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/socket.h>
65 #include <sys/socketvar.h>
67 #include <machine/limits.h>
70 #include <vm/vm_object.h>
71 #include <vm/vm_page.h>
72 #include <vm/vm_pager.h>
73 #include <vm/vnode_pager.h>
75 #include <sys/file2.h>
76 #include <sys/thread2.h>
78 static int journal_attach(struct mount
*mp
);
79 static void journal_detach(struct mount
*mp
);
80 static int journal_install_vfs_journal(struct mount
*mp
, struct file
*fp
,
81 const struct mountctl_install_journal
*info
);
82 static int journal_restart_vfs_journal(struct mount
*mp
, struct file
*fp
,
83 const struct mountctl_restart_journal
*info
);
84 static int journal_remove_vfs_journal(struct mount
*mp
,
85 const struct mountctl_remove_journal
*info
);
86 static int journal_restart(struct mount
*mp
, struct file
*fp
,
87 struct journal
*jo
, int flags
);
88 static int journal_destroy(struct mount
*mp
, struct journal
*jo
, int flags
);
89 static int journal_resync_vfs_journal(struct mount
*mp
, const void *ctl
);
90 static int journal_status_vfs_journal(struct mount
*mp
,
91 const struct mountctl_status_journal
*info
,
92 struct mountctl_journal_ret_status
*rstat
,
93 int buflen
, int *res
);
95 static void jrecord_undo_file(struct jrecord
*jrec
, struct vnode
*vp
,
96 int jrflags
, off_t off
, off_t bytes
);
98 static int journal_setattr(struct vop_setattr_args
*ap
);
99 static int journal_write(struct vop_write_args
*ap
);
100 static int journal_fsync(struct vop_fsync_args
*ap
);
101 static int journal_putpages(struct vop_putpages_args
*ap
);
102 static int journal_setacl(struct vop_setacl_args
*ap
);
103 static int journal_setextattr(struct vop_setextattr_args
*ap
);
104 static int journal_ncreate(struct vop_ncreate_args
*ap
);
105 static int journal_nmknod(struct vop_nmknod_args
*ap
);
106 static int journal_nlink(struct vop_nlink_args
*ap
);
107 static int journal_nsymlink(struct vop_nsymlink_args
*ap
);
108 static int journal_nwhiteout(struct vop_nwhiteout_args
*ap
);
109 static int journal_nremove(struct vop_nremove_args
*ap
);
110 static int journal_nmkdir(struct vop_nmkdir_args
*ap
);
111 static int journal_nrmdir(struct vop_nrmdir_args
*ap
);
112 static int journal_nrename(struct vop_nrename_args
*ap
);
114 #define JRUNDO_SIZE 0x00000001
115 #define JRUNDO_UID 0x00000002
116 #define JRUNDO_GID 0x00000004
117 #define JRUNDO_FSID 0x00000008
118 #define JRUNDO_MODES 0x00000010
119 #define JRUNDO_INUM 0x00000020
120 #define JRUNDO_ATIME 0x00000040
121 #define JRUNDO_MTIME 0x00000080
122 #define JRUNDO_CTIME 0x00000100
123 #define JRUNDO_GEN 0x00000200
124 #define JRUNDO_FLAGS 0x00000400
125 #define JRUNDO_UDEV 0x00000800
126 #define JRUNDO_NLINK 0x00001000
127 #define JRUNDO_FILEDATA 0x00010000
128 #define JRUNDO_GETVP 0x00020000
129 #define JRUNDO_CONDLINK 0x00040000 /* write file data if link count 1 */
130 #define JRUNDO_VATTR (JRUNDO_SIZE|JRUNDO_UID|JRUNDO_GID|JRUNDO_FSID|\
131 JRUNDO_MODES|JRUNDO_INUM|JRUNDO_ATIME|JRUNDO_MTIME|\
132 JRUNDO_CTIME|JRUNDO_GEN|JRUNDO_FLAGS|JRUNDO_UDEV|\
134 #define JRUNDO_ALL (JRUNDO_VATTR|JRUNDO_FILEDATA)
136 static struct vop_ops journal_vnode_vops
