2 * Copyright (c) 2004-2006 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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
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.
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
34 * $DragonFly: src/sys/kern/vfs_jops.c,v 1.36 2007/08/21 17:43:52 dillon Exp $
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
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
48 #include <sys/param.h>
49 #include <sys/systm.h>
52 #include <sys/kernel.h>
53 #include <sys/queue.h>
55 #include <sys/malloc.h>
56 #include <sys/mount.h>
57 #include <sys/unistd.h>
58 #include <sys/vnode.h>
60 #include <sys/mountctl.h>
61 #include <sys/journal.h>
64 #include <sys/msfbuf.h>
65 #include <sys/socket.h>
66 #include <sys/socketvar.h>
68 #include <machine/limits.h>
71 #include <vm/vm_object.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_pager.h>
74 #include <vm/vnode_pager.h>
76 #include <sys/file2.h>
77 #include <sys/thread2.h>
79 static int journal_attach(struct mount
*mp
);
80 static void journal_detach(struct mount
*mp
);
81 static int journal_install_vfs_journal(struct mount
*mp
, struct file
*fp
,
82 const struct mountctl_install_journal
*info
);
83 static int journal_restart_vfs_journal(struct mount
*mp
, struct file
*fp
,
84 const struct mountctl_restart_journal
*info
);
85 static int journal_remove_vfs_journal(struct mount
*mp
,
86 const struct mountctl_remove_journal
*info
);
87 static int journal_restart(struct mount
*mp
, struct file
*fp
,
88 struct journal
*jo
, int flags
);
89 static int journal_destroy(struct mount
*mp
, struct journal
*jo
, int flags
);
90 static int journal_resync_vfs_journal(struct mount
*mp
, const void *ctl
);
91 static int journal_status_vfs_journal(struct mount
*mp
,
92 const struct mountctl_status_journal
*info
,
93 struct mountctl_journal_ret_status
*rstat
,
94 int buflen
, int *res
);
96 static void jrecord_undo_file(struct jrecord
*jrec
, struct vnode
*vp
,
97 int jrflags
, off_t off
, off_t bytes
);
99 static int journal_setattr(struct vop_setattr_args
*ap
);
100 static int journal_write(struct vop_write_args
*ap
);
101 static int journal_fsync(struct vop_fsync_args
*ap
);
102 static int journal_putpages(struct vop_putpages_args
*ap
);
103 static int journal_setacl(struct vop_setacl_args
*ap
);
104 static int journal_setextattr(struct vop_setextattr_args
*ap
);
105 static int journal_ncreate(struct vop_ncreate_args
*ap
);
106 static int journal_nmknod(struct vop_nmknod_args
*ap
);
107 static int journal_nlink(struct vop_nlink_args
*ap
);
108 static int journal_nsymlink(struct vop_nsymlink_args
*ap
);
109 static int journal_nwhiteout(struct vop_nwhiteout_args
*ap
);
110 static int journal_nremove(struct vop_nremove_args
*ap
);
111 static int journal_nmkdir(struct vop_nmkdir_args
*ap
);
112 static int journal_nrmdir(struct vop_nrmdir_args
*ap
);
113 static int journal_nrename(struct vop_nrename_args
*ap
);
115 #define JRUNDO_SIZE 0x00000001
116 #define JRUNDO_UID 0x00000002
117 #define JRUNDO_GID 0x00000004
118 #define JRUNDO_FSID 0x00000008
119 #define JRUNDO_MODES 0x00000010
120 #define JRUNDO_INUM 0x00000020
121 #define JRUNDO_ATIME 0x00000040
122 #define JRUNDO_MTIME 0x00000080
123 #define JRUNDO_CTIME 0x00000100
124 #define JRUNDO_GEN 0x00000200
125 #define JRUNDO_FLAGS 0x00000400
126 #define JRUNDO_UDEV 0x00000800
127 #define JRUNDO_NLINK 0x00001000
128 #define JRUNDO_FILEDATA 0x00010000
129 #define JRUNDO_GETVP 0x00020000
130 #define JRUNDO_CONDLINK 0x00040000 /* write file data if link count 1 */
131 #define JRUNDO_VATTR (JRUNDO_SIZE|JRUNDO_UID|JRUNDO_GID|JRUNDO_FSID|\
132 JRUNDO_MODES|JRUNDO_INUM|JRUNDO_ATIME|JRUNDO_MTIME|\
133 JRUNDO_CTIME|JRUNDO_GEN|JRUNDO_FLAGS|JRUNDO_UDEV|\
135 #define JRUNDO_ALL (JRUNDO_VATTR|JRUNDO_FILEDATA)
137 static struct vop_ops journal_vnode_vops
= {
138 .vop_default
= vop_journal_operate_ap
,
139 .vop_mountctl
= journal_mountctl
,
140 .vop_setattr
= journal_setattr
,
141 .vop_write
= journal_write
,
142 .vop_fsync
= journal_fsync
,
143 .vop_putpages
= journal_putpages
,
144 .vop_setacl
= journal_setacl
,
145 .vop_setextattr
= journal_setextattr
,
146 .vop_ncreate
= journal_ncreate
,
147 .vop_nmknod
= journal_nmknod
,
148 .vop_nlink
= journal_nlink
,
149 .vop_nsymlink
= journal_nsymlink
,
150 .vop_nwhiteout
= journal_nwhiteout
,
151 .vop_nremove
= journal_nremove
,
152 .vop_nmkdir
= journal_nmkdir
,
153 .vop_nrmdir
= journal_nrmdir
,
154 .vop_nrename
= journal_nrename
