2 * Copyright (c) 2007-2008 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/vfs/hammer/hammer_vnops.c,v 1.44 2008/05/02 06:51:57 dillon Exp $
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
40 #include <sys/fcntl.h>
41 #include <sys/namecache.h>
42 #include <sys/vnode.h>
43 #include <sys/lockf.h>
44 #include <sys/event.h>
46 #include <sys/dirent.h>
47 #include <vm/vm_extern.h>
48 #include <vfs/fifofs/fifo.h>
54 /*static int hammer_vop_vnoperate(struct vop_generic_args *);*/
55 static int hammer_vop_fsync(struct vop_fsync_args
*);
56 static int hammer_vop_read(struct vop_read_args
*);
57 static int hammer_vop_write(struct vop_write_args
*);
58 static int hammer_vop_access(struct vop_access_args
*);
59 static int hammer_vop_advlock(struct vop_advlock_args
*);
60 static int hammer_vop_close(struct vop_close_args
*);
61 static int hammer_vop_ncreate(struct vop_ncreate_args
*);
62 static int hammer_vop_getattr(struct vop_getattr_args
*);
63 static int hammer_vop_nresolve(struct vop_nresolve_args
*);
64 static int hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args
*);
65 static int hammer_vop_nlink(struct vop_nlink_args
*);
66 static int hammer_vop_nmkdir(struct vop_nmkdir_args
*);
67 static int hammer_vop_nmknod(struct vop_nmknod_args
*);
68 static int hammer_vop_open(struct vop_open_args
*);
69 static int hammer_vop_pathconf(struct vop_pathconf_args
*);
70 static int hammer_vop_print(struct vop_print_args
*);
71 static int hammer_vop_readdir(struct vop_readdir_args
*);
72 static int hammer_vop_readlink(struct vop_readlink_args
*);
73 static int hammer_vop_nremove(struct vop_nremove_args
*);
74 static int hammer_vop_nrename(struct vop_nrename_args
*);
75 static int hammer_vop_nrmdir(struct vop_nrmdir_args
*);
76 static int hammer_vop_setattr(struct vop_setattr_args
*);
77 static int hammer_vop_strategy(struct vop_strategy_args
*);
78 static int hammer_vop_nsymlink(struct vop_nsymlink_args
*);
79 static int hammer_vop_nwhiteout(struct vop_nwhiteout_args
*);
80 static int hammer_vop_ioctl(struct vop_ioctl_args
*);
81 static int hammer_vop_mountctl(struct vop_mountctl_args
*);
83 static int hammer_vop_fifoclose (struct vop_close_args
*);
84 static int hammer_vop_fiforead (struct vop_read_args
*);
85 static int hammer_vop_fifowrite (struct vop_write_args
*);
87 static int hammer_vop_specclose (struct vop_close_args
*);
88 static int hammer_vop_specread (struct vop_read_args
*);
89 static int hammer_vop_specwrite (struct vop_write_args
*);
91 struct vop_ops hammer_vnode_vops
= {
92 .vop_default
= vop_defaultop
,
93 .vop_fsync
= hammer_vop_fsync
,
94 .vop_getpages
= vop_stdgetpages
,
95 .vop_putpages
= vop_stdputpages
,
96 .vop_read
= hammer_vop_read
,
97 .vop_write
= hammer_vop_write
,
98 .vop_access
= hammer_vop_access
,
99 .vop_advlock
= hammer_vop_advlock
,
100 .vop_close
= hammer_vop_close
,
101 .vop_ncreate
= hammer_vop_ncreate
,
102 .vop_getattr
= hammer_vop_getattr
,
103 .vop_inactive
= hammer_vop_inactive
,
104 .vop_reclaim
= hammer_vop_reclaim
,
105 .vop_nresolve
= hammer_vop_nresolve
,
106 .vop_nlookupdotdot
= hammer_vop_nlookupdotdot
,
107 .vop_nlink
= hammer_vop_nlink
,
108 .vop_nmkdir
= hammer_vop_nmkdir
,
109 .vop_nmknod
= hammer_vop_nmknod
,
110 .vop_open
= hammer_vop_open
,
111 .vop_pathconf
= hammer_vop_pathconf
,
112 .vop_print
= hammer_vop_print
,
113 .vop_readdir
= hammer_vop_readdir
,
114 .vop_readlink
= hammer_vop_readlink
,
115 .vop_nremove
= hammer_vop_nremove
,
116 .vop_nrename
= hammer_vop_nrename
,
117 .vop_nrmdir
= hammer_vop_nrmdir
,
118 .vop_setattr
= hammer_vop_setattr
,
119 .vop_strategy
= hammer_vop_strategy
,
120 .vop_nsymlink
= hammer_vop_nsymlink
,
121 .vop_nwhiteout
= hammer_vop_nwhiteout
,
122 .vop_ioctl
= hammer_vop_ioctl
,
123 .vop_mountctl
= hammer_vop_mountctl
126 struct vop_ops hammer_spec_vops
= {
127 .vop_default
= spec_vnoperate
,
128 .vop_fsync
= hammer_vop_fsync
,
129 .vop_read
= hammer_vop_specread
,
130 .vop_write
= hammer_vop_specwrite
,
131 .vop_access
= hammer_vop_access
,
132 .vop_close
= hammer_vop_specclose
,
133 .vop_getattr
= hammer_vop_getattr
,
134 .vop_inactive
= hammer_vop_inactive
,
135 .vop_reclaim
= hammer_vop_reclaim
,
136 .vop_setattr
= hammer_vop_setattr
139 struct vop_ops hammer_fifo_vops
= {
140 .vop_default
= fifo_vnoperate
,
141 .vop_fsync
= hammer_vop_fsync
,
142 .vop_read
= hammer_vop_fiforead
,
143 .vop_write
= hammer_vop_fifowrite
,