= {
137 .vop_default
= vop_journal_operate_ap
,
138 .vop_mountctl
= journal_mountctl
,
139 .vop_setattr
= journal_setattr
,
140 .vop_write
= journal_write
,
141 .vop_fsync
= journal_fsync
,
142 .vop_putpages
= journal_putpages
,
143 .vop_setacl
= journal_setacl
,
144 .vop_setextattr
= journal_setextattr
,
145 .vop_ncreate
= journal_ncreate
,
146 .vop_nmknod
= journal_nmknod
,
147 .vop_nlink
= journal_nlink
,
148 .vop_nsymlink
= journal_nsymlink
,
149 .vop_nwhiteout
= journal_nwhiteout
,
150 .vop_nremove
= journal_nremove
,
151 .vop_nmkdir
= journal_nmkdir
,
152 .vop_nrmdir
= journal_nrmdir
,
153 .vop_nrename
= journal_nrename
157 journal_mountctl(struct vop_mountctl_args
*ap
)
162 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
165 if (mp
->mnt_vn_journal_ops
== NULL
) {
167 case MOUNTCTL_INSTALL_VFS_JOURNAL
:
168 error
= journal_attach(mp
);
169 if (error
== 0 && ap
->a_ctllen
!= sizeof(struct mountctl_install_journal
))
171 if (error
== 0 && ap
->a_fp
== NULL
)
174 error
= journal_install_vfs_journal(mp
, ap
->a_fp
, ap
->a_ctl
);
175 if (TAILQ_EMPTY(&mp
->mnt_jlist
))
178 case MOUNTCTL_RESTART_VFS_JOURNAL
:
179 case MOUNTCTL_REMOVE_VFS_JOURNAL
:
180 case MOUNTCTL_RESYNC_VFS_JOURNAL
:
181 case MOUNTCTL_STATUS_VFS_JOURNAL
:
190 case MOUNTCTL_INSTALL_VFS_JOURNAL
:
191 if (ap
->a_ctllen
!= sizeof(struct mountctl_install_journal
))
193 if (error
== 0 && ap
->a_fp
== NULL
)
196 error
= journal_install_vfs_journal(mp
, ap
->a_fp
, ap
->a_ctl
);
198 case MOUNTCTL_RESTART_VFS_JOURNAL
:
199 if (ap
->a_ctllen
!= sizeof(struct mountctl_restart_journal
))
201 if (error
== 0 && ap
->a_fp
== NULL
)
204 error
= journal_restart_vfs_journal(mp
, ap
->a_fp
, ap
->a_ctl
);
206 case MOUNTCTL_REMOVE_VFS_JOURNAL
:
207 if (ap
->a_ctllen
!= sizeof(struct mountctl_remove_journal
))
210 error
= journal_remove_vfs_journal(mp
, ap
->a_ctl
);
211 if (TAILQ_EMPTY(&mp
->mnt_jlist
))
214 case MOUNTCTL_RESYNC_VFS_JOURNAL
:
215 if (ap
->a_ctllen
!= 0)
217 error
= journal_resync_vfs_journal(mp
, ap
->a_ctl
);
219 case MOUNTCTL_STATUS_VFS_JOURNAL
:
220 if (ap
->a_ctllen
!= sizeof(struct mountctl_status_journal
))
223 error
= journal_status_vfs_journal(mp
, ap
->a_ctl
,
224 ap
->a_buf
, ap
->a_buflen
, ap
->a_res
);
236 * High level mount point setup. When a
239 journal_attach(struct mount
*mp
)
241 KKASSERT(mp
->mnt_jbitmap
== NULL
);
242 vfs_add_vnodeops(mp
, &journal_vnode_vops
, &mp
->mnt_vn_journal_ops
);
243 mp
->mnt_jbitmap
= kmalloc(JREC_STREAMID_JMAX
/8, M_JOURNAL
, M_WAITOK
|M_ZERO
);
244 mp
->mnt_streamid
= JREC_STREAMID_JMIN
;
249 journal_detach(struct mount
*mp
)
251 KKASSERT(mp
->mnt_jbitmap
!= NULL
);
252 if (mp
->mnt_vn_journal_ops
)
253 vfs_rm_vnodeops(mp
, &journal_vnode_vops
, &mp
->mnt_vn_journal_ops
);
254 kfree(mp
->mnt_jbitmap
, M_JOURNAL
);
255 mp
->mnt_jbitmap
= NULL
;
259 * Install a journal on a mount point. Each journal has an associated worker
260 * thread which is responsible for buffering and spooling the data to the
261 * target. A mount point may have multiple journals attached to it. An
262 * initial start record is generated when the journal is associated.
