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 1.30 2006/09/05 03:48:12 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
157 static MALLOC_DEFINE(M_JOURNAL, "journal", "Journaling structures");
158 static MALLOC_DEFINE(M_JFIFO, "journal-fifo", "Journal FIFO");
161 journal_mountctl(struct vop_mountctl_args *ap)
166 mp = ap->a_head.a_ops->head.vv_mount;
169 if (mp->mnt_vn_journal_ops == NULL) {
171 case MOUNTCTL_INSTALL_VFS_JOURNAL:
172 error = journal_attach(mp);
173 if (error == 0 && ap->a_ctllen != sizeof(struct mountctl_install_journal))
175 if (error == 0 && ap->a_fp == NULL)
178 error = journal_install_vfs_journal(mp, ap->a_fp, ap->a_ctl);
179 if (TAILQ_EMPTY(&mp->mnt_jlist))
182 case MOUNTCTL_RESTART_VFS_JOURNAL:
183 case MOUNTCTL_REMOVE_VFS_JOURNAL:
184 case MOUNTCTL_RESYNC_VFS_JOURNAL:
185 case MOUNTCTL_STATUS_VFS_JOURNAL:
194 case MOUNTCTL_INSTALL_VFS_JOURNAL:
195 if (ap->a_ctllen != sizeof(struct mountctl_install_journal))
197 if (error == 0 && ap->a_fp == NULL)
200 error = journal_install_vfs_journal(mp, ap->a_fp, ap->a_ctl);
202 case MOUNTCTL_RESTART_VFS_JOURNAL:
203 if (ap->a_ctllen != sizeof(struct mountctl_restart_journal))
205 if (error == 0 && ap->a_fp == NULL)
208 error = journal_restart_vfs_journal(mp, ap->a_fp, ap->a_ctl);
210 case MOUNTCTL_REMOVE_VFS_JOURNAL:
211 if (ap->a_ctllen != sizeof(struct mountctl_remove_journal))
214 error = journal_remove_vfs_journal(mp, ap->a_ctl);
215 if (TAILQ_EMPTY(&mp->mnt_jlist))
218 case MOUNTCTL_RESYNC_VFS_JOURNAL:
219 if (ap->a_ctllen != 0)
221 error = journal_resync_vfs_journal(mp, ap->a_ctl);
223 case MOUNTCTL_STATUS_VFS_JOURNAL:
224 if (ap->a_ctllen != sizeof(struct mountctl_status_journal))
227 error = journal_status_vfs_journal(mp, ap->a_ctl,
228 ap->a_buf, ap->a_buflen, ap->a_res);
240 * High level mount point setup. When a
243 journal_attach(struct mount *mp)
245 KKASSERT(mp->mnt_jbitmap == NULL);
246 vfs_add_vnodeops(mp, &journal_vnode_vops, &mp->mnt_vn_journal_ops);
247 mp->mnt_jbitmap = kmalloc(JREC_STREAMID_JMAX/8, M_JOURNAL, M_WAITOK|M_ZERO);
248 mp->mnt_streamid = JREC_STREAMID_JMIN;
253 journal_detach(struct mount *mp)
255 KKASSERT(mp->mnt_jbitmap != NULL);
256 if (mp->mnt_vn_journal_ops)
257 vfs_rm_vnodeops(mp, &journal_vnode_vops, &mp->mnt_vn_journal_ops);
258 kfree(mp->mnt_jbitmap, M_JOURNAL);
259 mp->mnt_jbitmap = NULL;
263 * Install a journal on a mount point. Each journal has an associated worker
264 * thread which is responsible for buffering and spooling the data to the
265 * target. A mount point may have multiple journals attached to it. An
266 * initial start record is generated when the journal is associated.