144 .vop_access
= hammer_vop_access
,
145 .vop_close
= hammer_vop_fifoclose
,
146 .vop_getattr
= hammer_vop_getattr
,
147 .vop_inactive
= hammer_vop_inactive
,
148 .vop_reclaim
= hammer_vop_reclaim
,
149 .vop_setattr
= hammer_vop_setattr
152 static int hammer_dounlink(hammer_transaction_t trans
, struct nchandle
*nch
,
153 struct vnode
*dvp
, struct ucred
*cred
, int flags
);
154 static int hammer_vop_strategy_read(struct vop_strategy_args
*ap
);
155 static int hammer_vop_strategy_write(struct vop_strategy_args
*ap
);
160 hammer_vop_vnoperate(struct vop_generic_args
*)
162 return (VOCALL(&hammer_vnode_vops
, ap
));
167 * hammer_vop_fsync { vp, waitfor }
171 hammer_vop_fsync(struct vop_fsync_args
*ap
)
173 hammer_inode_t ip
= VTOI(ap
->a_vp
);
175 hammer_flush_inode(ip
, HAMMER_FLUSH_SIGNAL
);
176 if (ap
->a_waitfor
== MNT_WAIT
)
177 hammer_wait_inode(ip
);
182 * hammer_vop_read { vp, uio, ioflag, cred }
186 hammer_vop_read(struct vop_read_args
*ap
)
188 struct hammer_transaction trans
;
197 if (ap
->a_vp
->v_type
!= VREG
)
201 seqcount
= ap
->a_ioflag
>> 16;
203 hammer_start_transaction(&trans
, ip
->hmp
);
206 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
209 while (uio
->uio_resid
> 0 && uio
->uio_offset
< ip
->ino_rec
.ino_size
) {
210 offset
= uio
->uio_offset
& HAMMER_BUFMASK
;
212 error
= cluster_read(ap
->a_vp
, ip
->ino_rec
.ino_size
,
213 uio
->uio_offset
- offset
, HAMMER_BUFSIZE
,
214 MAXBSIZE
, seqcount
, &bp
);
216 error
= bread(ap
->a_vp
, uio
->uio_offset
- offset
,
217 HAMMER_BUFSIZE
, &bp
);
222 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
223 n
= HAMMER_BUFSIZE
- offset
;
224 if (n
> uio
->uio_resid
)
226 if (n
> ip
->ino_rec
.ino_size
- uio
->uio_offset
)
227 n
= (int)(ip
->ino_rec
.ino_size
- uio
->uio_offset
);
228 error
= uiomove((char *)bp
->b_data
+ offset
, n
, uio
);
235 if ((ip
->flags
& HAMMER_INODE_RO
) == 0 &&
236 (ip
->hmp
->mp
->mnt_flag
& MNT_NOATIME
) == 0) {
237 ip
->ino_rec
.ino_atime
= trans
.time
;
238 hammer_modify_inode(&trans
, ip
, HAMMER_INODE_ITIMES
);
240 hammer_done_transaction(&trans
);
245 * hammer_vop_write { vp, uio, ioflag, cred }
249 hammer_vop_write(struct vop_write_args
*ap
)
251 struct hammer_transaction trans
;
252 struct hammer_inode
*ip
;
261 if (ap
->a_vp
->v_type
!= VREG
)
266 if (ip
->flags
& HAMMER_INODE_RO
)
270 * Create a transaction to cover the operations we perform.
272 hammer_start_transaction(&trans
, ip
->hmp
);
278 if (ap
->a_ioflag
& IO_APPEND
)
279 uio
->uio_offset
= ip
->ino_rec
.ino_size
;
282 * Check for illegal write offsets. Valid range is 0...2^63-1
284 if (uio
->uio_offset
< 0 || uio
->uio_offset
+ uio
->uio_resid
<= 0) {
285 hammer_done_transaction(&trans
);
290 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
293 while (uio
->uio_resid
> 0) {
297 * Do not allow huge writes to deadlock the buffer cache
299 if ((++count
& 15) == 0) {
301 if ((ap
->a_ioflag
& IO_NOBWILL
) == 0)
303 vn_lock(ap
->a_vp
, LK_EXCLUSIVE
|LK_RETRY
);
306 offset
= uio
->uio_offset
& HAMMER_BUFMASK
;
307 n
= HAMMER_BUFSIZE
- offset
;
308 if (n
> uio
->uio_resid
)
310 if (uio
->uio_offset
+ n
> ip
->ino_rec
.ino_size
) {
311 vnode_pager_setsize(ap
->a_vp
, uio
->uio_offset
+ n
);
315 if (uio
->uio_segflg
== UIO_NOCOPY
) {
317 * Issuing a write with the same data backing the
318 * buffer. Instantiate the buffer to collect the
319 * backing vm pages, then read-in any missing bits.
321 * This case is used by vop_stdputpages().
323 bp
= getblk(ap
->a_vp
, uio
->uio_offset
- offset
,
324 HAMMER_BUFSIZE
, GETBLK_BHEAVY
, 0);
325 if ((bp
->b_flags
& B_CACHE
) == 0) {
327 error
= bread(ap
->a_vp
,
328 uio
->uio_offset
- offset
,
329 HAMMER_BUFSIZE
, &bp
);
331 } else if (offset
== 0 && uio
->uio_resid
>= HAMMER_BUFSIZE
) {
333 * Even though we are entirely overwriting the buffer
334 * we may still have to zero it out to avoid a
335 * mmap/write visibility issue.
337 bp
= getblk(ap
->a_vp
, uio
->uio_offset
- offset
,
338 HAMMER_BUFSIZE
, GETBLK_BHEAVY
, 0);
339 if ((bp
->b_flags
& B_CACHE
) == 0)
341 } else if (uio
->uio_offset
- offset
>= ip
->ino_rec
.ino_size
) {
343 * If the base offset of the buffer is beyond the
344 * file EOF, we don't have to issue a read.
346 bp
= getblk(ap
->a_vp
, uio
->uio_offset
- offset
,
347 HAMMER_BUFSIZE
, GETBLK_BHEAVY
, 0);
351 * Partial overwrite, read in any missing bits then
352 * replace the portion being written.