158 journal_mountctl(struct vop_mountctl_args
*ap
)
163 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
166 if (mp
->mnt_vn_journal_ops
== NULL
) {
168 case MOUNTCTL_INSTALL_VFS_JOURNAL
:
169 error
= journal_attach(mp
);
170 if (error
== 0 && ap
->a_ctllen
!= sizeof(struct mountctl_install_journal
))
172 if (error
== 0 && ap
->a_fp
== NULL
)
175 error
= journal_install_vfs_journal(mp
, ap
->a_fp
, ap
->a_ctl
);
176 if (TAILQ_EMPTY(&mp
->mnt_jlist
))
179 case MOUNTCTL_RESTART_VFS_JOURNAL
:
180 case MOUNTCTL_REMOVE_VFS_JOURNAL
:
181 case MOUNTCTL_RESYNC_VFS_JOURNAL
:
182 case MOUNTCTL_STATUS_VFS_JOURNAL
:
191 case MOUNTCTL_INSTALL_VFS_JOURNAL
:
192 if (ap
->a_ctllen
!= sizeof(struct mountctl_install_journal
))
194 if (error
== 0 && ap
->a_fp
== NULL
)
197 error
= journal_install_vfs_journal(mp
, ap
->a_fp
, ap
->a_ctl
);
199 case MOUNTCTL_RESTART_VFS_JOURNAL
:
200 if (ap
->a_ctllen
!= sizeof(struct mountctl_restart_journal
))
202 if (error
== 0 && ap
->a_fp
== NULL
)
205 error
= journal_restart_vfs_journal(mp
, ap
->a_fp
, ap
->a_ctl
);
207 case MOUNTCTL_REMOVE_VFS_JOURNAL
:
208 if (ap
->a_ctllen
!= sizeof(struct mountctl_remove_journal
))
211 error
= journal_remove_vfs_journal(mp
, ap
->a_ctl
);
212 if (TAILQ_EMPTY(&mp
->mnt_jlist
))
215 case MOUNTCTL_RESYNC_VFS_JOURNAL
:
216 if (ap
->a_ctllen
!= 0)
218 error
= journal_resync_vfs_journal(mp
, ap
->a_ctl
);
220 case MOUNTCTL_STATUS_VFS_JOURNAL
:
221 if (ap
->a_ctllen
!= sizeof(struct mountctl_status_journal
))
224 error
= journal_status_vfs_journal(mp
, ap
->a_ctl
,
225 ap
->a_buf
, ap
->a_buflen
, ap
->a_res
);
237 * High level mount point setup. When a
240 journal_attach(struct mount
*mp
)
242 KKASSERT(mp
->mnt_jbitmap
== NULL
);
243 vfs_add_vnodeops(mp
, &journal_vnode_vops
, &mp
->mnt_vn_journal_ops
);
244 mp
->mnt_jbitmap
= kmalloc(JREC_STREAMID_JMAX
/8, M_JOURNAL
, M_WAITOK
|M_ZERO
);
245 mp
->mnt_streamid
= JREC_STREAMID_JMIN
;
250 journal_detach(struct mount
*mp
)
252 KKASSERT(mp
->mnt_jbitmap
!= NULL
);
253 if (mp
->mnt_vn_journal_ops
)
254 vfs_rm_vnodeops(mp
, &journal_vnode_vops
, &mp
->mnt_vn_journal_ops
);
255 kfree(mp
->mnt_jbitmap
, M_JOURNAL
);
256 mp
->mnt_jbitmap
= NULL
;
260 * Install a journal on a mount point. Each journal has an associated worker
261 * thread which is responsible for buffering and spooling the data to the
262 * target. A mount point may have multiple journals attached to it. An
263 * initial start record is generated when the journal is associated.
266 journal_install_vfs_journal(struct mount
*mp
, struct file
*fp
,
267 const struct mountctl_install_journal
*info
)
274 jo
= kmalloc(sizeof(struct journal
), M_JOURNAL
, M_WAITOK
|M_ZERO
);
275 bcopy(info
->id
, jo
->id
, sizeof(jo
->id
));
276 jo
->flags
= info
->flags
& ~(MC_JOURNAL_WACTIVE
| MC_JOURNAL_RACTIVE
|
277 MC_JOURNAL_STOP_REQ
);
280 * Memory FIFO size, round to nearest power of 2
282 if (info
->membufsize
) {
283 if (info
->membufsize
< 65536)
285 else if (info
->membufsize
> 128 * 1024 * 1024)
286 size
= 128 * 1024 * 1024;
288 size
= (int)info
->membufsize
;
293 while (jo
->fifo
.size
< size
)
297 * Other parameters. If not specified the starting transaction id
298 * will be the current date.
301 jo
->transid
= info
->transid
;
305 jo
->transid
= ((int64_t)ts
.tv_sec
<< 30) | ts
.tv_nsec
;
311 * Allocate the memory FIFO
313 jo
->fifo
.mask
= jo
->fifo
.size
- 1;
314 jo
->fifo
.membase
= kmalloc(jo
->fifo
.size
, M_JFIFO
, M_WAITOK
|M_ZERO
|M_NULLOK
);
315 if (jo
->fifo
.membase
== NULL
)
319 * Create the worker threads and generate the association record.
322 kfree(jo
, M_JOURNAL
);
325 journal_create_threads(jo
);
326 jrecord_init(jo
, &jrec
, JREC_STREAMID_DISCONT
);
327 jrecord_write(&jrec
, JTYPE_ASSOCIATE
, 0);
328 jrecord_done(&jrec
, 0);
329 TAILQ_INSERT_TAIL(&mp
->mnt_jlist
, jo
, jentry
);
335 * Restart a journal with a new descriptor. The existing reader and writer
336 * threads are terminated and a new descriptor is associated with the
337 * journal. The FIFO rindex is reset to xindex and the threads are then
341 journal_restart_vfs_journal(struct mount
*mp
, struct file
*fp
,
342 const struct mountctl_restart_journal
*info
)
347 TAILQ_FOREACH(jo
, &mp
->mnt_jlist
, jentry
) {
348 if (bcmp(jo
->id
, info
->id
, sizeof(jo
->id
)) == 0)
352 error
= journal_restart(mp
, fp
, jo
, info
->flags
);
359 journal_restart(struct mount
*mp
, struct file
*fp
,
360 struct journal
*jo
, int flags
)
368 * Record the fact that we are doing a restart in the journal.