269 journal_install_vfs_journal(struct mount *mp, struct file *fp,
270 const struct mountctl_install_journal *info)
277 jo = kmalloc(sizeof(struct journal), M_JOURNAL, M_WAITOK|M_ZERO);
278 bcopy(info->id, jo->id, sizeof(jo->id));
279 jo->flags = info->flags & ~(MC_JOURNAL_WACTIVE | MC_JOURNAL_RACTIVE |
280 MC_JOURNAL_STOP_REQ);
283 * Memory FIFO size, round to nearest power of 2
285 if (info->membufsize) {
286 if (info->membufsize < 65536)
288 else if (info->membufsize > 128 * 1024 * 1024)
289 size = 128 * 1024 * 1024;
291 size = (int)info->membufsize;
296 while (jo->fifo.size < size)
300 * Other parameters. If not specified the starting transaction id
301 * will be the current date.
304 jo->transid = info->transid;
308 jo->transid = ((int64_t)ts.tv_sec << 30) | ts.tv_nsec;
314 * Allocate the memory FIFO
316 jo->fifo.mask = jo->fifo.size - 1;
317 jo->fifo.membase = kmalloc(jo->fifo.size, M_JFIFO, M_WAITOK|M_ZERO|M_NULLOK);
318 if (jo->fifo.membase == NULL)
322 * Create the worker threads and generate the association record.
325 kfree(jo, M_JOURNAL);
328 journal_create_threads(jo);
329 jrecord_init(jo, &jrec, JREC_STREAMID_DISCONT);
330 jrecord_write(&jrec, JTYPE_ASSOCIATE, 0);
331 jrecord_done(&jrec, 0);
332 TAILQ_INSERT_TAIL(&mp->mnt_jlist, jo, jentry);
338 * Restart a journal with a new descriptor. The existing reader and writer
339 * threads are terminated and a new descriptor is associated with the
340 * journal. The FIFO rindex is reset to xindex and the threads are then
344 journal_restart_vfs_journal(struct mount *mp, struct file *fp,
345 const struct mountctl_restart_journal *info)
350 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
351 if (bcmp(jo->id, info->id, sizeof(jo->id)) == 0)
355 error = journal_restart(mp, fp, jo, info->flags);
362 journal_restart(struct mount *mp, struct file *fp,
363 struct journal *jo, int flags)
371 * Record the fact that we are doing a restart in the journal.
372 * XXX it isn't safe to do this if the journal is being restarted
373 * because it was locked up and the writer thread has already exited.
375 jrecord_init(jo, &jrec, JREC_STREAMID_RESTART);
376 jrecord_write(&jrec, JTYPE_DISASSOCIATE, 0);
377 jrecord_done(&jrec, 0);
381 * Stop the reader and writer threads and clean up the current
384 printf("RESTART WITH FP %p KILLING %p\n", fp, jo->fp);
385 journal_destroy_threads(jo, flags);
391 * Associate the new descriptor, reset the FIFO index, and recreate
396 jo->fifo.rindex = jo->fifo.xindex;
397 journal_create_threads(jo);
403 * Disassociate a journal from a mount point and terminate its worker thread.
404 * A final termination record is written out before the file pointer is
408 journal_remove_vfs_journal(struct mount *mp,
409 const struct mountctl_remove_journal *info)
414 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
415 if (bcmp(jo->id, info->id, sizeof(jo->id)) == 0)
419 error = journal_destroy(mp, jo, info->flags);
426 * Remove all journals associated with a mount point. Usually called
427 * by the umount code.
