist_done(&jreclist, error);
558b8e00
MD
2119 return (error);
2120}
2121
b2f7ec6c
MD
2122/*
2123 * Journal vop_write { a_vp, a_uio, a_ioflag, a_cred }
2124 */
558b8e00
MD
2125static
2126int
2127journal_write(struct vop_write_args *ap)
2128{
26e603ed
MD
2129 struct jrecord_list jreclist;
2130 struct jrecord jreccache;
2131 struct jrecord *jrec;
558b8e00 2132 struct mount *mp;
9578bde0
MD
2133 struct uio uio_copy;
2134 struct iovec uio_one_iovec;
558b8e00
MD
2135 int error;
2136
9578bde0
MD
2137 /*
2138 * This is really nasty. UIO's don't retain sufficient information to
2139 * be reusable once they've gone through the VOP chain. The iovecs get
2140 * cleared, so we have to copy the UIO.
2141 *
2142 * XXX fix the UIO code to not destroy iov's during a scan so we can
2143 * reuse the uio over and over again.
d0887c34
MD
2144 *
2145 * XXX UNDO code needs to journal the old data prior to the write.
9578bde0
MD
2146 */
2147 uio_copy = *ap->a_uio;
2148 if (uio_copy.uio_iovcnt == 1) {
2149 uio_one_iovec = ap->a_uio->uio_iov[0];
2150 uio_copy.uio_iov = &uio_one_iovec;
2151 } else {
2152 uio_copy.uio_iov = malloc(uio_copy.uio_iovcnt * sizeof(struct iovec),
2153 M_JOURNAL, M_WAITOK);
2154 bcopy(ap->a_uio->uio_iov, uio_copy.uio_iov,
2155 uio_copy.uio_iovcnt * sizeof(struct iovec));
2156 }
2157
26e603ed
MD
2158 /*
2159 * Write out undo data. Note that uio_offset is incorrect if
2160 * IO_APPEND is set, but fortunately we have no undo file data to
2161 * write out in that case.
2162 */
2163 mp = ap->a_head.a_ops->vv_mount;
2164 if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_WRITE)) {
2165 if (ap->a_ioflag & IO_APPEND) {
2166 jreclist_undo_file(&jreclist, ap->a_vp, JRUNDO_SIZE|JRUNDO_MTIME, 0, 0);
2167 } else {
2168 jreclist_undo_file(&jreclist, ap->a_vp,
2169 JRUNDO_FILEDATA|JRUNDO_SIZE|JRUNDO_MTIME,
2170 uio_copy.uio_offset, uio_copy.uio_resid);
2171 }
2172 }
558b8e00 2173 error = vop_journal_operate_ap(&ap->a_head);
d0887c34
MD
2174
2175 /*
2176 * XXX bad hack to figure out the offset for O_APPEND writes (note:
2177 * uio field state after the VFS operation).
2178 */
2179 uio_copy.uio_offset = ap->a_uio->uio_offset -
26e603ed 2180 (uio_copy.uio_resid - ap->a_uio->uio_resid);
d0887c34 2181
26e603ed
MD
2182 /*
2183 * Output the write data to the journal.
2184 */
558b8e00 2185 if (error == 0) {
26e603ed
MD
2186 TAILQ_FOREACH(jrec, &jreclist, user_entry) {
2187 jrecord_write_cred(jrec, NULL, ap->a_cred);
2188 jrecord_write_vnode_ref(jrec, ap->a_vp);
2189 jrecord_write_uio(jrec, JLEAF_FILEDATA, &uio_copy);
558b8e00
MD
2190 }
2191 }
26e603ed 2192 jreclist_done(&jreclist, error);
9578bde0
MD
2193
2194 if (uio_copy.uio_iov != &uio_one_iovec)
2195 free(uio_copy.uio_iov, M_JOURNAL);
558b8e00
MD
2196 return (error);
2197}
2198
b2f7ec6c
MD
2199/*
2200 * Journal vop_fsync { a_vp, a_waitfor, a_td }
2201 */
558b8e00
MD
2202static
2203int
2204journal_fsync(struct vop_fsync_args *ap)
2205{
26e603ed 2206#if 0
558b8e00
MD
2207 struct mount *mp;
2208 struct journal *jo;
26e603ed 2209#endif
558b8e00
MD
2210 int error;
2211
2212 error = vop_journal_operate_ap(&ap->a_head);
26e603ed 2213#if 0
558b8e00
MD
2214 mp = ap->a_head.a_ops->vv_mount;
2215 if (error == 0) {
2216 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
2217 /* XXX synchronize pending journal records */
2218 }
2219 }
26e603ed 2220#endif
558b8e00
MD
2221 return (error);
2222}
2223
b2f7ec6c
MD
2224/*
2225 * Journal vop_putpages { a_vp, a_m, a_count, a_sync, a_rtvals, a_offset }
143c4f15
MD
2226 *
2227 * note: a_count is in bytes.