354 error
= bread(ap
->a_vp
, uio
->uio_offset
- offset
,
355 HAMMER_BUFSIZE
, &bp
);
360 error
= uiomove((char *)bp
->b_data
+ offset
, n
, uio
);
363 * If we screwed up we have to undo any VM size changes we
369 vtruncbuf(ap
->a_vp
, ip
->ino_rec
.ino_size
,
374 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
375 hammer_lock_sh(&ip
->lock
);
376 if (ip
->ino_rec
.ino_size
< uio
->uio_offset
) {
377 ip
->ino_rec
.ino_size
= uio
->uio_offset
;
378 flags
= HAMMER_INODE_RDIRTY
;
379 vnode_pager_setsize(ap
->a_vp
, ip
->ino_rec
.ino_size
);
383 ip
->ino_rec
.ino_mtime
= trans
.time
;
384 flags
|= HAMMER_INODE_ITIMES
| HAMMER_INODE_BUFS
;
385 hammer_modify_inode(&trans
, ip
, flags
);
386 hammer_unlock(&ip
->lock
);
388 if (ap
->a_ioflag
& IO_SYNC
) {
390 } else if (ap
->a_ioflag
& IO_DIRECT
) {
393 } else if ((ap
->a_ioflag
>> 16) == IO_SEQMAX
&&
394 (uio
->uio_offset
& HAMMER_BUFMASK
) == 0) {
396 * XXX HAMMER can only fsync the whole inode,
397 * doing it on every buffer would be a bad idea.
400 * If seqcount indicates sequential operation and
401 * we just finished filling a buffer, push it out
402 * now to prevent the buffer cache from becoming
403 * too full, which would trigger non-optimal
412 hammer_done_transaction(&trans
);
417 * hammer_vop_access { vp, mode, cred }
421 hammer_vop_access(struct vop_access_args
*ap
)
423 struct hammer_inode
*ip
= VTOI(ap
->a_vp
);
428 uid
= hammer_to_unix_xid(&ip
->ino_data
.uid
);
429 gid
= hammer_to_unix_xid(&ip
->ino_data
.gid
);
431 error
= vop_helper_access(ap
, uid
, gid
, ip
->ino_data
.mode
,
432 ip
->ino_data
.uflags
);
437 * hammer_vop_advlock { vp, id, op, fl, flags }
441 hammer_vop_advlock(struct vop_advlock_args
*ap
)
443 struct hammer_inode
*ip
= VTOI(ap
->a_vp
);
445 return (lf_advlock(ap
, &ip
->advlock
, ip
->ino_rec
.ino_size
));
449 * hammer_vop_close { vp, fflag }
453 hammer_vop_close(struct vop_close_args
*ap
)
455 return (vop_stdclose(ap
));
459 * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
461 * The operating system has already ensured that the directory entry
462 * does not exist and done all appropriate namespace locking.
466 hammer_vop_ncreate(struct vop_ncreate_args
*ap
)
468 struct hammer_transaction trans
;
469 struct hammer_inode
*dip
;
470 struct hammer_inode
*nip
;
471 struct nchandle
*nch
;
475 dip
= VTOI(ap
->a_dvp
);
477 if (dip
->flags
& HAMMER_INODE_RO
)
481 * Create a transaction to cover the operations we perform.
483 hammer_start_transaction(&trans
, dip
->hmp
);
486 * Create a new filesystem object of the requested type. The
487 * returned inode will be referenced and shared-locked to prevent
488 * it from being moved to the flusher.
491 error
= hammer_create_inode(&trans
, ap
->a_vap
, ap
->a_cred
, dip
, &nip
);
493 kprintf("hammer_create_inode error %d\n", error
);
494 hammer_done_transaction(&trans
);
498 hammer_lock_sh(&nip
->lock
);
499 hammer_lock_sh(&dip
->lock
);
502 * Add the new filesystem object to the directory. This will also
503 * bump the inode's link count.
505 error
= hammer_ip_add_directory(&trans
, dip
, nch
->ncp
, nip
);
507 kprintf("hammer_ip_add_directory error %d\n", error
);
508 hammer_unlock(&dip
->lock
);
509 hammer_unlock(&nip
->lock
);
515 hammer_rel_inode(nip
, 0);
516 hammer_done_transaction(&trans
);
519 error
= hammer_get_vnode(nip
, LK_EXCLUSIVE
, ap
->a_vpp
);
520 hammer_done_transaction(&trans
);
521 hammer_rel_inode(nip
, 0);
523 cache_setunresolved(ap
->a_nch
);
524 cache_setvp(ap
->a_nch
, *ap
->a_vpp
);
531 * hammer_vop_getattr { vp, vap }
535 hammer_vop_getattr(struct vop_getattr_args
*ap
)
537 struct hammer_inode
*ip
= VTOI(ap
->a_vp
);
538 struct vattr
*vap
= ap
->a_vap
;
541 if (cache_check_fsmid_vp(ap
->a_vp
, &ip
->fsmid
) &&
542 (vp
->v_mount
->mnt_flag
& MNT_RDONLY
) == 0 &&
547 hammer_itimes(ap
->a_vp
);
550 vap
->va_fsid
= ip
->hmp
->fsid_udev
;
551 vap
->va_fileid
= ip
->ino_rec
.base
.base
.obj_id
;
552 vap
->va_mode
= ip
->ino_data
.mode
;
553 vap
->va_nlink
= ip
->ino_rec
.ino_nlinks
;
554 vap
->va_uid
= hammer_to_unix_xid(&ip
->ino_data
.uid
);
555 vap
->va_gid
= hammer_to_unix_xid(&ip
->ino_data
.gid
);
558 vap
->va_size
= ip
->ino_rec
.ino_size
;
559 hammer_to_timespec(ip
->ino_rec
.ino_atime
, &vap
->va_atime
);
560 hammer_to_timespec(ip
->ino_rec
.ino_mtime
, &vap
->va_mtime
);
561 hammer_to_timespec(ip
->ino_data
.ctime
, &vap
->va_ctime
);
562 vap
->va_flags
= ip
->ino_data
.uflags
;
563 vap
->va_gen
= 1; /* hammer inums are unique for all time */
564 vap
->va_blocksize
= HAMMER_BUFSIZE
;
565 vap
->va_bytes
= (ip
->ino_rec
.ino_size
+ 63) & ~63;
566 vap
->va_type
= hammer_get_vnode_type(ip
->ino_rec
.base
.base
.obj_type
);
567 vap
->va_filerev
= 0; /* XXX */
568 /* mtime uniquely identifies any adjustments made to the file */
569 vap
->va_fsmid
= ip
->ino_rec
.ino_mtime
;
570 vap
->va_uid_uuid
= ip
->ino_data
.uid
;
571 vap
->va_gid_uuid
= ip
->ino_data
.gid
;
572 vap
->va_fsid_uuid
= ip
->hmp
->fsid
;
573 vap
->va_vaflags
= VA_UID_UUID_VALID
| VA_GID_UUID_VALID
|
576 switch (ip
->ino_rec
.base
.base
.obj_type
) {
577 case HAMMER_OBJTYPE_CDEV
:
578 case HAMMER_OBJTYPE_BDEV
:
579 vap
->va_rmajor
= ip
->ino_data
.rmajor
;
580 vap
->va_rminor
= ip
->ino_data
.rminor
;
590 * hammer_vop_nresolve { nch, dvp, cred }
592 * Locate the requested directory entry.