430 journal_remove_all_journals(struct mount *mp, int flags)
434 while ((jo = TAILQ_FIRST(&mp->mnt_jlist)) != NULL) {
435 journal_destroy(mp, jo, flags);
440 journal_destroy(struct mount *mp, struct journal *jo, int flags)
444 TAILQ_REMOVE(&mp->mnt_jlist, jo, jentry);
446 jrecord_init(jo, &jrec, JREC_STREAMID_DISCONT);
447 jrecord_write(&jrec, JTYPE_DISASSOCIATE, 0);
448 jrecord_done(&jrec, 0);
450 journal_destroy_threads(jo, flags);
454 if (jo->fifo.membase)
455 kfree(jo->fifo.membase, M_JFIFO);
456 kfree(jo, M_JOURNAL);
462 journal_resync_vfs_journal(struct mount *mp, const void *ctl)
468 journal_status_vfs_journal(struct mount *mp,
469 const struct mountctl_status_journal *info,
470 struct mountctl_journal_ret_status *rstat,
471 int buflen, int *res)
479 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
480 if (info->index == MC_JOURNAL_INDEX_ID) {
481 if (bcmp(jo->id, info->id, sizeof(jo->id)) != 0)
483 } else if (info->index >= 0) {
484 if (info->index < index)
486 } else if (info->index != MC_JOURNAL_INDEX_ALL) {
489 if (buflen < sizeof(*rstat)) {
491 rstat[-1].flags |= MC_JOURNAL_STATUS_MORETOCOME;
496 bzero(rstat, sizeof(*rstat));
497 rstat->recsize = sizeof(*rstat);
498 bcopy(jo->id, rstat->id, sizeof(jo->id));
499 rstat->index = index;
500 rstat->membufsize = jo->fifo.size;
501 rstat->membufused = jo->fifo.windex - jo->fifo.xindex;
502 rstat->membufunacked = jo->fifo.rindex - jo->fifo.xindex;
503 rstat->bytessent = jo->total_acked;
504 rstat->fifostalls = jo->fifostalls;
507 *res += sizeof(*rstat);
508 buflen -= sizeof(*rstat);
513 /************************************************************************
514 * PARALLEL TRANSACTION SUPPORT ROUTINES *
515 ************************************************************************
517 * JRECLIST_*() - routines which create and iterate over jrecord structures,
518 * because a mount point may have multiple attached journals.
522 * Initialize the passed jrecord_list and create a jrecord for each
523 * journal we need to write to. Unnecessary mallocs are avoided by
524 * using the passed jrecord structure as the first jrecord in the list.
525 * A starting transaction is pushed for each jrecord.
527 * Returns non-zero if any of the journals require undo records.
531 jreclist_init(struct mount *mp, struct jrecord_list *jreclist,
532 struct jrecord *jreccache, int16_t rectype)
535 struct jrecord *jrec;
540 TAILQ_INIT(&jreclist->list);
543 * Select the stream ID to use for the transaction. We must select
544 * a stream ID that is not currently in use by some other parallel
547 * Don't bother calculating the next streamid when reassigning
548 * mnt_streamid, since parallel transactions are fairly rare. This
549 * also allows someone observing the raw records to clearly see
550 * when parallel transactions occur.
552 streamid = mp->mnt_streamid;
554 while (mp->mnt_jbitmap[streamid >> 3] & (1 << (streamid & 7))) {
555 if (++streamid == JREC_STREAMID_JMAX)
556 streamid = JREC_STREAMID_JMIN;
557 if (++count == JREC_STREAMID_JMAX - JREC_STREAMID_JMIN) {
558 printf("jreclist_init: all streamid's in use! sleeping\n");
559 tsleep(jreclist, 0, "jsidfl", hz * 10);
563 mp->mnt_jbitmap[streamid >> 3] |= 1 << (streamid & 7);
564 mp->mnt_streamid = streamid;
565 jreclist->streamid = streamid;
568 * Now initialize a stream on each journal.
572 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
576 jrec = kmalloc(sizeof(*jrec), M_JOURNAL, M_WAITOK);
577 jrecord_init(jo, jrec, streamid);
578 jrec->user_save = jrecord_push(jrec, rectype);
579 TAILQ_INSERT_TAIL(&jreclist->list, jrec, user_entry);
580 if (jo->flags & MC_JOURNAL_WANT_REVERSABLE)
588 * Terminate the journaled transactions started by jreclist_init(). If
589 * an error occured, the transaction records will be aborted.
593 jreclist_done(struct mount *mp, struct jrecord_list *jreclist, int error)
595 struct jrecord *jrec;
599 * Cleanup the jrecord state on each journal.