369 * XXX it isn't safe to do this if the journal is being restarted
370 * because it was locked up and the writer thread has already exited.
372 jrecord_init(jo
, &jrec
, JREC_STREAMID_RESTART
);
373 jrecord_write(&jrec
, JTYPE_DISASSOCIATE
, 0);
374 jrecord_done(&jrec
, 0);
378 * Stop the reader and writer threads and clean up the current
381 kprintf("RESTART WITH FP %p KILLING %p\n", fp
, jo
->fp
);
382 journal_destroy_threads(jo
, flags
);
388 * Associate the new descriptor, reset the FIFO index, and recreate
393 jo
->fifo
.rindex
= jo
->fifo
.xindex
;
394 journal_create_threads(jo
);
400 * Disassociate a journal from a mount point and terminate its worker thread.
401 * A final termination record is written out before the file pointer is
405 journal_remove_vfs_journal(struct mount
*mp
,
406 const struct mountctl_remove_journal
*info
)
411 TAILQ_FOREACH(jo
, &mp
->mnt_jlist
, jentry
) {
412 if (bcmp(jo
->id
, info
->id
, sizeof(jo
->id
)) == 0)
416 error
= journal_destroy(mp
, jo
, info
->flags
);
423 * Remove all journals associated with a mount point. Usually called
424 * by the umount code.
427 journal_remove_all_journals(struct mount
*mp
, int flags
)
431 while ((jo
= TAILQ_FIRST(&mp
->mnt_jlist
)) != NULL
) {
432 journal_destroy(mp
, jo
, flags
);
437 journal_destroy(struct mount
*mp
, struct journal
*jo
, int flags
)
441 TAILQ_REMOVE(&mp
->mnt_jlist
, jo
, jentry
);
443 jrecord_init(jo
, &jrec
, JREC_STREAMID_DISCONT
);
444 jrecord_write(&jrec
, JTYPE_DISASSOCIATE
, 0);
445 jrecord_done(&jrec
, 0);
447 journal_destroy_threads(jo
, flags
);
451 if (jo
->fifo
.membase
)
452 kfree(jo
->fifo
.membase
, M_JFIFO
);
453 kfree(jo
, M_JOURNAL
);
459 journal_resync_vfs_journal(struct mount
*mp
, const void *ctl
)
465 journal_status_vfs_journal(struct mount
*mp
,
466 const struct mountctl_status_journal
*info
,
467 struct mountctl_journal_ret_status
*rstat
,
468 int buflen
, int *res
)
476 TAILQ_FOREACH(jo
, &mp
->mnt_jlist
, jentry
) {
477 if (info
->index
== MC_JOURNAL_INDEX_ID
) {
478 if (bcmp(jo
->id
, info
->id
, sizeof(jo
->id
)) != 0)
480 } else if (info
->index
>= 0) {
481 if (info
->index
< index
)
483 } else if (info
->index
!= MC_JOURNAL_INDEX_ALL
) {
486 if (buflen
< sizeof(*rstat
)) {
488 rstat
[-1].flags
|= MC_JOURNAL_STATUS_MORETOCOME
;
493 bzero(rstat
, sizeof(*rstat
));
494 rstat
->recsize
= sizeof(*rstat
);
495 bcopy(jo
->id
, rstat
->id
, sizeof(jo
->id
));
496 rstat
->index
= index
;
497 rstat
->membufsize
= jo
->fifo
.size
;
498 rstat
->membufused
= jo
->fifo
.windex
- jo
->fifo
.xindex
;
499 rstat
->membufunacked
= jo
->fifo
.rindex
- jo
->fifo
.xindex
;
500 rstat
->bytessent
= jo
->total_acked
;
501 rstat
->fifostalls
= jo
->fifostalls
;
504 *res
+= sizeof(*rstat
);
505 buflen
-= sizeof(*rstat
);
510 /************************************************************************
511 * PARALLEL TRANSACTION SUPPORT ROUTINES *
512 ************************************************************************
514 * JRECLIST_*() - routines which create and iterate over jrecord structures,
515 * because a mount point may have multiple attached journals.
519 * Initialize the passed jrecord_list and create a jrecord for each
520 * journal we need to write to. Unnecessary mallocs are avoided by
521 * using the passed jrecord structure as the first jrecord in the list.
522 * A starting transaction is pushed for each jrecord.
524 * Returns non-zero if any of the journals require undo records.
528 jreclist_init(struct mount
*mp
, struct jrecord_list
*jreclist
,
529 struct jrecord
*jreccache
, int16_t rectype
)
532 struct jrecord
*jrec
;
537 TAILQ_INIT(&jreclist
->list
);
540 * Select the stream ID to use for the transaction. We must select
541 * a stream ID that is not currently in use by some other parallel
544 * Don't bother calculating the next streamid when reassigning
545 * mnt_streamid, since parallel transactions are fairly rare. This
546 * also allows someone observing the raw records to clearly see
547 * when parallel transactions occur.
549 streamid
= mp
->mnt_streamid
;
551 while (mp
->mnt_jbitmap
[streamid
>> 3] & (1 << (streamid
& 7))) {
552 if (++streamid
== JREC_STREAMID_JMAX
)
553 streamid
= JREC_STREAMID_JMIN
;
554 if (++count
== JREC_STREAMID_JMAX
- JREC_STREAMID_JMIN
) {
555 kprintf("jreclist_init: all streamid's in use! sleeping\n");
556 tsleep(jreclist
, 0, "jsidfl", hz
* 10);
560 mp
->mnt_jbitmap
[streamid
>> 3] |= 1 << (streamid
& 7);
561 mp
->mnt_streamid
= streamid
;
562 jreclist
->streamid
= streamid
;
565 * Now initialize a stream on each journal.