b2f7ec6c 2228 */
558b8e00
MD
2229static
2230int
2231journal_putpages(struct vop_putpages_args *ap)
2232{
26e603ed
MD
2233 struct jrecord_list jreclist;
2234 struct jrecord jreccache;
2235 struct jrecord *jrec;
558b8e00 2236 struct mount *mp;
558b8e00
MD
2237 int error;
2238
558b8e00 2239 mp = ap->a_head.a_ops->vv_mount;
26e603ed
MD
2240 if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_PUTPAGES) &&
2241 ap->a_count > 0
2242 ) {
2243 jreclist_undo_file(&jreclist, ap->a_vp,
2244 JRUNDO_FILEDATA|JRUNDO_SIZE|JRUNDO_MTIME,
2245 ap->a_offset, btoc(ap->a_count));
2246 }
2247 error = vop_journal_operate_ap(&ap->a_head);
143c4f15 2248 if (error == 0 && ap->a_count > 0) {
26e603ed
MD
2249 TAILQ_FOREACH(jrec, &jreclist, user_entry) {
2250 jrecord_write_vnode_ref(jrec, ap->a_vp);
2251 jrecord_write_pagelist(jrec, JLEAF_FILEDATA, ap->a_m, ap->a_rtvals,
2252 btoc(ap->a_count), ap->a_offset);
558b8e00
MD
2253 }
2254 }
26e603ed 2255 jreclist_done(&jreclist, error);
558b8e00
MD
2256 return (error);
2257}
2258
b2f7ec6c
MD
2259/*
2260 * Journal vop_setacl { a_vp, a_type, a_aclp, a_cred, a_td }
2261 */
558b8e00
MD
2262static
2263int
2264journal_setacl(struct vop_setacl_args *ap)
2265{
26e603ed
MD
2266 struct jrecord_list jreclist;
2267 struct jrecord jreccache;
2268 struct jrecord *jrec;
558b8e00 2269 struct mount *mp;
558b8e00
MD
2270 int error;
2271
558b8e00 2272 mp = ap->a_head.a_ops->vv_mount;
26e603ed
MD
2273 jreclist_init(mp, &jreclist, &jreccache, JTYPE_SETACL);
2274 error = vop_journal_operate_ap(&ap->a_head);
558b8e00 2275 if (error == 0) {
26e603ed
MD
2276 TAILQ_FOREACH(jrec, &jreclist, user_entry) {
2277#if 0
2278 if ((jo->flags & MC_JOURNAL_WANT_REVERSABLE))
2279 jrecord_undo_file(jrec, ap->a_vp, JRUNDO_XXX, 0, 0);
2280#endif
2281 jrecord_write_cred(jrec, ap->a_td, ap->a_cred);
2282 jrecord_write_vnode_ref(jrec, ap->a_vp);
b2f7ec6c 2283 /* XXX type, aclp */
558b8e00
MD
2284 }
2285 }
26e603ed 2286 jreclist_done(&jreclist, error);
558b8e00
MD
2287 return (error);
2288}
2289
b2f7ec6c
MD
2290/*
2291 * Journal vop_setextattr { a_vp, a_name, a_uio, a_cred, a_td }
2292 */
558b8e00
MD
2293static
2294int
2295journal_setextattr(struct vop_setextattr_args *ap)
2296{
26e603ed
MD
2297 struct jrecord_list jreclist;
2298 struct jrecord jreccache;
2299 struct jrecord *jrec;
558b8e00 2300 struct mount *mp;
558b8e00
MD
2301 int error;
2302
558b8e00 2303 mp = ap->a_head.a_ops->vv_mount;
26e603ed
MD
2304 jreclist_init(mp, &jreclist, &jreccache, JTYPE_SETEXTATTR);
2305 error = vop_journal_operate_ap(&ap->a_head);
558b8e00 2306 if (error == 0) {
26e603ed
MD
2307 TAILQ_FOREACH(jrec, &jreclist, user_entry) {
2308#if 0
2309 if ((jo->flags & MC_JOURNAL_WANT_REVERSABLE))
2310 jrecord_undo_file(jrec, ap->a_vp, JRUNDO_XXX, 0, 0);
2311#endif
2312 jrecord_write_cred(jrec, ap->a_td, ap->a_cred);
2313 jrecord_write_vnode_ref(jrec, ap->a_vp);
2314 jrecord_leaf(jrec, JLEAF_ATTRNAME, ap->a_name, strlen(ap->a_name));
2315 jrecord_write_uio(jrec, JLEAF_FILEDATA, ap->a_uio);
558b8e00
MD
2316 }
2317 }