596 hammer_vop_nresolve(struct vop_nresolve_args
*ap
)
598 struct hammer_transaction trans
;
599 struct namecache
*ncp
;
603 struct hammer_cursor cursor
;
604 union hammer_record_ondisk
*rec
;
614 * Misc initialization, plus handle as-of name extensions. Look for
615 * the '@@' extension. Note that as-of files and directories cannot
618 dip
= VTOI(ap
->a_dvp
);
619 ncp
= ap
->a_nch
->ncp
;
620 asof
= dip
->obj_asof
;
624 hammer_simple_transaction(&trans
, dip
->hmp
);
626 for (i
= 0; i
< nlen
; ++i
) {
627 if (ncp
->nc_name
[i
] == '@' && ncp
->nc_name
[i
+1] == '@') {
628 asof
= hammer_str_to_tid(ncp
->nc_name
+ i
+ 2);
629 flags
|= HAMMER_INODE_RO
;
636 * If there is no path component the time extension is relative to
640 ip
= hammer_get_inode(&trans
, &dip
->cache
[1], dip
->obj_id
,
641 asof
, flags
, &error
);
643 error
= hammer_get_vnode(ip
, LK_EXCLUSIVE
, &vp
);
644 hammer_rel_inode(ip
, 0);
650 cache_setvp(ap
->a_nch
, vp
);
657 * Calculate the namekey and setup the key range for the scan. This
658 * works kinda like a chained hash table where the lower 32 bits
659 * of the namekey synthesize the chain.
661 * The key range is inclusive of both key_beg and key_end.
663 namekey
= hammer_directory_namekey(ncp
->nc_name
, nlen
);
665 error
= hammer_init_cursor(&trans
, &cursor
, &dip
->cache
[0]);
666 cursor
.key_beg
.obj_id
= dip
->obj_id
;
667 cursor
.key_beg
.key
= namekey
;
668 cursor
.key_beg
.create_tid
= 0;
669 cursor
.key_beg
.delete_tid
= 0;
670 cursor
.key_beg
.rec_type
= HAMMER_RECTYPE_DIRENTRY
;
671 cursor
.key_beg
.obj_type
= 0;
673 cursor
.key_end
= cursor
.key_beg
;
674 cursor
.key_end
.key
|= 0xFFFFFFFFULL
;
676 cursor
.flags
|= HAMMER_CURSOR_END_INCLUSIVE
| HAMMER_CURSOR_ASOF
;
679 * Scan all matching records (the chain), locate the one matching
680 * the requested path component.
682 * The hammer_ip_*() functions merge in-memory records with on-disk
683 * records for the purposes of the search.
686 error
= hammer_ip_first(&cursor
, dip
);
692 error
= hammer_ip_resolve_data(&cursor
);
696 if (nlen
== rec
->entry
.base
.data_len
&&
697 bcmp(ncp
->nc_name
, cursor
.data
, nlen
) == 0) {
698 obj_id
= rec
->entry
.obj_id
;
701 error
= hammer_ip_next(&cursor
);
703 hammer_done_cursor(&cursor
);
705 ip
= hammer_get_inode(&trans
, &dip
->cache
[1],
706 obj_id
, asof
, flags
, &error
);
708 error
= hammer_get_vnode(ip
, LK_EXCLUSIVE
, &vp
);
709 hammer_rel_inode(ip
, 0);
715 cache_setvp(ap
->a_nch
, vp
);
718 } else if (error
== ENOENT
) {
719 cache_setvp(ap
->a_nch
, NULL
);
722 hammer_done_transaction(&trans
);
727 * hammer_vop_nlookupdotdot { dvp, vpp, cred }
729 * Locate the parent directory of a directory vnode.
731 * dvp is referenced but not locked. *vpp must be returned referenced and
732 * locked. A parent_obj_id of 0 does not necessarily indicate that we are
733 * at the root, instead it could indicate that the directory we were in was
736 * NOTE: as-of sequences are not linked into the directory structure. If
737 * we are at the root with a different asof then the mount point, reload
738 * the same directory with the mount point's asof. I'm not sure what this
739 * will do to NFS. We encode ASOF stamps in NFS file handles so it might not
740 * get confused, but it hasn't been tested.