601 TAILQ_FOREACH(jrec, &jreclist->list, user_entry) {
602 jrecord_pop(jrec, jrec->user_save);
603 jrecord_done(jrec, error);
607 * Free allocated jrec's (the first is always supplied)
610 while ((jrec = TAILQ_FIRST(&jreclist->list)) != NULL) {
611 TAILQ_REMOVE(&jreclist->list, jrec, user_entry);
613 kfree(jrec, M_JOURNAL);
618 * Clear the streamid so it can be reused.
620 mp->mnt_jbitmap[jreclist->streamid >> 3] &= ~(1 << (jreclist->streamid & 7));
624 * This procedure writes out UNDO records for available reversable
627 * XXX could use improvement. There is no need to re-read the file
632 jreclist_undo_file(struct jrecord_list *jreclist, struct vnode *vp,
633 int jrflags, off_t off, off_t bytes)
635 struct jrecord *jrec;
639 if (jrflags & JRUNDO_GETVP)
640 error = vget(vp, LK_SHARED);
642 TAILQ_FOREACH(jrec, &jreclist->list, user_entry) {
643 if (jrec->jo->flags & MC_JOURNAL_WANT_REVERSABLE) {
644 jrecord_undo_file(jrec, vp, jrflags, off, bytes);
648 if (error == 0 && jrflags & JRUNDO_GETVP)
652 /************************************************************************
653 * LOW LEVEL UNDO SUPPORT ROUTINE *
654 ************************************************************************
656 * This function is used to support UNDO records. It will generate an
657 * appropriate record with the requested portion of the file data. Note
658 * that file data is only recorded if JRUNDO_FILEDATA is passed. If bytes
659 * is -1, it will be set to the size of the file.
662 jrecord_undo_file(struct jrecord *jrec, struct vnode *vp, int jrflags,
663 off_t off, off_t bytes)
666 void *save1; /* warning, save pointers do not always remain valid */
671 * Setup. Start the UNDO record, obtain a shared lock on the vnode,
672 * and retrieve attribute info.
674 save1 = jrecord_push(jrec, JTYPE_UNDO);
675 error = VOP_GETATTR(vp, &attr);
680 * Generate UNDO records as requested.
682 if (jrflags & JRUNDO_VATTR) {
683 save2 = jrecord_push(jrec, JTYPE_VATTR);
684 jrecord_leaf(jrec, JLEAF_VTYPE, &attr.va_type, sizeof(attr.va_type));
685 if ((jrflags & JRUNDO_NLINK) && attr.va_nlink != VNOVAL)
686 jrecord_leaf(jrec, JLEAF_NLINK, &attr.va_nlink, sizeof(attr.va_nlink));
687 if ((jrflags & JRUNDO_SIZE) && attr.va_size != VNOVAL)
688 jrecord_leaf(jrec, JLEAF_SIZE, &attr.va_size, sizeof(attr.va_size));
689 if ((jrflags & JRUNDO_UID) && attr.va_uid != VNOVAL)
690 jrecord_leaf(jrec, JLEAF_UID, &attr.va_uid, sizeof(attr.va_uid));
691 if ((jrflags & JRUNDO_GID) && attr.va_gid != VNOVAL)
692 jrecord_leaf(jrec, JLEAF_GID, &attr.va_gid, sizeof(attr.va_gid));
693 if ((jrflags & JRUNDO_FSID) && attr.va_fsid != VNOVAL)
694 jrecord_leaf(jrec, JLEAF_FSID, &attr.va_fsid, sizeof(attr.va_fsid));
695 if ((jrflags & JRUNDO_MODES) && attr.va_mode != (mode_t)VNOVAL)
696 jrecord_leaf(jrec, JLEAF_MODES, &attr.va_mode, sizeof(attr.va_mode));
697 if ((jrflags & JRUNDO_INUM) && attr.va_fileid != VNOVAL)
698 jrecord_leaf(jrec, JLEAF_INUM, &attr.va_fileid, sizeof(attr.va_fileid));
699 if ((jrflags & JRUNDO_ATIME) && attr.va_atime.tv_sec != VNOVAL)
700 jrecord_leaf(jrec, JLEAF_ATIME, &attr.va_atime, sizeof(attr.va_atime));