265 journal_install_vfs_journal(struct mount
*mp
, struct file
*fp
,
266 const struct mountctl_install_journal
*info
)
273 jo
= kmalloc(sizeof(struct journal
), M_JOURNAL
, M_WAITOK
|M_ZERO
);
274 bcopy(info
->id
, jo
->id
, sizeof(jo
->id
));
275 jo
->flags
= info
->flags
& ~(MC_JOURNAL_WACTIVE
| MC_JOURNAL_RACTIVE
|
276 MC_JOURNAL_STOP_REQ
);
279 * Memory FIFO size, round to nearest power of 2
281 if (info
->membufsize
) {
282 if (info
->membufsize
< 65536)
284 else if (info
->membufsize
> 128 * 1024 * 1024)
285 size
= 128 * 1024 * 1024;
287 size
= (int)info
->membufsize
;
292 while (jo
->fifo
.size
< size
)
296 * Other parameters. If not specified the starting transaction id
297 * will be the current date.
300 jo
->transid
= info
->transid
;
304 jo
->transid
= ((int64_t)ts
.tv_sec
<< 30) | ts
.tv_nsec
;
310 * Allocate the memory FIFO
312 jo
->fifo
.mask
= jo
->fifo
.size
- 1;
313 jo
->fifo
.membase
= kmalloc(jo
->fifo
.size
, M_JFIFO
, M_WAITOK
|M_ZERO
|M_NULLOK
);
314 if (jo
->fifo
.membase
== NULL
)
318 * Create the worker threads and generate the association record.
321 kfree(jo
, M_JOURNAL
);
324 journal_create_threads(jo
);
325 jrecord_init(jo
, &jrec
, JREC_STREAMID_DISCONT
);
326 jrecord_write(&jrec
, JTYPE_ASSOCIATE
, 0);
327 jrecord_done(&jrec
, 0);
328 TAILQ_INSERT_TAIL(&mp
->mnt_jlist
, jo
, jentry
);
334 * Restart a journal with a new descriptor. The existing reader and writer
335 * threads are terminated and a new descriptor is associated with the
336 * journal. The FIFO rindex is reset to xindex and the threads are then
340 journal_restart_vfs_journal(struct mount
*mp
, struct file
*fp
,
341 const struct mountctl_restart_journal
*info
)
346 TAILQ_FOREACH(jo
, &mp
->mnt_jlist
, jentry
) {
347 if (bcmp(jo
->id
, info
->id
, sizeof(jo
->id
)) == 0)
351 error
= journal_restart(mp
, fp
, jo
, info
->flags
);
358 journal_restart(struct mount
*mp
, struct file
*fp
,
359 struct journal
*jo
, int flags
)
367 * Record the fact that we are doing a restart in the journal.
368 * XXX it isn't safe to do this if the journal is being restarted
369 * because it was locked up and the writer thread has already exited.
371 jrecord_init(jo
, &jrec
, JREC_STREAMID_RESTART
);
372 jrecord_write(&jrec
, JTYPE_DISASSOCIATE
, 0);
373 jrecord_done(&jrec
, 0);
377 * Stop the reader and writer threads and clean up the current
380 kprintf("RESTART WITH FP %p KILLING %p\n", fp
, jo
->fp
);
381 journal_destroy_threads(jo
, flags
);
387 * Associate the new descriptor, reset the FIFO index, and recreate
392 jo
->fifo
.rindex
= jo
->fifo
.xindex
;
393 journal_create_threads(jo
);
399 * Disassociate a journal from a mount point and terminate its worker thread.
400 * A final termination record is written out before the file pointer is
404 journal_remove_vfs_journal(struct mount
*mp
,
405 const struct mountctl_remove_journal
*info
)
410 TAILQ_FOREACH(jo
, &mp
->mnt_jlist
, jentry
) {
411 if (bcmp(jo
->id
, info
->id
, sizeof(jo
->id
)) == 0)
415 error
= journal_destroy(mp
, jo
, info
->flags
);
422 * Remove all journals associated with a mount point. Usually called
423 * by the umount code.