744 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args
*ap
)
746 struct hammer_transaction trans
;
747 struct hammer_inode
*dip
;
748 struct hammer_inode
*ip
;
749 int64_t parent_obj_id
;
753 dip
= VTOI(ap
->a_dvp
);
754 asof
= dip
->obj_asof
;
755 parent_obj_id
= dip
->ino_data
.parent_obj_id
;
757 if (parent_obj_id
== 0) {
758 if (dip
->obj_id
== HAMMER_OBJID_ROOT
&&
759 asof
!= dip
->hmp
->asof
) {
760 parent_obj_id
= dip
->obj_id
;
761 asof
= dip
->hmp
->asof
;
762 *ap
->a_fakename
= kmalloc(19, M_TEMP
, M_WAITOK
);
763 ksnprintf(*ap
->a_fakename
, 19, "0x%016llx",
771 hammer_simple_transaction(&trans
, dip
->hmp
);
773 ip
= hammer_get_inode(&trans
, &dip
->cache
[1], parent_obj_id
,
774 asof
, dip
->flags
, &error
);
776 error
= hammer_get_vnode(ip
, LK_EXCLUSIVE
, ap
->a_vpp
);
777 hammer_rel_inode(ip
, 0);
781 hammer_done_transaction(&trans
);
786 * hammer_vop_nlink { nch, dvp, vp, cred }
790 hammer_vop_nlink(struct vop_nlink_args
*ap
)
792 struct hammer_transaction trans
;
793 struct hammer_inode
*dip
;
794 struct hammer_inode
*ip
;
795 struct nchandle
*nch
;
799 dip
= VTOI(ap
->a_dvp
);
802 if (dip
->flags
& HAMMER_INODE_RO
)
804 if (ip
->flags
& HAMMER_INODE_RO
)
808 * Create a transaction to cover the operations we perform.
810 hammer_start_transaction(&trans
, dip
->hmp
);
813 * Add the filesystem object to the directory. Note that neither
814 * dip nor ip are referenced or locked, but their vnodes are
815 * referenced. This function will bump the inode's link count.
817 hammer_lock_sh(&ip
->lock
);
818 hammer_lock_sh(&dip
->lock
);
819 error
= hammer_ip_add_directory(&trans
, dip
, nch
->ncp
, ip
);
820 hammer_unlock(&dip
->lock
);
821 hammer_unlock(&ip
->lock
);
827 cache_setunresolved(nch
);
828 cache_setvp(nch
, ap
->a_vp
);
830 hammer_done_transaction(&trans
);
835 * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
837 * The operating system has already ensured that the directory entry
838 * does not exist and done all appropriate namespace locking.
842 hammer_vop_nmkdir(struct vop_nmkdir_args
*ap
)
844 struct hammer_transaction trans
;
845 struct hammer_inode
*dip
;
846 struct hammer_inode
*nip
;
847 struct nchandle
*nch
;
851 dip
= VTOI(ap
->a_dvp
);
853 if (dip
->flags
& HAMMER_INODE_RO
)
857 * Create a transaction to cover the operations we perform.
859 hammer_start_transaction(&trans
, dip
->hmp
);
862 * Create a new filesystem object of the requested type. The
863 * returned inode will be referenced but not locked.
865 error
= hammer_create_inode(&trans
, ap
->a_vap
, ap
->a_cred
, dip
, &nip
);
867 kprintf("hammer_mkdir error %d\n", error
);
868 hammer_done_transaction(&trans
);
873 * Add the new filesystem object to the directory. This will also
874 * bump the inode's link count.
876 hammer_lock_sh(&nip
->lock
);
877 hammer_lock_sh(&dip
->lock
);
878 error
= hammer_ip_add_directory(&trans
, dip
, nch
->ncp
, nip
);
879 hammer_unlock(&dip
->lock
);
880 hammer_unlock(&nip
->lock
);
882 kprintf("hammer_mkdir (add) error %d\n", error
);
888 hammer_rel_inode(nip
, 0);
891 error
= hammer_get_vnode(nip
, LK_EXCLUSIVE
, ap
->a_vpp
);
892 hammer_rel_inode(nip
, 0);
894 cache_setunresolved(ap
->a_nch
);
895 cache_setvp(ap
->a_nch
, *ap
->a_vpp
);
898 hammer_done_transaction(&trans
);
903 * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
905 * The operating system has already ensured that the directory entry
906 * does not exist and done all appropriate namespace locking.
910 hammer_vop_nmknod(struct vop_nmknod_args
*ap
)
912 struct hammer_transaction trans
;
913 struct hammer_inode
*dip
;
914 struct hammer_inode
*nip
;
915 struct nchandle
*nch
;
919 dip
= VTOI(ap
->a_dvp
);
921 if (dip
->flags
& HAMMER_INODE_RO
)
925 * Create a transaction to cover the operations we perform.
927 hammer_start_transaction(&trans
, dip
->hmp
);
930 * Create a new filesystem object of the requested type. The
931 * returned inode will be referenced but not locked.
933 error
= hammer_create_inode(&trans
, ap
->a_vap
, ap
->a_cred
, dip
, &nip
);
935 hammer_done_transaction(&trans
);
941 * Add the new filesystem object to the directory. This will also
942 * bump the inode's link count.
944 hammer_lock_sh(&nip
->lock
);
945 hammer_lock_sh(&dip
->lock
);
946 error
= hammer_ip_add_directory(&trans
, dip
, nch
->ncp
, nip
);
947 hammer_unlock(&dip
->lock
);
948 hammer_unlock(&nip
->lock
);
954 hammer_rel_inode(nip
, 0);
957 error
= hammer_get_vnode(nip
, LK_EXCLUSIVE
, ap
->a_vpp
);
958 hammer_rel_inode(nip
, 0);
960 cache_setunresolved(ap
->a_nch
);
961 cache_setvp(ap
->a_nch
, *ap
->a_vpp
);
964 hammer_done_transaction(&trans
);
969 * hammer_vop_open { vp, mode, cred, fp }
973 hammer_vop_open(struct vop_open_args
*ap
)
975 if ((ap
->a_mode
& FWRITE
) && (VTOI(ap
->a_vp
)->flags
& HAMMER_INODE_RO
))
978 return(vop_stdopen(ap
));
982 * hammer_vop_pathconf { vp, name, retval }
986 hammer_vop_pathconf(struct vop_pathconf_args
*ap
)
992 * hammer_vop_print { vp }
996 hammer_vop_print(struct vop_print_args
*ap
)
1002 * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
1006 hammer_vop_readdir(struct vop_readdir_args
*ap
)
1008 struct hammer_transaction trans
;
1009 struct hammer_cursor cursor
;
1010 struct hammer_inode
*ip
;
1012 hammer_record_ondisk_t rec
;
1013 hammer_base_elm_t base
;
1021 ip
= VTOI(ap
->a_vp
);
1023 saveoff
= uio
->uio_offset
;
1025 if (ap
->a_ncookies
) {
1026 ncookies
= uio
->uio_resid
/ 16 + 1;
1027 if (ncookies
> 1024)
1029 cookies
= kmalloc(ncookies
* sizeof(off_t
), M_TEMP
, M_WAITOK
);
1037 hammer_simple_transaction(&trans
, ip
->hmp
);
1040 * Handle artificial entries
1044 r
= vop_write_dirent(&error
, uio
, ip
->obj_id
, DT_DIR
, 1, ".");
1048 cookies
[cookie_index
] = saveoff
;
1051 if (cookie_index
== ncookies
)
1055 if (ip
->ino_data
.parent_obj_id
) {
1056 r
= vop_write_dirent(&error
, uio
,
1057 ip
->ino_data
.parent_obj_id
,
1060 r
= vop_write_dirent(&error
, uio
,
1061 ip
->obj_id
, DT_DIR
, 2, "..");
1066 cookies
[cookie_index
] = saveoff
;
1069 if (cookie_index
== ncookies
)
1074 * Key range (begin and end inclusive) to scan. Directory keys
1075 * directly translate to a 64 bit 'seek' position.