569 TAILQ_FOREACH(jo
, &mp
->mnt_jlist
, jentry
) {
573 jrec
= kmalloc(sizeof(*jrec
), M_JOURNAL
, M_WAITOK
);
574 jrecord_init(jo
, jrec
, streamid
);
575 jrec
->user_save
= jrecord_push(jrec
, rectype
);
576 TAILQ_INSERT_TAIL(&jreclist
->list
, jrec
, user_entry
);
577 if (jo
->flags
& MC_JOURNAL_WANT_REVERSABLE
)
585 * Terminate the journaled transactions started by jreclist_init(). If
586 * an error occured, the transaction records will be aborted.
590 jreclist_done(struct mount
*mp
, struct jrecord_list
*jreclist
, int error
)
592 struct jrecord
*jrec
;
596 * Cleanup the jrecord state on each journal.
598 TAILQ_FOREACH(jrec
, &jreclist
->list
, user_entry
) {
599 jrecord_pop(jrec
, jrec
->user_save
);
600 jrecord_done(jrec
, error
);
604 * Free allocated jrec's (the first is always supplied)
607 while ((jrec
= TAILQ_FIRST(&jreclist
->list
)) != NULL
) {
608 TAILQ_REMOVE(&jreclist
->list
, jrec
, user_entry
);
610 kfree(jrec
, M_JOURNAL
);
615 * Clear the streamid so it can be reused.
617 mp
->mnt_jbitmap
[jreclist
->streamid
>> 3] &= ~(1 << (jreclist
->streamid
& 7));
621 * This procedure writes out UNDO records for available reversable
624 * XXX could use improvement. There is no need to re-read the file
629 jreclist_undo_file(struct jrecord_list
*jreclist
, struct vnode
*vp
,
630 int jrflags
, off_t off
, off_t bytes
)
632 struct jrecord
*jrec
;
636 if (jrflags
& JRUNDO_GETVP
)
637 error
= vget(vp
, LK_SHARED
);
639 TAILQ_FOREACH(jrec
, &jreclist
->list
, user_entry
) {
640 if (jrec
->jo
->flags
& MC_JOURNAL_WANT_REVERSABLE
) {
641 jrecord_undo_file(jrec
, vp
, jrflags
, off
, bytes
);
645 if (error
== 0 && jrflags
& JRUNDO_GETVP
)
649 /************************************************************************
650 * LOW LEVEL UNDO SUPPORT ROUTINE *
651 ************************************************************************
653 * This function is used to support UNDO records. It will generate an
654 * appropriate record with the requested portion of the file data. Note
655 * that file data is only recorded if JRUNDO_FILEDATA is passed. If bytes
656 * is -1, it will be set to the size of the file.
659 jrecord_undo_file(struct jrecord
*jrec
, struct vnode
*vp
, int jrflags
,
660 off_t off
, off_t bytes
)
663 void *save1
; /* warning, save pointers do not always remain valid */
668 * Setup. Start the UNDO record, obtain a shared lock on the vnode,
669 * and retrieve attribute info.
671 save1
= jrecord_push(jrec
, JTYPE_UNDO
);
672 error
= VOP_GETATTR(vp
, &attr
);
677 * Generate UNDO records as requested.
679 if (jrflags
& JRUNDO_VATTR
) {
680 save2
= jrecord_push(jrec
, JTYPE_VATTR
);
681 jrecord_leaf(jrec
, JLEAF_VTYPE
, &attr
.va_type
, sizeof(attr
.va_type
));
682 if ((jrflags
& JRUNDO_NLINK
) && attr
.va_nlink
!= VNOVAL
)
683 jrecord_leaf(jrec
, JLEAF_NLINK
, &attr
.va_nlink
, sizeof(attr
.va_nlink
));
684 if ((jrflags
& JRUNDO_SIZE
) && attr
.va_size
!= VNOVAL
)
685 jrecord_leaf(jrec
, JLEAF_SIZE
, &attr
.va_size
, sizeof(attr
.va_size
));
686 if ((jrflags
& JRUNDO_UID
) && attr
.va_uid
!= VNOVAL
)
687 jrecord_leaf(jrec
, JLEAF_UID
, &attr
.va_uid
, sizeof(attr
.va_uid
));
688 if ((jrflags
& JRUNDO_GID
) && attr
.va_gid
!= VNOVAL
)
689 jrecord_leaf(jrec
, JLEAF_GID
, &attr
.va_gid
, sizeof(attr
.va_gid
));
690 if ((jrflags
& JRUNDO_FSID
) && attr
.va_fsid
!= VNOVAL
)
691 jrecord_leaf(jrec
, JLEAF_FSID
, &attr
.va_fsid
, sizeof(attr
.va_fsid
));
692 if ((jrflags
& JRUNDO_MODES
) && attr
.va_mode
!= (mode_t
)VNOVAL
)
693 jrecord_leaf(jrec
, JLEAF_MODES
, &attr
.va_mode
, sizeof(attr
.va_mode
));
694 if ((jrflags
& JRUNDO_INUM
) && attr
.va_fileid
!= VNOVAL
)
695 jrecord_leaf(jrec
, JLEAF_INUM
, &attr
.va_fileid
, sizeof(attr
.va_fileid
));
696 if ((jrflags
& JRUNDO_ATIME
) && attr
.va_atime
.tv_sec
!= VNOVAL
)
697 jrecord_leaf(jrec
, JLEAF_ATIME
, &attr
.va_atime
, sizeof(attr
.va_atime
));
698 if ((jrflags
& JRUNDO_MTIME
) && attr
.va_mtime
.tv_sec
!= VNOVAL
)
699 jrecord_leaf(jrec
, JLEAF_MTIME
, &attr
.va_mtime
, sizeof(attr
.va_mtime
));
700 if ((jrflags
& JRUNDO_CTIME
) && attr
.va_ctime
.tv_sec
!= VNOVAL
)
701 jrecord_leaf(jrec
, JLEAF_CTIME
, &attr
.va_ctime
, sizeof(attr
.va_ctime
));
702 if ((jrflags
& JRUNDO_GEN
) && attr
.va_gen
!= VNOVAL
)
703 jrecord_leaf(jrec
, JLEAF_GEN
, &attr
.va_gen
, sizeof(attr
.va_gen
));
704 if ((jrflags
& JRUNDO_FLAGS
) && attr
.va_flags
!= VNOVAL
)
705 jrecord_leaf(jrec
, JLEAF_FLAGS
, &attr
.va_flags
, sizeof(attr
.va_flags
));
706 if ((jrflags
& JRUNDO_UDEV
) && attr
.va_rmajor
!= VNOVAL
) {
707 udev_t rdev
= makeudev(attr
.va_rmajor
, attr
.va_rminor
);
708 jrecord_leaf(jrec
, JLEAF_UDEV
, &rdev
, sizeof(rdev
));
709 jrecord_leaf(jrec
, JLEAF_UMAJOR
, &attr
.va_rmajor
, sizeof(attr
.va_rmajor
));
710 jrecord_leaf(jrec
, JLEAF_UMINOR
, &attr
.va_rminor
, sizeof(attr
.va_rminor
));
712 jrecord_pop(jrec
, save2
);
716 * Output the file data being overwritten by reading the file and
717 * writing it out to the journal prior to the write operation. We
718 * do not need to write out data past the current file EOF.