426 journal_remove_all_journals(struct mount
*mp
, int flags
)
430 while ((jo
= TAILQ_FIRST(&mp
->mnt_jlist
)) != NULL
) {
431 journal_destroy(mp
, jo
, flags
);
436 journal_destroy(struct mount
*mp
, struct journal
*jo
, int flags
)
440 TAILQ_REMOVE(&mp
->mnt_jlist
, jo
, jentry
);
442 jrecord_init(jo
, &jrec
, JREC_STREAMID_DISCONT
);
443 jrecord_write(&jrec
, JTYPE_DISASSOCIATE
, 0);
444 jrecord_done(&jrec
, 0);
446 journal_destroy_threads(jo
, flags
);
450 if (jo
->fifo
.membase
)
451 kfree(jo
->fifo
.membase
, M_JFIFO
);
452 kfree(jo
, M_JOURNAL
);
458 journal_resync_vfs_journal(struct mount
*mp
, const void *ctl
)
464 journal_status_vfs_journal(struct mount
*mp
,
465 const struct mountctl_status_journal
*info
,
466 struct mountctl_journal_ret_status
*rstat
,
467 int buflen
, int *res
)
475 TAILQ_FOREACH(jo
, &mp
->mnt_jlist
, jentry
) {
476 if (info
->index
== MC_JOURNAL_INDEX_ID
) {
477 if (bcmp(jo
->id
, info
->id
, sizeof(jo
->id
)) != 0)
479 } else if (info
->index
>= 0) {
480 if (info
->index
< index
)
482 } else if (info
->index
!= MC_JOURNAL_INDEX_ALL
) {
485 if (buflen
< sizeof(*rstat
)) {
487 rstat
[-1].flags
|= MC_JOURNAL_STATUS_MORETOCOME
;
492 bzero(rstat
, sizeof(*rstat
));
493 rstat
->recsize
= sizeof(*rstat
);
494 bcopy(jo
->id
, rstat
->id
, sizeof(jo
->id
));
495 rstat
->index
= index
;
496 rstat
->membufsize
= jo
->fifo
.size
;
497 rstat
->membufused
= jo
->fifo
.windex
- jo
->fifo
.xindex
;
498 rstat
->membufunacked
= jo
->fifo
.rindex
- jo
->fifo
.xindex
;
499 rstat
->bytessent
= jo
->total_acked
;
500 rstat
->fifostalls
= jo
->fifostalls
;
503 *res
+= sizeof(*rstat
);
504 buflen
-= sizeof(*rstat
);
509 /************************************************************************
510 * PARALLEL TRANSACTION SUPPORT ROUTINES *
511 ************************************************************************
513 * JRECLIST_*() - routines which create and iterate over jrecord structures,
514 * because a mount point may have multiple attached journals.
518 * Initialize the passed jrecord_list and create a jrecord for each
519 * journal we need to write to. Unnecessary mallocs are avoided by
520 * using the passed jrecord structure as the first jrecord in the list.
521 * A starting transaction is pushed for each jrecord.
523 * Returns non-zero if any of the journals require undo records.
527 jreclist_init(struct mount
*mp
, struct jrecord_list
*jreclist
,
528 struct jrecord
*jreccache
, int16_t rectype
)
531 struct jrecord
*jrec
;
536 TAILQ_INIT(&jreclist
->list
);
539 * Select the stream ID to use for the transaction. We must select
540 * a stream ID that is not currently in use by some other parallel
543 * Don't bother calculating the next streamid when reassigning
544 * mnt_streamid, since parallel transactions are fairly rare. This
545 * also allows someone observing the raw records to clearly see
546 * when parallel transactions occur.
548 streamid
= mp
->mnt_streamid
;
550 while (mp
->mnt_jbitmap
[streamid
>> 3] & (1 << (streamid
& 7))) {
551 if (++streamid
== JREC_STREAMID_JMAX
)
552 streamid
= JREC_STREAMID_JMIN
;
553 if (++count
== JREC_STREAMID_JMAX
- JREC_STREAMID_JMIN
) {
554 kprintf("jreclist_init: all streamid's in use! sleeping\n");
555 tsleep(jreclist
, 0, "jsidfl", hz
* 10);
559 mp
->mnt_jbitmap
[streamid
>> 3] |= 1 << (streamid
& 7);
560 mp
->mnt_streamid
= streamid
;
561 jreclist
->streamid
= streamid
;
564 * Now initialize a stream on each journal.
568 TAILQ_FOREACH(jo
, &mp
->mnt_jlist
, jentry
) {
572 jrec
= kmalloc(sizeof(*jrec
), M_JOURNAL
, M_WAITOK
);
573 jrecord_init(jo
, jrec
, streamid
);
574 jrec
->user_save
= jrecord_push(jrec
, rectype
);
575 TAILQ_INSERT_TAIL(&jreclist
->list
, jrec
, user_entry
);
576 if (jo
->flags
& MC_JOURNAL_WANT_REVERSABLE
)
584 * Terminate the journaled transactions started by jreclist_init(). If
585 * an error occured, the transaction records will be aborted.
589 jreclist_done(struct mount
*mp
, struct jrecord_list
*jreclist
, int error
)
591 struct jrecord
*jrec
;
595 * Cleanup the jrecord state on each journal.