701 if ((jrflags & JRUNDO_MTIME) && attr.va_mtime.tv_sec != VNOVAL)
702 jrecord_leaf(jrec, JLEAF_MTIME, &attr.va_mtime, sizeof(attr.va_mtime));
703 if ((jrflags & JRUNDO_CTIME) && attr.va_ctime.tv_sec != VNOVAL)
704 jrecord_leaf(jrec, JLEAF_CTIME, &attr.va_ctime, sizeof(attr.va_ctime));
705 if ((jrflags & JRUNDO_GEN) && attr.va_gen != VNOVAL)
706 jrecord_leaf(jrec, JLEAF_GEN, &attr.va_gen, sizeof(attr.va_gen));
707 if ((jrflags & JRUNDO_FLAGS) && attr.va_flags != VNOVAL)
708 jrecord_leaf(jrec, JLEAF_FLAGS, &attr.va_flags, sizeof(attr.va_flags));
709 if ((jrflags & JRUNDO_UDEV) && attr.va_rdev != VNOVAL)
710 jrecord_leaf(jrec, JLEAF_UDEV, &attr.va_rdev, sizeof(attr.va_rdev));
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_settattr { a_vp, a_vap, a_cred, a_td }
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 * This is really nasty. UIO's don't retain sufficient information to
845 * be reusable once they've gone through the VOP chain. The iovecs get
846 * cleared, so we have to copy the UIO.
848 * XXX fix the UIO code to not destroy iov's during a scan so we can
849 * reuse the uio over and over again.
851 * XXX UNDO code needs to journal the old data prior to the write.
853 uio_copy = *ap->a_uio;
854 if (uio_copy.uio_iovcnt == 1) {
855 uio_one_iovec = ap->a_uio->uio_iov[0];
856 uio_copy.uio_iov = &uio_one_iovec;
858 uio_copy.uio_iov = kmalloc(uio_copy.uio_iovcnt * sizeof(struct iovec),
859 M_JOURNAL, M_WAITOK);
860 bcopy(ap->a_uio->uio_iov, uio_copy.uio_iov,
861 uio_copy.uio_iovcnt * sizeof(struct iovec));
865 * Write out undo data. Note that uio_offset is incorrect if
866 * IO_APPEND is set, but fortunately we have no undo file data to
867 * write out in that case.
869 mp = ap->a_head.a_ops->head.vv_mount;
870 if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_WRITE)) {
871 if (ap->a_ioflag & IO_APPEND) {
872 jreclist_undo_file(&jreclist, ap->a_vp, JRUNDO_SIZE|JRUNDO_MTIME, 0, 0);
874 jreclist_undo_file(&jreclist, ap->a_vp,
875 JRUNDO_FILEDATA|JRUNDO_SIZE|JRUNDO_MTIME,
876 uio_copy.uio_offset, uio_copy.uio_resid);
879 error = vop_journal_operate_ap(&ap->a_head);
882 * XXX bad hack to figure out the offset for O_APPEND writes (note:
883 * uio field state after the VFS operation).
885 uio_copy.uio_offset = ap->a_uio->uio_offset -
886 (uio_copy.uio_resid - ap->a_uio->uio_resid);
889 * Output the write data to the journal.
892 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
893 jrecord_write_cred(jrec, NULL, ap->a_cred);
894 jrecord_write_vnode_ref(jrec, ap->a_vp);
895 save = jrecord_push(jrec, JTYPE_REDO);
896 jrecord_write_uio(jrec, JLEAF_FILEDATA, &uio_copy);
897 jrecord_pop(jrec, save);
900 jreclist_done(mp, &jreclist, error);
902 if (uio_copy.uio_iov != &uio_one_iovec)
903 kfree(uio_copy.uio_iov, M_JOURNAL);
908 * Journal vop_fsync { a_vp, a_waitfor, a_td }
912 journal_fsync(struct vop_fsync_args *ap)
920 error = vop_journal_operate_ap(&ap->a_head);
922 mp = ap->a_head.a_ops->head.vv_mount;
924 TAILQ_FOREACH(jo, &mp->mnt_jlist, jentry) {
925 /* XXX synchronize pending journal records */
933 * Journal vop_putpages { a_vp, a_m, a_count, a_sync, a_rtvals, a_offset }
935 * note: a_count is in bytes.