720 * XXX support JRUNDO_CONDLINK - do not write out file data for files
721 * with a link count > 1. The undo code needs to locate the inode and
722 * regenerate the hardlink.
724 if ((jrflags
& JRUNDO_FILEDATA
) && attr
.va_type
== VREG
) {
725 if (attr
.va_size
!= VNOVAL
) {
727 bytes
= attr
.va_size
- off
;
728 if (off
+ bytes
> attr
.va_size
)
729 bytes
= attr
.va_size
- off
;
731 jrecord_file_data(jrec
, vp
, off
, bytes
);
736 if ((jrflags
& JRUNDO_FILEDATA
) && attr
.va_type
== VLNK
) {
741 buf
= kmalloc(PATH_MAX
, M_JOURNAL
, M_WAITOK
);
743 aiov
.iov_len
= PATH_MAX
;
744 auio
.uio_iov
= &aiov
;
747 auio
.uio_rw
= UIO_READ
;
748 auio
.uio_segflg
= UIO_SYSSPACE
;
749 auio
.uio_td
= curthread
;
750 auio
.uio_resid
= PATH_MAX
;
751 error
= VOP_READLINK(vp
, &auio
, proc0
.p_ucred
);
753 jrecord_leaf(jrec
, JLEAF_SYMLINKDATA
, buf
,
754 PATH_MAX
- auio
.uio_resid
);
756 kfree(buf
, M_JOURNAL
);
760 jrecord_leaf(jrec
, JLEAF_ERROR
, &error
, sizeof(error
));
761 jrecord_pop(jrec
, save1
);
764 /************************************************************************
766 ************************************************************************
768 * These are function shims replacing the normal filesystem ops. We become
769 * responsible for calling the underlying filesystem ops. We have the choice
770 * of executing the underlying op first and then generating the journal entry,
771 * or starting the journal entry, executing the underlying op, and then
772 * either completing or aborting it.
774 * The journal is supposed to be a high-level entity, which generally means
775 * identifying files by name rather then by inode. Supplying both allows
776 * the journal to be used both for inode-number-compatible 'mirrors' and
777 * for simple filesystem replication.
779 * Writes are particularly difficult to deal with because a single write may
780 * represent a hundred megabyte buffer or more, and both writes and truncations
781 * require the 'old' data to be written out as well as the new data if the
782 * log is reversable. Other issues:
784 * - How to deal with operations on unlinked files (no path available),
785 * but which may still be filesystem visible due to hard links.
787 * - How to deal with modifications made via a memory map.
789 * - Future cache coherency support will require cache coherency API calls
790 * both prior to and after the call to the underlying VFS.
792 * ALSO NOTE: We do not have to shim compatibility VOPs like MKDIR which have
793 * new VFS equivalents (NMKDIR).
797 * Journal vop_setattr { a_vp, a_vap, a_cred }
801 journal_setattr(struct vop_setattr_args
*ap
)
803 struct jrecord_list jreclist
;
804 struct jrecord jreccache
;
805 struct jrecord
*jrec
;
810 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
811 if (jreclist_init(mp
, &jreclist
, &jreccache
, JTYPE_SETATTR
)) {
812 jreclist_undo_file(&jreclist
, ap
->a_vp
, JRUNDO_VATTR
, 0, 0);
814 error
= vop_journal_operate_ap(&ap
->a_head
);
816 TAILQ_FOREACH(jrec
, &jreclist
.list
, user_entry
) {
817 jrecord_write_cred(jrec
, curthread
, ap
->a_cred
);
818 jrecord_write_vnode_ref(jrec
, ap
->a_vp
);
819 save
= jrecord_push(jrec
, JTYPE_REDO
);
820 jrecord_write_vattr(jrec
, ap
->a_vap
);
821 jrecord_pop(jrec
, save
);
824 jreclist_done(mp
, &jreclist
, error
);
829 * Journal vop_write { a_vp, a_uio, a_ioflag, a_cred }
833 journal_write(struct vop_write_args
*ap
)
835 struct jrecord_list jreclist
;
836 struct jrecord jreccache
;
837 struct jrecord
*jrec
;
840 struct iovec uio_one_iovec
;
845 * Special synchronizing writes for VM backing store do not supply any
848 if (ap
->a_uio
->uio_segflg
== UIO_NOCOPY
) {
849 error
= vop_journal_operate_ap(&ap
->a_head
);
854 * This is really nasty. UIO's don't retain sufficient information to
855 * be reusable once they've gone through the VOP chain. The iovecs get
856 * cleared, so we have to copy the UIO.