597 TAILQ_FOREACH(jrec
, &jreclist
->list
, user_entry
) {
598 jrecord_pop(jrec
, jrec
->user_save
);
599 jrecord_done(jrec
, error
);
603 * Free allocated jrec's (the first is always supplied)
606 while ((jrec
= TAILQ_FIRST(&jreclist
->list
)) != NULL
) {
607 TAILQ_REMOVE(&jreclist
->list
, jrec
, user_entry
);
609 kfree(jrec
, M_JOURNAL
);
614 * Clear the streamid so it can be reused.
616 mp
->mnt_jbitmap
[jreclist
->streamid
>> 3] &= ~(1 << (jreclist
->streamid
& 7));
620 * This procedure writes out UNDO records for available reversable
623 * XXX could use improvement. There is no need to re-read the file
628 jreclist_undo_file(struct jrecord_list
*jreclist
, struct vnode
*vp
,
629 int jrflags
, off_t off
, off_t bytes
)
631 struct jrecord
*jrec
;
635 if (jrflags
& JRUNDO_GETVP
)
636 error
= vget(vp
, LK_SHARED
);
638 TAILQ_FOREACH(jrec
, &jreclist
->list
, user_entry
) {
639 if (jrec
->jo
->flags
& MC_JOURNAL_WANT_REVERSABLE
) {
640 jrecord_undo_file(jrec
, vp
, jrflags
, off
, bytes
);
644 if (error
== 0 && jrflags
& JRUNDO_GETVP
)
648 /************************************************************************
649 * LOW LEVEL UNDO SUPPORT ROUTINE *
650 ************************************************************************
652 * This function is used to support UNDO records. It will generate an
653 * appropriate record with the requested portion of the file data. Note
654 * that file data is only recorded if JRUNDO_FILEDATA is passed. If bytes
655 * is -1, it will be set to the size of the file.
658 jrecord_undo_file(struct jrecord
*jrec
, struct vnode
*vp
, int jrflags
,
659 off_t off
, off_t bytes
)
662 void *save1
; /* warning, save pointers do not always remain valid */
667 * Setup. Start the UNDO record, obtain a shared lock on the vnode,
668 * and retrieve attribute info.
670 save1
= jrecord_push(jrec
, JTYPE_UNDO
);
671 error
= VOP_GETATTR(vp
, &attr
);
676 * Generate UNDO records as requested.
678 if (jrflags
& JRUNDO_VATTR
) {
679 save2
= jrecord_push(jrec
, JTYPE_VATTR
);
680 jrecord_leaf(jrec
, JLEAF_VTYPE
, &attr
.va_type
, sizeof(attr
.va_type
));
681 if ((jrflags
& JRUNDO_NLINK
) && attr
.va_nlink
!= VNOVAL
)
682 jrecord_leaf(jrec
, JLEAF_NLINK
, &attr
.va_nlink
, sizeof(attr
.va_nlink
));
683 if ((jrflags
& JRUNDO_SIZE
) && attr
.va_size
!= VNOVAL
)
684 jrecord_leaf(jrec
, JLEAF_SIZE
, &attr
.va_size
, sizeof(attr
.va_size
));
685 if ((jrflags
& JRUNDO_UID
) && attr
.va_uid
!= VNOVAL
)
686 jrecord_leaf(jrec
, JLEAF_UID
, &attr
.va_uid
, sizeof(attr
.va_uid
));
687 if ((jrflags
& JRUNDO_GID
) && attr
.va_gid
!= VNOVAL
)
688 jrecord_leaf(jrec
, JLEAF_GID
, &attr
.va_gid
, sizeof(attr
.va_gid
));
689 if ((jrflags
& JRUNDO_FSID
) && attr
.va_fsid
!= VNOVAL
)
690 jrecord_leaf(jrec
, JLEAF_FSID
, &attr
.va_fsid
, sizeof(attr
.va_fsid
));
691 if ((jrflags
& JRUNDO_MODES
) && attr
.va_mode
!= (mode_t
)VNOVAL
)
692 jrecord_leaf(jrec
, JLEAF_MODES
, &attr
.va_mode
, sizeof(attr
.va_mode
));
693 if ((jrflags
& JRUNDO_INUM
) && attr
.va_fileid
!= VNOVAL
)
694 jrecord_leaf(jrec
, JLEAF_INUM
, &attr
.va_fileid
, sizeof(attr
.va_fileid
));
695 if ((jrflags
& JRUNDO_ATIME
) && attr
.va_atime
.tv_sec
!= VNOVAL
)
696 jrecord_leaf(jrec
, JLEAF_ATIME
, &attr
.va_atime
, sizeof(attr
.va_atime
));
697 if ((jrflags
& JRUNDO_MTIME
) && attr
.va_mtime
.tv_sec
!= VNOVAL
)