1077 hammer_init_cursor(&trans
, &cursor
, &ip
->cache
[0]);
1078 cursor
.key_beg
.obj_id
= ip
->obj_id
;
1079 cursor
.key_beg
.create_tid
= 0;
1080 cursor
.key_beg
.delete_tid
= 0;
1081 cursor
.key_beg
.rec_type
= HAMMER_RECTYPE_DIRENTRY
;
1082 cursor
.key_beg
.obj_type
= 0;
1083 cursor
.key_beg
.key
= saveoff
;
1085 cursor
.key_end
= cursor
.key_beg
;
1086 cursor
.key_end
.key
= HAMMER_MAX_KEY
;
1087 cursor
.asof
= ip
->obj_asof
;
1088 cursor
.flags
|= HAMMER_CURSOR_END_INCLUSIVE
| HAMMER_CURSOR_ASOF
;
1090 error
= hammer_ip_first(&cursor
, ip
);
1092 while (error
== 0) {
1093 error
= hammer_ip_resolve_record_and_data(&cursor
);
1096 rec
= cursor
.record
;
1097 base
= &rec
->base
.base
;
1098 saveoff
= base
->key
;
1100 if (base
->obj_id
!= ip
->obj_id
)
1101 panic("readdir: bad record at %p", cursor
.node
);
1103 r
= vop_write_dirent(
1104 &error
, uio
, rec
->entry
.obj_id
,
1105 hammer_get_dtype(rec
->entry
.base
.base
.obj_type
),
1106 rec
->entry
.base
.data_len
,
1107 (void *)cursor
.data
);
1112 cookies
[cookie_index
] = base
->key
;
1114 if (cookie_index
== ncookies
)
1116 error
= hammer_ip_next(&cursor
);
1118 hammer_done_cursor(&cursor
);
1121 hammer_done_transaction(&trans
);
1124 *ap
->a_eofflag
= (error
== ENOENT
);
1125 uio
->uio_offset
= saveoff
;
1126 if (error
&& cookie_index
== 0) {
1127 if (error
== ENOENT
)
1130 kfree(cookies
, M_TEMP
);
1131 *ap
->a_ncookies
= 0;
1132 *ap
->a_cookies
= NULL
;
1135 if (error
== ENOENT
)
1138 *ap
->a_ncookies
= cookie_index
;
1139 *ap
->a_cookies
= cookies
;
1146 * hammer_vop_readlink { vp, uio, cred }
1150 hammer_vop_readlink(struct vop_readlink_args
*ap
)
1152 struct hammer_transaction trans
;
1153 struct hammer_cursor cursor
;
1154 struct hammer_inode
*ip
;
1157 ip
= VTOI(ap
->a_vp
);
1159 hammer_simple_transaction(&trans
, ip
->hmp
);
1161 hammer_init_cursor(&trans
, &cursor
, &ip
->cache
[0]);
1164 * Key range (begin and end inclusive) to scan. Directory keys
1165 * directly translate to a 64 bit 'seek' position.
1167 cursor
.key_beg
.obj_id
= ip
->obj_id
;
1168 cursor
.key_beg
.create_tid
= 0;
1169 cursor
.key_beg
.delete_tid
= 0;
1170 cursor
.key_beg
.rec_type
= HAMMER_RECTYPE_FIX
;
1171 cursor
.key_beg
.obj_type
= 0;
1172 cursor
.key_beg
.key
= HAMMER_FIXKEY_SYMLINK
;
1173 cursor
.asof
= ip
->obj_asof
;
1174 cursor
.flags
|= HAMMER_CURSOR_ASOF
;
1176 error
= hammer_ip_lookup(&cursor
, ip
);
1178 error
= hammer_ip_resolve_data(&cursor
);
1180 error
= uiomove((char *)cursor
.data
,
1181 cursor
.record
->base
.data_len
,
1185 hammer_done_cursor(&cursor
);
1186 hammer_done_transaction(&trans
);
1191 * hammer_vop_nremove { nch, dvp, cred }
1195 hammer_vop_nremove(struct vop_nremove_args
*ap
)
1197 struct hammer_transaction trans
;
1200 hammer_start_transaction(&trans
, VTOI(ap
->a_dvp
)->hmp
);
1201 error
= hammer_dounlink(&trans
, ap
->a_nch
, ap
->a_dvp
, ap
->a_cred
, 0);
1202 hammer_done_transaction(&trans
);
1208 * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1212 hammer_vop_nrename(struct vop_nrename_args
*ap
)
1214 struct hammer_transaction trans
;
1215 struct namecache
*fncp
;
1216 struct namecache
*tncp
;
1217 struct hammer_inode
*fdip
;
1218 struct hammer_inode
*tdip
;
1219 struct hammer_inode
*ip
;
1220 struct hammer_cursor cursor
;
1221 union hammer_record_ondisk
*rec
;
1225 fdip
= VTOI(ap
->a_fdvp
);
1226 tdip
= VTOI(ap
->a_tdvp
);
1227 fncp
= ap
->a_fnch
->ncp
;
1228 tncp
= ap
->a_tnch
->ncp
;
1229 ip
= VTOI(fncp
->nc_vp
);
1230 KKASSERT(ip
!= NULL
);
1232 if (fdip
->flags
& HAMMER_INODE_RO
)
1234 if (tdip
->flags
& HAMMER_INODE_RO
)
1236 if (ip
->flags
& HAMMER_INODE_RO
)
1239 hammer_start_transaction(&trans
, fdip
->hmp
);
1241 hammer_lock_sh(&ip
->lock
);
1242 if (fdip
->obj_id
< tdip
->obj_id
) {
1243 hammer_lock_sh(&fdip
->lock
);
1244 hammer_lock_sh(&tdip
->lock
);
1246 hammer_lock_sh(&tdip
->lock
);
1247 hammer_lock_sh(&fdip
->lock
);
1251 * Remove tncp from the target directory and then link ip as
1252 * tncp. XXX pass trans to dounlink
1254 * Force the inode sync-time to match the transaction so it is
1255 * in-sync with the creation of the target directory entry.