939 journal_putpages(struct vop_putpages_args *ap)
941 struct jrecord_list jreclist;
942 struct jrecord jreccache;
943 struct jrecord *jrec;
948 mp = ap->a_head.a_ops->head.vv_mount;
949 if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_PUTPAGES) &&
952 jreclist_undo_file(&jreclist, ap->a_vp,
953 JRUNDO_FILEDATA|JRUNDO_SIZE|JRUNDO_MTIME,
954 ap->a_offset, btoc(ap->a_count));
956 error = vop_journal_operate_ap(&ap->a_head);
957 if (error == 0 && ap->a_count > 0) {
958 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
959 jrecord_write_vnode_ref(jrec, ap->a_vp);
960 save = jrecord_push(jrec, JTYPE_REDO);
961 jrecord_write_pagelist(jrec, JLEAF_FILEDATA, ap->a_m, ap->a_rtvals,
962 btoc(ap->a_count), ap->a_offset);
963 jrecord_pop(jrec, save);
966 jreclist_done(mp, &jreclist, error);
971 * Journal vop_setacl { a_vp, a_type, a_aclp, a_cred, a_td }
975 journal_setacl(struct vop_setacl_args *ap)
977 struct jrecord_list jreclist;
978 struct jrecord jreccache;
979 struct jrecord *jrec;
983 mp = ap->a_head.a_ops->head.vv_mount;
984 jreclist_init(mp, &jreclist, &jreccache, JTYPE_SETACL);
985 error = vop_journal_operate_ap(&ap->a_head);
987 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
989 if ((jo->flags & MC_JOURNAL_WANT_REVERSABLE))
990 jrecord_undo_file(jrec, ap->a_vp, JRUNDO_XXX, 0, 0);
992 jrecord_write_cred(jrec, curthread, ap->a_cred);
993 jrecord_write_vnode_ref(jrec, ap->a_vp);
995 save = jrecord_push(jrec, JTYPE_REDO);
997 jrecord_pop(jrec, save);
1001 jreclist_done(mp, &jreclist, error);
1006 * Journal vop_setextattr { a_vp, a_name, a_uio, a_cred, a_td }
1010 journal_setextattr(struct vop_setextattr_args *ap)
1012 struct jrecord_list jreclist;
1013 struct jrecord jreccache;
1014 struct jrecord *jrec;
1019 mp = ap->a_head.a_ops->head.vv_mount;
1020 jreclist_init(mp, &jreclist, &jreccache, JTYPE_SETEXTATTR);
1021 error = vop_journal_operate_ap(&ap->a_head);
1023 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
1025 if ((jo->flags & MC_JOURNAL_WANT_REVERSABLE))
1026 jrecord_undo_file(jrec, ap->a_vp, JRUNDO_XXX, 0, 0);
1028 jrecord_write_cred(jrec, curthread, ap->a_cred);
1029 jrecord_write_vnode_ref(jrec, ap->a_vp);
1030 jrecord_leaf(jrec, JLEAF_ATTRNAME, ap->a_name, strlen(ap->a_name));
1031 save = jrecord_push(jrec, JTYPE_REDO);
1032 jrecord_write_uio(jrec, JLEAF_FILEDATA, ap->a_uio);
1033 jrecord_pop(jrec, save);
1036 jreclist_done(mp, &jreclist, error);
1041 * Journal vop_ncreate { a_ncp, a_vpp, a_cred, a_vap }
1045 journal_ncreate(struct vop_ncreate_args *ap)
1047 struct jrecord_list jreclist;
1048 struct jrecord jreccache;
1049 struct jrecord *jrec;
1054 mp = ap->a_head.a_ops->head.vv_mount;
1055 jreclist_init(mp, &jreclist, &jreccache, JTYPE_CREATE);
1056 error = vop_journal_operate_ap(&ap->a_head);
1058 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
1059 jrecord_write_cred(jrec, NULL, ap->a_cred);
1060 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_ncp);
1062 jrecord_write_vnode_ref(jrec, *ap->a_vpp);