26e603ed 2318 jreclist_done(&jreclist, error);
558b8e00
MD
2319 return (error);
2320}
2321
b2f7ec6c
MD
2322/*
2323 * Journal vop_ncreate { a_ncp, a_vpp, a_cred, a_vap }
2324 */
558b8e00
MD
2325static
2326int
2327journal_ncreate(struct vop_ncreate_args *ap)
2328{
26e603ed
MD
2329 struct jrecord_list jreclist;
2330 struct jrecord jreccache;
2331 struct jrecord *jrec;
558b8e00 2332 struct mount *mp;
558b8e00
MD
2333 int error;
2334
558b8e00 2335 mp = ap->a_head.a_ops->vv_mount;
26e603ed
MD
2336 jreclist_init(mp, &jreclist, &jreccache, JTYPE_CREATE);
2337 error = vop_journal_operate_ap(&ap->a_head);
558b8e00 2338 if (error == 0) {
26e603ed
MD
2339 TAILQ_FOREACH(jrec, &jreclist, user_entry) {
2340 jrecord_write_cred(jrec, NULL, ap->a_cred);
2341 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_ncp);
b2f7ec6c 2342 if (*ap->a_vpp)
26e603ed
MD
2343 jrecord_write_vnode_ref(jrec, *ap->a_vpp);
2344 jrecord_write_vattr(jrec, ap->a_vap);
558b8e00
MD
2345 }
2346 }
26e603ed 2347 jreclist_done(&jreclist, error);
558b8e00
MD
2348 return (error);
2349}
2350
b2f7ec6c
MD
2351/*
2352 * Journal vop_nmknod { a_ncp, a_vpp, a_cred, a_vap }
2353 */
558b8e00
MD
2354static
2355int
2356journal_nmknod(struct vop_nmknod_args *ap)
2357{
26e603ed
MD
2358 struct jrecord_list jreclist;
2359 struct jrecord jreccache;
2360 struct jrecord *jrec;
558b8e00 2361 struct mount *mp;
558b8e00
MD
2362 int error;
2363
558b8e00 2364 mp = ap->a_head.a_ops->vv_mount;
26e603ed
MD
2365 jreclist_init(mp, &jreclist, &jreccache, JTYPE_MKNOD);
2366 error = vop_journal_operate_ap(&ap->a_head);
558b8e00 2367 if (error == 0) {
26e603ed
MD
2368 TAILQ_FOREACH(jrec, &jreclist, user_entry) {
2369 jrecord_write_cred(jrec, NULL, ap->a_cred);
2370 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_ncp);
2371 jrecord_write_vattr(jrec, ap->a_vap);
b2f7ec6c 2372 if (*ap->a_vpp)
26e603ed 2373 jrecord_write_vnode_ref(jrec, *ap->a_vpp);
558b8e00
MD
2374 }
2375 }
26e603ed 2376 jreclist_done(&jreclist, error);
558b8e00
MD
2377 return (error);
2378}
2379
b2f7ec6c
MD
2380/*
2381 * Journal vop_nlink { a_ncp, a_vp, a_cred }
2382 */
558b8e00
MD
2383static
2384int
2385journal_nlink(struct vop_nlink_args *ap)
2386{
26e603ed
MD
2387 struct jrecord_list jreclist;
2388 struct jrecord jreccache;
2389 struct jrecord *jrec;
558b8e00 2390 struct mount *mp;
558b8e00
MD
2391 int error;
2392
558b8e00 2393 mp = ap->a_head.a_ops->vv_mount;
26e603ed
MD
2394 jreclist_init(mp, &jreclist, &jreccache, JTYPE_LINK);
2395 error = vop_journal_operate_ap(&ap->a_head);
558b8e00 2396 if (error == 0) {
26e603ed
MD
2397 TAILQ_FOREACH(jrec, &jreclist, user_entry) {
2398 jrecord_write_cred(jrec, NULL, ap->a_cred);
2399 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_ncp);
b2f7ec6c 2400 /* XXX PATH to VP and inode number */
f4659a6c
MD
2401 /* XXX this call may not record the correct path when
2402 * multiple paths are available */
26e603ed 2403 jrecord_write_vnode_link(jrec, ap->a_vp, ap->a_ncp);
558b8e00
MD
2404 }
2405 }
26e603ed 2406 jreclist_done(&jreclist, error);
558b8e00