698 jrecord_leaf(jrec
, JLEAF_MTIME
, &attr
.va_mtime
, sizeof(attr
.va_mtime
));
699 if ((jrflags
& JRUNDO_CTIME
) && attr
.va_ctime
.tv_sec
!= VNOVAL
)
700 jrecord_leaf(jrec
, JLEAF_CTIME
, &attr
.va_ctime
, sizeof(attr
.va_ctime
));
701 if ((jrflags
& JRUNDO_GEN
) && attr
.va_gen
!= VNOVAL
)
702 jrecord_leaf(jrec
, JLEAF_GEN
, &attr
.va_gen
, sizeof(attr
.va_gen
));
703 if ((jrflags
& JRUNDO_FLAGS
) && attr
.va_flags
!= VNOVAL
)
704 jrecord_leaf(jrec
, JLEAF_FLAGS
, &attr
.va_flags
, sizeof(attr
.va_flags
));
705 if ((jrflags
& JRUNDO_UDEV
) && attr
.va_rmajor
!= VNOVAL
) {
706 udev_t rdev
= makeudev(attr
.va_rmajor
, attr
.va_rminor
);
707 jrecord_leaf(jrec
, JLEAF_UDEV
, &rdev
, sizeof(rdev
));
708 jrecord_leaf(jrec
, JLEAF_UMAJOR
, &attr
.va_rmajor
, sizeof(attr
.va_rmajor
));
709 jrecord_leaf(jrec
, JLEAF_UMINOR
, &attr
.va_rminor
, sizeof(attr
.va_rminor
));
711 jrecord_pop(jrec
, save2
);
715 * Output the file data being overwritten by reading the file and
716 * writing it out to the journal prior to the write operation. We
717 * do not need to write out data past the current file EOF.
719 * XXX support JRUNDO_CONDLINK - do not write out file data for files
720 * with a link count > 1. The undo code needs to locate the inode and
721 * regenerate the hardlink.
723 if ((jrflags
& JRUNDO_FILEDATA
) && attr
.va_type
== VREG
) {
724 if (attr
.va_size
!= VNOVAL
) {
726 bytes
= attr
.va_size
- off
;
727 if (off
+ bytes
> attr
.va_size
)
728 bytes
= attr
.va_size
- off
;
730 jrecord_file_data(jrec
, vp
, off
, bytes
);
735 if ((jrflags
& JRUNDO_FILEDATA
) && attr
.va_type
== VLNK
) {
740 buf
= kmalloc(PATH_MAX
, M_JOURNAL
, M_WAITOK
);
742 aiov
.iov_len
= PATH_MAX
;
743 auio
.uio_iov
= &aiov
;
746 auio
.uio_rw
= UIO_READ
;
747 auio
.uio_segflg
= UIO_SYSSPACE
;
748 auio
.uio_td
= curthread
;
749 auio
.uio_resid
= PATH_MAX
;
750 error
= VOP_READLINK(vp
, &auio
, proc0
.p_ucred
);
752 jrecord_leaf(jrec
, JLEAF_SYMLINKDATA
, buf
,
753 PATH_MAX
- auio
.uio_resid
);
755 kfree(buf
, M_JOURNAL
);
759 jrecord_leaf(jrec
, JLEAF_ERROR
, &error
, sizeof(error
));
760 jrecord_pop(jrec
, save1
);
763 /************************************************************************
765 ************************************************************************
767 * These are function shims replacing the normal filesystem ops. We become
768 * responsible for calling the underlying filesystem ops. We have the choice
769 * of executing the underlying op first and then generating the journal entry,
770 * or starting the journal entry, executing the underlying op, and then
771 * either completing or aborting it.
773 * The journal is supposed to be a high-level entity, which generally means
774 * identifying files by name rather then by inode. Supplying both allows
775 * the journal to be used both for inode-number-compatible 'mirrors' and
776 * for simple filesystem replication.
778 * Writes are particularly difficult to deal with because a single write may
779 * represent a hundred megabyte buffer or more, and both writes and truncations
780 * require the 'old' data to be written out as well as the new data if the
781 * log is reversable. Other issues:
783 * - How to deal with operations on unlinked files (no path available),
784 * but which may still be filesystem visible due to hard links.