1257 error
= hammer_dounlink(&trans
, ap
->a_tnch
, ap
->a_tdvp
, ap
->a_cred
, 0);
1258 if (error
== 0 || error
== ENOENT
) {
1259 error
= hammer_ip_add_directory(&trans
, tdip
, tncp
, ip
);
1261 ip
->ino_data
.parent_obj_id
= tdip
->obj_id
;
1262 hammer_modify_inode(&trans
, ip
, HAMMER_INODE_DDIRTY
);
1266 goto failed
; /* XXX */
1269 * Locate the record in the originating directory and remove it.
1271 * Calculate the namekey and setup the key range for the scan. This
1272 * works kinda like a chained hash table where the lower 32 bits
1273 * of the namekey synthesize the chain.
1275 * The key range is inclusive of both key_beg and key_end.
1277 namekey
= hammer_directory_namekey(fncp
->nc_name
, fncp
->nc_nlen
);
1279 hammer_init_cursor(&trans
, &cursor
, &fdip
->cache
[0]);
1280 cursor
.key_beg
.obj_id
= fdip
->obj_id
;
1281 cursor
.key_beg
.key
= namekey
;
1282 cursor
.key_beg
.create_tid
= 0;
1283 cursor
.key_beg
.delete_tid
= 0;
1284 cursor
.key_beg
.rec_type
= HAMMER_RECTYPE_DIRENTRY
;
1285 cursor
.key_beg
.obj_type
= 0;
1287 cursor
.key_end
= cursor
.key_beg
;
1288 cursor
.key_end
.key
|= 0xFFFFFFFFULL
;
1289 cursor
.asof
= fdip
->obj_asof
;
1290 cursor
.flags
|= HAMMER_CURSOR_END_INCLUSIVE
| HAMMER_CURSOR_ASOF
;
1293 * Scan all matching records (the chain), locate the one matching
1294 * the requested path component.
1296 * The hammer_ip_*() functions merge in-memory records with on-disk
1297 * records for the purposes of the search.
1299 error
= hammer_ip_first(&cursor
, fdip
);
1300 while (error
== 0) {
1301 if (hammer_ip_resolve_data(&cursor
) != 0)
1303 rec
= cursor
.record
;
1304 if (fncp
->nc_nlen
== rec
->entry
.base
.data_len
&&
1305 bcmp(fncp
->nc_name
, cursor
.data
, fncp
->nc_nlen
) == 0) {
1308 error
= hammer_ip_next(&cursor
);
1312 * If all is ok we have to get the inode so we can adjust nlinks.
1314 * WARNING: hammer_ip_del_directory() may have to terminate the
1315 * cursor to avoid a recursion. It's ok to call hammer_done_cursor()
1319 error
= hammer_ip_del_directory(&trans
, &cursor
, fdip
, ip
);
1322 * XXX A deadlock here will break rename's atomicy for the purposes
1323 * of crash recovery.
1325 if (error
== EDEADLK
) {
1326 hammer_unlock(&ip
->lock
);
1327 hammer_unlock(&fdip
->lock
);
1328 hammer_unlock(&tdip
->lock
);
1329 hammer_done_cursor(&cursor
);
1330 hammer_lock_sh(&ip
->lock
);
1331 if (fdip
->obj_id
< tdip
->obj_id
) {
1332 hammer_lock_sh(&fdip
->lock
);
1333 hammer_lock_sh(&tdip
->lock
);
1335 hammer_lock_sh(&tdip
->lock
);
1336 hammer_lock_sh(&fdip
->lock
);
1342 * Cleanup and tell the kernel that the rename succeeded.