MD
2407 return (error);
2408}
2409
b2f7ec6c
MD
2410/*
2411 * Journal vop_symlink { a_ncp, a_vpp, a_cred, a_vap, a_target }
2412 */
558b8e00
MD
2413static
2414int
2415journal_nsymlink(struct vop_nsymlink_args *ap)
2416{
26e603ed
MD
2417 struct jrecord_list jreclist;
2418 struct jrecord jreccache;
2419 struct jrecord *jrec;
558b8e00 2420 struct mount *mp;
558b8e00
MD
2421 int error;
2422
558b8e00 2423 mp = ap->a_head.a_ops->vv_mount;
26e603ed
MD
2424 jreclist_init(mp, &jreclist, &jreccache, JTYPE_SYMLINK);
2425 error = vop_journal_operate_ap(&ap->a_head);
558b8e00 2426 if (error == 0) {
26e603ed
MD
2427 TAILQ_FOREACH(jrec, &jreclist, user_entry) {
2428 jrecord_write_cred(jrec, NULL, ap->a_cred);
2429 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_ncp);
2430 jrecord_leaf(jrec, JLEAF_SYMLINKDATA,
b2f7ec6c
MD
2431 ap->a_target, strlen(ap->a_target));
2432 if (*ap->a_vpp)
26e603ed 2433 jrecord_write_vnode_ref(jrec, *ap->a_vpp);
558b8e00
MD
2434 }
2435 }
26e603ed 2436 jreclist_done(&jreclist, error);
558b8e00
MD
2437 return (error);
2438}
2439
b2f7ec6c
MD
2440/*
2441 * Journal vop_nwhiteout { a_ncp, a_cred, a_flags }
2442 */
558b8e00
MD
2443static
2444int
2445journal_nwhiteout(struct vop_nwhiteout_args *ap)
2446{
26e603ed
MD
2447 struct jrecord_list jreclist;
2448 struct jrecord jreccache;
2449 struct jrecord *jrec;
558b8e00 2450 struct mount *mp;
558b8e00
MD
2451 int error;
2452
558b8e00 2453 mp = ap->a_head.a_ops->vv_mount;
26e603ed
MD
2454 jreclist_init(mp, &jreclist, &jreccache, JTYPE_WHITEOUT);
2455 error = vop_journal_operate_ap(&ap->a_head);
558b8e00 2456 if (error == 0) {
26e603ed
MD
2457 TAILQ_FOREACH(jrec, &jreclist, user_entry) {
2458 jrecord_write_cred(jrec, NULL, ap->a_cred);
2459 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_ncp);
558b8e00
MD
2460 }
2461 }
26e603ed 2462 jreclist_done(&jreclist, error);
558b8e00
MD
2463 return (error);
2464}
2465
b2f7ec6c
MD
2466/*
2467 * Journal vop_nremove { a_ncp, a_cred }
2468 */
558b8e00
MD
2469static
2470int
2471journal_nremove(struct vop_nremove_args *ap)
2472{
26e603ed
MD
2473 struct jrecord_list jreclist;
2474 struct jrecord jreccache;
2475 struct jrecord *jrec;
558b8e00 2476 struct mount *mp;
558b8e00
MD
2477 int error;
2478
558b8e00 2479 mp = ap->a_head.a_ops->vv_mount;
26e603ed
MD
2480 if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_REMOVE) &&
2481 ap->a_ncp->nc_vp
2482 ) {
2483 jreclist_undo_file(&jreclist, ap->a_ncp->nc_vp,
2484 JRUNDO_ALL|JRUNDO_GETVP|JRUNDO_CONDLINK, 0, -1);
2485 }
2486 error = vop_journal_operate_ap(&ap->a_head);
558b8e00 2487 if (error == 0) {
26e603ed
MD
2488 TAILQ_FOREACH(jrec, &jreclist, user_entry) {
2489 jrecord_write_cred(jrec, NULL, ap->a_cred);
2490 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_ncp);
558b8e00
MD
2491 }
2492 }
26e603ed 2493 jreclist_done(&jreclist, error);
558b8e00
MD
2494 return (error);
2495}
2281065e 2496
b2f7ec6c
MD
2497/*
2498 * Journal vop_nmkdir { a_ncp, a_vpp, a_cred, a_vap }
2499 */
2281065e
MD
2500static
2501int