1063 save = jrecord_push(jrec, JTYPE_REDO);
1064 jrecord_write_vattr(jrec, ap->a_vap);
1065 jrecord_pop(jrec, save);
1068 jreclist_done(mp, &jreclist, error);
1073 * Journal vop_nmknod { a_ncp, a_vpp, a_cred, a_vap }
1077 journal_nmknod(struct vop_nmknod_args *ap)
1079 struct jrecord_list jreclist;
1080 struct jrecord jreccache;
1081 struct jrecord *jrec;
1086 mp = ap->a_head.a_ops->head.vv_mount;
1087 jreclist_init(mp, &jreclist, &jreccache, JTYPE_MKNOD);
1088 error = vop_journal_operate_ap(&ap->a_head);
1090 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
1091 jrecord_write_cred(jrec, NULL, ap->a_cred);
1092 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_ncp);
1093 save = jrecord_push(jrec, JTYPE_REDO);
1094 jrecord_write_vattr(jrec, ap->a_vap);
1095 jrecord_pop(jrec, save);
1097 jrecord_write_vnode_ref(jrec, *ap->a_vpp);
1100 jreclist_done(mp, &jreclist, error);
1105 * Journal vop_nlink { a_ncp, a_vp, a_cred }
1109 journal_nlink(struct vop_nlink_args *ap)
1111 struct jrecord_list jreclist;
1112 struct jrecord jreccache;
1113 struct jrecord *jrec;
1118 mp = ap->a_head.a_ops->head.vv_mount;
1119 jreclist_init(mp, &jreclist, &jreccache, JTYPE_LINK);
1120 error = vop_journal_operate_ap(&ap->a_head);
1122 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
1123 jrecord_write_cred(jrec, NULL, ap->a_cred);
1124 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_ncp);
1125 /* XXX PATH to VP and inode number */
1126 /* XXX this call may not record the correct path when
1127 * multiple paths are available */
1128 save = jrecord_push(jrec, JTYPE_REDO);
1129 jrecord_write_vnode_link(jrec, ap->a_vp, ap->a_ncp);
1130 jrecord_pop(jrec, save);
1133 jreclist_done(mp, &jreclist, error);
1138 * Journal vop_symlink { a_ncp, a_vpp, a_cred, a_vap, a_target }
1142 journal_nsymlink(struct vop_nsymlink_args *ap)
1144 struct jrecord_list jreclist;
1145 struct jrecord jreccache;
1146 struct jrecord *jrec;
1151 mp = ap->a_head.a_ops->head.vv_mount;
1152 jreclist_init(mp, &jreclist, &jreccache, JTYPE_SYMLINK);
1153 error = vop_journal_operate_ap(&ap->a_head);
1155 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
1156 jrecord_write_cred(jrec, NULL, ap->a_cred);
1157 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_ncp);
1158 save = jrecord_push(jrec, JTYPE_REDO);
1159 jrecord_leaf(jrec, JLEAF_SYMLINKDATA,
1160 ap->a_target, strlen(ap->a_target));
1161 jrecord_pop(jrec, save);
1163 jrecord_write_vnode_ref(jrec, *ap->a_vpp);
1166 jreclist_done(mp, &jreclist, error);
1171 * Journal vop_nwhiteout { a_ncp, a_cred, a_flags }
1175 journal_nwhiteout(struct vop_nwhiteout_args *ap)
1177 struct jrecord_list jreclist;
1178 struct jrecord jreccache;
1179 struct jrecord *jrec;
1183 mp = ap->a_head.a_ops->head.vv_mount;
1184 jreclist_init(mp, &jreclist, &jreccache, JTYPE_WHITEOUT);
1185 error = vop_journal_operate_ap(&ap->a_head);
1187 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
1188 jrecord_write_cred(jrec, NULL, ap->a_cred);