786 * - How to deal with modifications made via a memory map.
788 * - Future cache coherency support will require cache coherency API calls
789 * both prior to and after the call to the underlying VFS.
791 * ALSO NOTE: We do not have to shim compatibility VOPs like MKDIR which have
792 * new VFS equivalents (NMKDIR).
796 * Journal vop_setattr { a_vp, a_vap, a_cred }
800 journal_setattr(struct vop_setattr_args
*ap
)
802 struct jrecord_list jreclist
;
803 struct jrecord jreccache
;
804 struct jrecord
*jrec
;
809 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
810 if (jreclist_init(mp
, &jreclist
, &jreccache
, JTYPE_SETATTR
)) {
811 jreclist_undo_file(&jreclist
, ap
->a_vp
, JRUNDO_VATTR
, 0, 0);
813 error
= vop_journal_operate_ap(&ap
->a_head
);
815 TAILQ_FOREACH(jrec
, &jreclist
.list
, user_entry
) {
816 jrecord_write_cred(jrec
, curthread
, ap
->a_cred
);
817 jrecord_write_vnode_ref(jrec
, ap
->a_vp
);
818 save
= jrecord_push(jrec
, JTYPE_REDO
);
819 jrecord_write_vattr(jrec
, ap
->a_vap
);
820 jrecord_pop(jrec
, save
);
823 jreclist_done(mp
, &jreclist
, error
);
828 * Journal vop_write { a_vp, a_uio, a_ioflag, a_cred }
832 journal_write(struct vop_write_args
*ap
)
834 struct jrecord_list jreclist
;
835 struct jrecord jreccache
;
836 struct jrecord
*jrec
;
839 struct iovec uio_one_iovec
;
844 * Special synchronizing writes for VM backing store do not supply any
847 if (ap
->a_uio
->uio_segflg
== UIO_NOCOPY
) {
848 error
= vop_journal_operate_ap(&ap
->a_head
);
853 * This is really nasty. UIO's don't retain sufficient information to
854 * be reusable once they've gone through the VOP chain. The iovecs get
855 * cleared, so we have to copy the UIO.
857 * XXX fix the UIO code to not destroy iov's during a scan so we can
858 * reuse the uio over and over again.
860 * XXX UNDO code needs to journal the old data prior to the write.
862 uio_copy
= *ap
->a_uio
;
863 if (uio_copy
.uio_iovcnt
== 1) {
864 uio_one_iovec
= ap
->a_uio
->uio_iov
[0];
865 uio_copy
.uio_iov
= &uio_one_iovec
;
867 uio_copy
.uio_iov
= kmalloc(uio_copy
.uio_iovcnt
* sizeof(struct iovec
),
868 M_JOURNAL
, M_WAITOK
);
869 bcopy(ap
->a_uio
->uio_iov
, uio_copy
.uio_iov
,
870 uio_copy
.uio_iovcnt
* sizeof(struct iovec
));
874 * Write out undo data. Note that uio_offset is incorrect if
875 * IO_APPEND is set, but fortunately we have no undo file data to
876 * write out in that case.
878 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
879 if (jreclist_init(mp
, &jreclist
, &jreccache
, JTYPE_WRITE
)) {
880 if (ap
->a_ioflag
& IO_APPEND
) {
881 jreclist_undo_file(&jreclist
, ap
->a_vp
, JRUNDO_SIZE
|JRUNDO_MTIME
, 0, 0);
883 jreclist_undo_file(&jreclist
, ap
->a_vp
,
884 JRUNDO_FILEDATA
|JRUNDO_SIZE
|JRUNDO_MTIME
,
885 uio_copy
.uio_offset
, uio_copy
.uio_resid
);
888 error
= vop_journal_operate_ap(&ap
->a_head
);
891 * XXX bad hack to figure out the offset for O_APPEND writes (note:
892 * uio field state after the VFS operation).
894 uio_copy
.uio_offset
= ap
->a_uio
->uio_offset
-
895 (uio_copy
.uio_resid
- ap
->a_uio
->uio_resid
);
898 * Output the write data to the journal.