2502journal_nmkdir(struct vop_nmkdir_args *ap)
2503{
26e603ed
MD
2504 struct jrecord_list jreclist;
2505 struct jrecord jreccache;
2506 struct jrecord *jrec;
82eaef15 2507 struct mount *mp;
2281065e
MD
2508 int error;
2509
82eaef15 2510 mp = ap->a_head.a_ops->vv_mount;
26e603ed
MD
2511 jreclist_init(mp, &jreclist, &jreccache, JTYPE_MKDIR);
2512 error = vop_journal_operate_ap(&ap->a_head);
82eaef15 2513 if (error == 0) {
26e603ed 2514 TAILQ_FOREACH(jrec, &jreclist, user_entry) {
82eaef15
MD
2515#if 0
2516 if (jo->flags & MC_JOURNAL_WANT_AUDIT) {
26e603ed 2517 jrecord_write_audit(jrec);
82eaef15
MD
2518 }
2519#endif
26e603ed
MD
2520 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_ncp);
2521 jrecord_write_cred(jrec, NULL, ap->a_cred);
2522 jrecord_write_vattr(jrec, ap->a_vap);
2523 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_ncp);
b2f7ec6c 2524 if (*ap->a_vpp)
26e603ed 2525 jrecord_write_vnode_ref(jrec, *ap->a_vpp);
82eaef15
MD
2526 }
2527 }
26e603ed 2528 jreclist_done(&jreclist, error);
2281065e 2529 return (error);
6ddb7618
MD
2530}
2531
b2f7ec6c
MD
2532/*
2533 * Journal vop_nrmdir { a_ncp, a_cred }
2534 */
558b8e00
MD
2535static
2536int
2537journal_nrmdir(struct vop_nrmdir_args *ap)
2538{
26e603ed
MD
2539 struct jrecord_list jreclist;
2540 struct jrecord jreccache;
2541 struct jrecord *jrec;
558b8e00 2542 struct mount *mp;
558b8e00
MD
2543 int error;
2544
558b8e00 2545 mp = ap->a_head.a_ops->vv_mount;
26e603ed
MD
2546 if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_RMDIR)) {
2547 jreclist_undo_file(&jreclist, ap->a_ncp->nc_vp,
2548 JRUNDO_VATTR|JRUNDO_GETVP, 0, 0);
2549 }
2550 error = vop_journal_operate_ap(&ap->a_head);
558b8e00 2551 if (error == 0) {
26e603ed
MD
2552 TAILQ_FOREACH(jrec, &jreclist, user_entry) {
2553 jrecord_write_cred(jrec, NULL, ap->a_cred);
2554 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_ncp);
558b8e00
MD
2555 }
2556 }
26e603ed 2557 jreclist_done(&jreclist, error);
558b8e00
MD
2558 return (error);
2559}
2560
b2f7ec6c
MD
2561/*
2562 * Journal vop_nrename { a_fncp, a_tncp, a_cred }
2563 */
558b8e00
MD
2564static
2565int
2566journal_nrename(struct vop_nrename_args *ap)
2567{
26e603ed
MD
2568 struct jrecord_list jreclist;
2569 struct jrecord jreccache;
2570 struct jrecord *jrec;
558b8e00 2571 struct mount *mp;
558b8e00
MD
2572 int error;
2573
558b8e00 2574 mp = ap->a_head.a_ops->vv_mount;
26e603ed
MD
2575 if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_RENAME) &&
2576 ap->a_tncp->nc_vp
2577 ) {
2578 jreclist_undo_file(&jreclist, ap->a_tncp->nc_vp,
2579 JRUNDO_ALL|JRUNDO_GETVP|JRUNDO_CONDLINK, 0, -1);
2580 }
2581 error = vop_journal_operate_ap(&ap->a_head);
558b8e00 2582 if (error == 0) {
26e603ed
MD
2583 TAILQ_FOREACH(jrec, &jreclist, user_entry) {
2584 jrecord_write_cred(jrec, NULL, ap->a_cred);
2585 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_fncp);
2586 jrecord_write_path(jrec, JLEAF_PATH2, ap->a_tncp);
558b8e00
MD
2587 }
2588 }
26e603ed 2589 jreclist_done(&jreclist, error);
558b8e00
MD
2590 return (error);
2591}
2592