858 * XXX fix the UIO code to not destroy iov's during a scan so we can
859 * reuse the uio over and over again.
861 * XXX UNDO code needs to journal the old data prior to the write.
863 uio_copy
= *ap
->a_uio
;
864 if (uio_copy
.uio_iovcnt
== 1) {
865 uio_one_iovec
= ap
->a_uio
->uio_iov
[0];
866 uio_copy
.uio_iov
= &uio_one_iovec
;
868 uio_copy
.uio_iov
= kmalloc(uio_copy
.uio_iovcnt
* sizeof(struct iovec
),
869 M_JOURNAL
, M_WAITOK
);
870 bcopy(ap
->a_uio
->uio_iov
, uio_copy
.uio_iov
,
871 uio_copy
.uio_iovcnt
* sizeof(struct iovec
));
875 * Write out undo data. Note that uio_offset is incorrect if
876 * IO_APPEND is set, but fortunately we have no undo file data to
877 * write out in that case.
879 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
880 if (jreclist_init(mp
, &jreclist
, &jreccache
, JTYPE_WRITE
)) {
881 if (ap
->a_ioflag
& IO_APPEND
) {
882 jreclist_undo_file(&jreclist
, ap
->a_vp
, JRUNDO_SIZE
|JRUNDO_MTIME
, 0, 0);
884 jreclist_undo_file(&jreclist
, ap
->a_vp
,
885 JRUNDO_FILEDATA
|JRUNDO_SIZE
|JRUNDO_MTIME
,
886 uio_copy
.uio_offset
, uio_copy
.uio_resid
);
889 error
= vop_journal_operate_ap(&ap
->a_head
);
892 * XXX bad hack to figure out the offset for O_APPEND writes (note:
893 * uio field state after the VFS operation).
895 uio_copy
.uio_offset
= ap
->a_uio
->uio_offset
-
896 (uio_copy
.uio_resid
- ap
->a_uio
->uio_resid
);
899 * Output the write data to the journal.
902 TAILQ_FOREACH(jrec
, &jreclist
.list
, user_entry
) {
903 jrecord_write_cred(jrec
, NULL
, ap
->a_cred
);
904 jrecord_write_vnode_ref(jrec
, ap
->a_vp
);
905 save
= jrecord_push(jrec
, JTYPE_REDO
);
906 jrecord_write_uio(jrec
, JLEAF_FILEDATA
, &uio_copy
);
907 jrecord_pop(jrec
, save
);
910 jreclist_done(mp
, &jreclist
, error
);
912 if (uio_copy
.uio_iov
!= &uio_one_iovec
)
913 kfree(uio_copy
.uio_iov
, M_JOURNAL
);
918 * Journal vop_fsync { a_vp, a_waitfor }
922 journal_fsync(struct vop_fsync_args
*ap
)
930 error
= vop_journal_operate_ap(&ap
->a_head
);
932 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
934 TAILQ_FOREACH(jo
, &mp
->mnt_jlist
, jentry
) {
935 /* XXX synchronize pending journal records */
943 * Journal vop_putpages { a_vp, a_m, a_count, a_sync, a_rtvals, a_offset }
945 * note: a_count is in bytes.
949 journal_putpages(struct vop_putpages_args
*ap
)
951 struct jrecord_list jreclist
;
952 struct jrecord jreccache
;
953 struct jrecord
*jrec
;
958 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
959 if (jreclist_init(mp
, &jreclist
, &jreccache
, JTYPE_PUTPAGES
) &&
962 jreclist_undo_file(&jreclist
, ap
->a_vp
,
963 JRUNDO_FILEDATA
|JRUNDO_SIZE
|JRUNDO_MTIME
,
964 ap
->a_offset
, btoc(ap
->a_count
));
966 error
= vop_journal_operate_ap(&ap
->a_head
);
967 if (error
== 0 && ap
->a_count
> 0) {
968 TAILQ_FOREACH(jrec
, &jreclist
.list
, user_entry
) {
969 jrecord_write_vnode_ref(jrec
, ap
->a_vp
);
970 save
= jrecord_push(jrec
, JTYPE_REDO
);
971 jrecord_write_pagelist(jrec
, JLEAF_FILEDATA
, ap
->a_m
, ap
->a_rtvals
,
972 btoc(ap
->a_count
), ap
->a_offset
);
973 jrecord_pop(jrec
, save
);
976 jreclist_done(mp
, &jreclist
, error
);
981 * Journal vop_setacl { a_vp, a_type, a_aclp, a_cred }
985 journal_setacl(struct vop_setacl_args
*ap
)
987 struct jrecord_list jreclist
;
988 struct jrecord jreccache
;
989 struct jrecord
*jrec
;
993 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
994 jreclist_init(mp
, &jreclist
, &jreccache
, JTYPE_SETACL
);
995 error
= vop_journal_operate_ap(&ap
->a_head
);
997 TAILQ_FOREACH(jrec
, &jreclist
.list
, user_entry
) {
999 if ((jo
->flags
& MC_JOURNAL_WANT_REVERSABLE
))
1000 jrecord_undo_file(jrec
, ap
->a_vp
, JRUNDO_XXX
, 0, 0);
1002 jrecord_write_cred(jrec
, curthread
, ap
->a_cred
);
1003 jrecord_write_vnode_ref(jrec
, ap
->a_vp
);
1005 save
= jrecord_push(jrec
, JTYPE_REDO
);
1006 /* XXX type, aclp */
1007 jrecord_pop(jrec
, save
);
1011 jreclist_done(mp
, &jreclist
, error
);
1016 * Journal vop_setextattr { a_vp, a_name, a_uio, a_cred }