1189 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_ncp);
1192 jreclist_done(mp, &jreclist, error);
1197 * Journal vop_nremove { a_ncp, a_cred }
1201 journal_nremove(struct vop_nremove_args *ap)
1203 struct jrecord_list jreclist;
1204 struct jrecord jreccache;
1205 struct jrecord *jrec;
1209 mp = ap->a_head.a_ops->head.vv_mount;
1210 if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_REMOVE) &&
1213 jreclist_undo_file(&jreclist, ap->a_ncp->nc_vp,
1214 JRUNDO_ALL|JRUNDO_GETVP|JRUNDO_CONDLINK, 0, -1);
1216 error = vop_journal_operate_ap(&ap->a_head);
1218 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
1219 jrecord_write_cred(jrec, NULL, ap->a_cred);
1220 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_ncp);
1223 jreclist_done(mp, &jreclist, error);
1228 * Journal vop_nmkdir { a_ncp, a_vpp, a_cred, a_vap }
1232 journal_nmkdir(struct vop_nmkdir_args *ap)
1234 struct jrecord_list jreclist;
1235 struct jrecord jreccache;
1236 struct jrecord *jrec;
1240 mp = ap->a_head.a_ops->head.vv_mount;
1241 jreclist_init(mp, &jreclist, &jreccache, JTYPE_MKDIR);
1242 error = vop_journal_operate_ap(&ap->a_head);
1244 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
1246 if (jo->flags & MC_JOURNAL_WANT_AUDIT) {
1247 jrecord_write_audit(jrec);
1250 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_ncp);
1251 jrecord_write_cred(jrec, NULL, ap->a_cred);
1252 jrecord_write_vattr(jrec, ap->a_vap);
1253 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_ncp);
1255 jrecord_write_vnode_ref(jrec, *ap->a_vpp);
1258 jreclist_done(mp, &jreclist, error);
1263 * Journal vop_nrmdir { a_ncp, a_cred }
1267 journal_nrmdir(struct vop_nrmdir_args *ap)
1269 struct jrecord_list jreclist;
1270 struct jrecord jreccache;
1271 struct jrecord *jrec;
1275 mp = ap->a_head.a_ops->head.vv_mount;
1276 if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_RMDIR)) {
1277 jreclist_undo_file(&jreclist, ap->a_ncp->nc_vp,
1278 JRUNDO_VATTR|JRUNDO_GETVP, 0, 0);
1280 error = vop_journal_operate_ap(&ap->a_head);
1282 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
1283 jrecord_write_cred(jrec, NULL, ap->a_cred);
1284 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_ncp);
1287 jreclist_done(mp, &jreclist, error);
1292 * Journal vop_nrename { a_fncp, a_tncp, a_cred }
1296 journal_nrename(struct vop_nrename_args *ap)
1298 struct jrecord_list jreclist;
1299 struct jrecord jreccache;
1300 struct jrecord *jrec;
1304 mp = ap->a_head.a_ops->head.vv_mount;
1305 if (jreclist_init(mp, &jreclist, &jreccache, JTYPE_RENAME) &&
1308 jreclist_undo_file(&jreclist, ap->a_tncp->nc_vp,
1309 JRUNDO_ALL|JRUNDO_GETVP|JRUNDO_CONDLINK, 0, -1);
1311 error = vop_journal_operate_ap(&ap->a_head);
1313 TAILQ_FOREACH(jrec, &jreclist.list, user_entry) {
1314 jrecord_write_cred(jrec, NULL, ap->a_cred);
1315 jrecord_write_path(jrec, JLEAF_PATH1, ap->a_fncp);
1316 jrecord_write_path(jrec, JLEAF_PATH2, ap->a_tncp);
1319 jreclist_done(mp, &jreclist, error);