901 TAILQ_FOREACH(jrec
, &jreclist
.list
, user_entry
) {
902 jrecord_write_cred(jrec
, NULL
, ap
->a_cred
);
903 jrecord_write_vnode_ref(jrec
, ap
->a_vp
);
904 save
= jrecord_push(jrec
, JTYPE_REDO
);
905 jrecord_write_uio(jrec
, JLEAF_FILEDATA
, &uio_copy
);
906 jrecord_pop(jrec
, save
);
909 jreclist_done(mp
, &jreclist
, error
);
911 if (uio_copy
.uio_iov
!= &uio_one_iovec
)
912 kfree(uio_copy
.uio_iov
, M_JOURNAL
);
917 * Journal vop_fsync { a_vp, a_waitfor }
921 journal_fsync(struct vop_fsync_args
*ap
)
929 error
= vop_journal_operate_ap(&ap
->a_head
);
931 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
933 TAILQ_FOREACH(jo
, &mp
->mnt_jlist
, jentry
) {
934 /* XXX synchronize pending journal records */
942 * Journal vop_putpages { a_vp, a_m, a_count, a_sync, a_rtvals, a_offset }
944 * note: a_count is in bytes.
948 journal_putpages(struct vop_putpages_args
*ap
)
950 struct jrecord_list jreclist
;
951 struct jrecord jreccache
;
952 struct jrecord
*jrec
;
957 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
958 if (jreclist_init(mp
, &jreclist
, &jreccache
, JTYPE_PUTPAGES
) &&
961 jreclist_undo_file(&jreclist
, ap
->a_vp
,
962 JRUNDO_FILEDATA
|JRUNDO_SIZE
|JRUNDO_MTIME
,
963 ap
->a_offset
, btoc(ap
->a_count
));
965 error
= vop_journal_operate_ap(&ap
->a_head
);
966 if (error
== 0 && ap
->a_count
> 0) {
967 TAILQ_FOREACH(jrec
, &jreclist
.list
, user_entry
) {
968 jrecord_write_vnode_ref(jrec
, ap
->a_vp
);
969 save
= jrecord_push(jrec
, JTYPE_REDO
);
970 jrecord_write_pagelist(jrec
, JLEAF_FILEDATA
, ap
->a_m
, ap
->a_rtvals
,
971 btoc(ap
->a_count
), ap
->a_offset
);
972 jrecord_pop(jrec
, save
);
975 jreclist_done(mp
, &jreclist
, error
);
980 * Journal vop_setacl { a_vp, a_type, a_aclp, a_cred }
984 journal_setacl(struct vop_setacl_args
*ap
)
986 struct jrecord_list jreclist
;
987 struct jrecord jreccache
;
988 struct jrecord
*jrec
;
992 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
993 jreclist_init(mp
, &jreclist
, &jreccache
, JTYPE_SETACL
);
994 error
= vop_journal_operate_ap(&ap
->a_head
);
996 TAILQ_FOREACH(jrec
, &jreclist
.list
, user_entry
) {
998 if ((jo
->flags
& MC_JOURNAL_WANT_REVERSABLE
))
999 jrecord_undo_file(jrec
, ap
->a_vp
, JRUNDO_XXX
, 0, 0);
1001 jrecord_write_cred(jrec
, curthread
, ap
->a_cred
);
1002 jrecord_write_vnode_ref(jrec
, ap
->a_vp
);
1004 save
= jrecord_push(jrec
, JTYPE_REDO
);
1005 /* XXX type, aclp */
1006 jrecord_pop(jrec
, save
);
1010 jreclist_done(mp
, &jreclist
, error
);
1015 * Journal vop_setextattr { a_vp, a_name, a_uio, a_cred }
1019 journal_setextattr(struct vop_setextattr_args
*ap
)
1021 struct jrecord_list jreclist
;
1022 struct jrecord jreccache
;
1023 struct jrecord
*jrec
;
1028 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
1029 jreclist_init(mp
, &jreclist
, &jreccache
, JTYPE_SETEXTATTR
);
1030 error
= vop_journal_operate_ap(&ap
->a_head
);
1032 TAILQ_FOREACH(jrec
, &jreclist
.list
, user_entry
) {
1034 if ((jo
->flags
& MC_JOURNAL_WANT_REVERSABLE
))
1035 jrecord_undo_file(jrec
, ap
->a_vp
, JRUNDO_XXX
, 0, 0);
1037 jrecord_write_cred(jrec
, curthread
, ap
->a_cred
);
1038 jrecord_write_vnode_ref(jrec
, ap
->a_vp
);
1039 jrecord_leaf(jrec
, JLEAF_ATTRNAME
, ap
->a_attrname
,
1040 strlen(ap
->a_attrname
));
1041 save
= jrecord_push(jrec
, JTYPE_REDO
);
1042 jrecord_write_uio(jrec
, JLEAF_FILEDATA
, ap
->a_uio
);
1043 jrecord_pop(jrec
, save
);
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
);