1020 journal_setextattr(struct vop_setextattr_args
*ap
)
1022 struct jrecord_list jreclist
;
1023 struct jrecord jreccache
;
1024 struct jrecord
*jrec
;
1029 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
1030 jreclist_init(mp
, &jreclist
, &jreccache
, JTYPE_SETEXTATTR
);
1031 error
= vop_journal_operate_ap(&ap
->a_head
);
1033 TAILQ_FOREACH(jrec
, &jreclist
.list
, user_entry
) {
1035 if ((jo
->flags
& MC_JOURNAL_WANT_REVERSABLE
))
1036 jrecord_undo_file(jrec
, ap
->a_vp
, JRUNDO_XXX
, 0, 0);
1038 jrecord_write_cred(jrec
, curthread
, ap
->a_cred
);
1039 jrecord_write_vnode_ref(jrec
, ap
->a_vp
);
1040 jrecord_leaf(jrec
, JLEAF_ATTRNAME
, ap
->a_name
, strlen(ap
->a_name
));
1041 save
= jrecord_push(jrec
, JTYPE_REDO
);
1042 jrecord_write_uio(jrec
, JLEAF_FILEDATA
, ap
->a_uio
);
1043 jrecord_pop(jrec
, save
);
1046 jreclist_done(mp
, &jreclist
, error
);
1051 * Journal vop_ncreate { a_nch, a_vpp, a_cred, a_vap }
1055 journal_ncreate(struct vop_ncreate_args
*ap
)
1057 struct jrecord_list jreclist
;
1058 struct jrecord jreccache
;
1059 struct jrecord
*jrec
;
1064 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
1065 jreclist_init(mp
, &jreclist
, &jreccache
, JTYPE_CREATE
);
1066 error
= vop_journal_operate_ap(&ap
->a_head
);
1068 TAILQ_FOREACH(jrec
, &jreclist
.list
, user_entry
) {
1069 jrecord_write_cred(jrec
, NULL
, ap
->a_cred
);
1070 jrecord_write_path(jrec
, JLEAF_PATH1
, ap
->a_nch
->ncp
);
1072 jrecord_write_vnode_ref(jrec
, *ap
->a_vpp
);
1073 save
= jrecord_push(jrec
, JTYPE_REDO
);
1074 jrecord_write_vattr(jrec
, ap
->a_vap
);
1075 jrecord_pop(jrec
, save
);
1078 jreclist_done(mp
, &jreclist
, error
);
1083 * Journal vop_nmknod { a_nch, a_vpp, a_cred, a_vap }
1087 journal_nmknod(struct vop_nmknod_args
*ap
)
1089 struct jrecord_list jreclist
;
1090 struct jrecord jreccache
;
1091 struct jrecord
*jrec
;
1096 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
1097 jreclist_init(mp
, &jreclist
, &jreccache
, JTYPE_MKNOD
);
1098 error
= vop_journal_operate_ap(&ap
->a_head
);
1100 TAILQ_FOREACH(jrec
, &jreclist
.list
, user_entry
) {
1101 jrecord_write_cred(jrec
, NULL
, ap
->a_cred
);
1102 jrecord_write_path(jrec
, JLEAF_PATH1
, ap
->a_nch
->ncp
);
1103 save
= jrecord_push(jrec
, JTYPE_REDO
);
1104 jrecord_write_vattr(jrec
, ap
->a_vap
);
1105 jrecord_pop(jrec
, save
);
1107 jrecord_write_vnode_ref(jrec
, *ap
->a_vpp
);
1110 jreclist_done(mp
, &jreclist
, error
);
1115 * Journal vop_nlink { a_nch, a_vp, a_cred }
1119 journal_nlink(struct vop_nlink_args
*ap
)
1121 struct jrecord_list jreclist
;
1122 struct jrecord jreccache
;
1123 struct jrecord
*jrec
;
1128 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
1129 jreclist_init(mp
, &jreclist
, &jreccache
, JTYPE_LINK
);
1130 error
= vop_journal_operate_ap(&ap
->a_head
);
1132 TAILQ_FOREACH(jrec
, &jreclist
.list
, user_entry
) {
1133 jrecord_write_cred(jrec
, NULL
, ap
->a_cred
);
1134 jrecord_write_path(jrec
, JLEAF_PATH1
, ap
->a_nch
->ncp
);
1135 /* XXX PATH to VP and inode number */
1136 /* XXX this call may not record the correct path when
1137 * multiple paths are available */
1138 save
= jrecord_push(jrec
, JTYPE_REDO
);
1139 jrecord_write_vnode_link(jrec
, ap
->a_vp
, ap
->a_nch
->ncp
);
1140 jrecord_pop(jrec
, save
);
1143 jreclist_done(mp
, &jreclist
, error
);
1148 * Journal vop_symlink { a_nch, a_vpp, a_cred, a_vap, a_target }
1152 journal_nsymlink(struct vop_nsymlink_args
*ap
)
1154 struct jrecord_list jreclist
;
1155 struct jrecord jreccache
;
1156 struct jrecord
*jrec
;
1161 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
1162 jreclist_init(mp
, &jreclist
, &jreccache
, JTYPE_SYMLINK
);
1163 error
= vop_journal_operate_ap(&ap
->a_head
);
1165 TAILQ_FOREACH(jrec
, &jreclist
.list
, user_entry
) {
1166 jrecord_write_cred(jrec
, NULL
, ap
->a_cred
);
1167 jrecord_write_path(jrec
, JLEAF_PATH1
, ap
->a_nch
->ncp
);
1168 save
= jrecord_push(jrec
, JTYPE_REDO
);
1169 jrecord_leaf(jrec
, JLEAF_SYMLINKDATA
,
1170 ap
->a_target
, strlen(ap
->a_target
));
1171 jrecord_pop(jrec
, save
);
1173 jrecord_write_vnode_ref(jrec
, *ap
->a_vpp
);
1176 jreclist_done(mp
, &jreclist
, error
);
1181 * Journal vop_nwhiteout { a_nch, a_cred, a_flags }
1185 journal_nwhiteout(struct vop_nwhiteout_args
*ap
)
1187 struct jrecord_list jreclist
;
1188 struct jrecord jreccache
;
1189 struct jrecord
*jrec
;
1193 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
1194 jreclist_init(mp
, &jreclist
, &jreccache
, JTYPE_WHITEOUT
);
1195 error
= vop_journal_operate_ap(&ap
->a_head
);
1197 TAILQ_FOREACH(jrec
, &jreclist
.list
, user_entry
) {
1198 jrecord_write_cred(jrec
, NULL
, ap
->a_cred
);
1199 jrecord_write_path(jrec
, JLEAF_PATH1
, ap
->a_nch
->ncp
);
1202 jreclist_done(mp
, &jreclist
, error
);
1207 * Journal vop_nremove { a_nch, a_cred }
1211 journal_nremove(struct vop_nremove_args
*ap
)
1213 struct jrecord_list jreclist
;
1214 struct jrecord jreccache
;
1215 struct jrecord
*jrec
;
1219 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
1220 if (jreclist_init(mp
, &jreclist
, &jreccache
, JTYPE_REMOVE
) &&
1221 ap
->a_nch
->ncp
->nc_vp
1223 jreclist_undo_file(&jreclist
, ap
->a_nch
->ncp
->nc_vp
,
1224 JRUNDO_ALL
|JRUNDO_GETVP
|JRUNDO_CONDLINK
, 0, -1);
1226 error
= vop_journal_operate_ap(&ap
->a_head
);
1228 TAILQ_FOREACH(jrec
, &jreclist
.list
, user_entry
) {
1229 jrecord_write_cred(jrec
, NULL
, ap
->a_cred
);
1230 jrecord_write_path(jrec
, JLEAF_PATH1
, ap
->a_nch
->ncp
);
1233 jreclist_done(mp
, &jreclist
, error
);
1238 * Journal vop_nmkdir { a_nch, a_vpp, a_cred, a_vap }
1242 journal_nmkdir(struct vop_nmkdir_args
*ap
)
1244 struct jrecord_list jreclist
;
1245 struct jrecord jreccache
;
1246 struct jrecord
*jrec
;
1250 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
1251 jreclist_init(mp
, &jreclist
, &jreccache
, JTYPE_MKDIR
);
1252 error
= vop_journal_operate_ap(&ap
->a_head
);
1254 TAILQ_FOREACH(jrec
, &jreclist
.list
, user_entry
) {
1256 if (jo
->flags
& MC_JOURNAL_WANT_AUDIT
) {
1257 jrecord_write_audit(jrec
);
1260 jrecord_write_path(jrec
, JLEAF_PATH1
, ap
->a_nch
->ncp
);
1261 jrecord_write_cred(jrec
, NULL
, ap
->a_cred
);
1262 jrecord_write_vattr(jrec
, ap
->a_vap
);
1263 jrecord_write_path(jrec
, JLEAF_PATH1
, ap
->a_nch
->ncp
);
1265 jrecord_write_vnode_ref(jrec
, *ap
->a_vpp
);
1268 jreclist_done(mp
, &jreclist
, error
);
1273 * Journal vop_nrmdir { a_nch, a_cred }
1277 journal_nrmdir(struct vop_nrmdir_args
*ap
)
1279 struct jrecord_list jreclist
;
1280 struct jrecord jreccache
;
1281 struct jrecord
*jrec
;
1285 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
1286 if (jreclist_init(mp
, &jreclist
, &jreccache
, JTYPE_RMDIR
)) {
1287 jreclist_undo_file(&jreclist
, ap
->a_nch
->ncp
->nc_vp
,
1288 JRUNDO_VATTR
|JRUNDO_GETVP
, 0, 0);
1290 error
= vop_journal_operate_ap(&ap
->a_head
);
1292 TAILQ_FOREACH(jrec
, &jreclist
.list
, user_entry
) {
1293 jrecord_write_cred(jrec
, NULL
, ap
->a_cred
);
1294 jrecord_write_path(jrec
, JLEAF_PATH1
, ap
->a_nch
->ncp
);
1297 jreclist_done(mp
, &jreclist
, error
);
1302 * Journal vop_nrename { a_fnch, a_tnch, a_cred }
1306 journal_nrename(struct vop_nrename_args
*ap
)
1308 struct jrecord_list jreclist
;
1309 struct jrecord jreccache
;
1310 struct jrecord
*jrec
;
1314 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
1315 if (jreclist_init(mp
, &jreclist
, &jreccache
, JTYPE_RENAME
) &&
1316 ap
->a_tnch
->ncp
->nc_vp
1318 jreclist_undo_file(&jreclist
, ap
->a_tnch
->ncp
->nc_vp
,
1319 JRUNDO_ALL
|JRUNDO_GETVP
|JRUNDO_CONDLINK
, 0, -1);
1321 error
= vop_journal_operate_ap(&ap
->a_head
);
1323 TAILQ_FOREACH(jrec
, &jreclist
.list
, user_entry
) {
1324 jrecord_write_cred(jrec
, NULL
, ap
->a_cred
);
1325 jrecord_write_path(jrec
, JLEAF_PATH1
, ap
->a_fnch
->ncp
);
1326 jrecord_write_path(jrec
, JLEAF_PATH2
, ap
->a_tnch
->ncp
);
1329 jreclist_done(mp
, &jreclist
, error
);