1344 hammer_done_cursor(&cursor
);
1346 cache_rename(ap
->a_fnch
, ap
->a_tnch
);
1349 hammer_unlock(&ip
->lock
);
1350 hammer_unlock(&fdip
->lock
);
1351 hammer_unlock(&tdip
->lock
);
1352 hammer_done_transaction(&trans
);
1357 * hammer_vop_nrmdir { nch, dvp, cred }
1361 hammer_vop_nrmdir(struct vop_nrmdir_args
*ap
)
1363 struct hammer_transaction trans
;
1366 hammer_start_transaction(&trans
, VTOI(ap
->a_dvp
)->hmp
);
1367 error
= hammer_dounlink(&trans
, ap
->a_nch
, ap
->a_dvp
, ap
->a_cred
, 0);
1368 hammer_done_transaction(&trans
);
1374 * hammer_vop_setattr { vp, vap, cred }
1378 hammer_vop_setattr(struct vop_setattr_args
*ap
)
1380 struct hammer_transaction trans
;
1382 struct hammer_inode
*ip
;
1391 ip
= ap
->a_vp
->v_data
;
1394 if (ap
->a_vp
->v_mount
->mnt_flag
& MNT_RDONLY
)
1396 if (ip
->flags
& HAMMER_INODE_RO
)
1399 hammer_start_transaction(&trans
, ip
->hmp
);
1400 hammer_lock_sh(&ip
->lock
);
1403 if (vap
->va_flags
!= VNOVAL
) {
1404 flags
= ip
->ino_data
.uflags
;
1405 error
= vop_helper_setattr_flags(&flags
, vap
->va_flags
,
1406 hammer_to_unix_xid(&ip
->ino_data
.uid
),
1409 if (ip
->ino_data
.uflags
!= flags
) {
1410 ip
->ino_data
.uflags
= flags
;
1411 modflags
|= HAMMER_INODE_DDIRTY
;
1413 if (ip
->ino_data
.uflags
& (IMMUTABLE
| APPEND
)) {
1420 if (ip
->ino_data
.uflags
& (IMMUTABLE
| APPEND
)) {
1424 if (vap
->va_uid
!= (uid_t
)VNOVAL
) {
1425 hammer_guid_to_uuid(&uuid
, vap
->va_uid
);
1426 if (bcmp(&uuid
, &ip
->ino_data
.uid
, sizeof(uuid
)) != 0) {
1427 ip
->ino_data
.uid
= uuid
;
1428 modflags
|= HAMMER_INODE_DDIRTY
;
1431 if (vap
->va_gid
!= (uid_t
)VNOVAL
) {
1432 hammer_guid_to_uuid(&uuid
, vap
->va_gid
);
1433 if (bcmp(&uuid
, &ip
->ino_data
.gid
, sizeof(uuid
)) != 0) {
1434 ip
->ino_data
.gid
= uuid
;
1435 modflags
|= HAMMER_INODE_DDIRTY
;
1438 while (vap
->va_size
!= VNOVAL
&& ip
->ino_rec
.ino_size
!= vap
->va_size
) {
1439 switch(ap
->a_vp
->v_type
) {
1441 if (vap
->va_size
== ip
->ino_rec
.ino_size
)
1444 * XXX break atomicy, we can deadlock the backend
1445 * if we do not release the lock. Probably not a
1448 hammer_unlock(&ip
->lock
);
1449 if (vap
->va_size
< ip
->ino_rec
.ino_size
) {
1450 vtruncbuf(ap
->a_vp
, vap
->va_size
,
1454 vnode_pager_setsize(ap
->a_vp
, vap
->va_size
);
1457 hammer_lock_sh(&ip
->lock
);
1458 ip
->ino_rec
.ino_size
= vap
->va_size
;
1459 modflags
|= HAMMER_INODE_RDIRTY
;
1460 aligned_size
= (vap
->va_size
+ HAMMER_BUFMASK
) &
1464 * on-media truncation is cached in the inode until
1465 * the inode is synchronized.
1468 if ((ip
->flags
& HAMMER_INODE_TRUNCATED
) == 0) {
1469 ip
->flags
|= HAMMER_INODE_TRUNCATED
;
1470 ip
->trunc_off
= vap
->va_size
;
1471 } else if (ip
->trunc_off
> vap
->va_size
) {
1472 ip
->trunc_off
= vap
->va_size
;
1477 * If truncating we have to clean out a portion of
1478 * the last block on-disk. We do this in the
1479 * front-end buffer cache.
1481 if (truncating
&& vap
->va_size
< aligned_size
) {
1485 offset
= vap
->va_size
& HAMMER_BUFMASK
;
1486 error
= bread(ap
->a_vp
,
1487 aligned_size
- HAMMER_BUFSIZE
,
1488 HAMMER_BUFSIZE
, &bp
);
1490 bzero(bp
->b_data
+ offset
,
1491 HAMMER_BUFSIZE
- offset
);
1499 if ((ip
->flags
& HAMMER_INODE_TRUNCATED
) == 0) {
1500 ip
->flags
|= HAMMER_INODE_TRUNCATED
;
1501 ip
->trunc_off
= vap
->va_size
;
1502 } else if (ip
->trunc_off
> vap
->va_size
) {
1503 ip
->trunc_off
= vap
->va_size
;
1505 ip
->ino_rec
.ino_size
= vap
->va_size
;
1506 modflags
|= HAMMER_INODE_RDIRTY
;
1514 if (vap
->va_atime
.tv_sec
!= VNOVAL
) {
1515 ip
->ino_rec
.ino_atime
=
1516 hammer_timespec_to_transid(&vap
->va_atime
);
1517 modflags
|= HAMMER_INODE_ITIMES
;
1519 if (vap
->va_mtime
.tv_sec
!= VNOVAL
) {
1520 ip
->ino_rec
.ino_mtime
=
1521 hammer_timespec_to_transid(&vap
->va_mtime
);
1522 modflags
|= HAMMER_INODE_ITIMES
;
1524 if (vap
->va_mode
!= (mode_t
)VNOVAL
) {
1525 if (ip
->ino_data
.mode
!= vap
->va_mode
) {
1526 ip
->ino_data
.mode
= vap
->va_mode
;
1527 modflags
|= HAMMER_INODE_DDIRTY
;
1532 hammer_modify_inode(&trans
, ip
, modflags
);
1533 hammer_unlock(&ip
->lock
);
1534 hammer_done_transaction(&trans
);
1539 * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1543 hammer_vop_nsymlink(struct vop_nsymlink_args
*ap
)
1545 struct hammer_transaction trans
;
1546 struct hammer_inode
*dip
;
1547 struct hammer_inode
*nip
;
1548 struct nchandle
*nch
;
1549 hammer_record_t record
;
1553 ap
->a_vap
->va_type
= VLNK
;
1556 dip
= VTOI(ap
->a_dvp
);
1558 if (dip
->flags
& HAMMER_INODE_RO
)
1562 * Create a transaction to cover the operations we perform.
1564 hammer_start_transaction(&trans
, dip
->hmp
);
1567 * Create a new filesystem object of the requested type. The
1568 * returned inode will be referenced but not locked.
1571 error
= hammer_create_inode(&trans
, ap
->a_vap
, ap
->a_cred
, dip
, &nip
);
1573 hammer_done_transaction(&trans
);
1579 * Add the new filesystem object to the directory. This will also
1580 * bump the inode's link count.
1582 hammer_lock_sh(&nip
->lock
);
1583 hammer_lock_sh(&dip
->lock
);
1586 * Add a record representing the symlink. symlink stores the link
1587 * as pure data, not a string, and is no \0 terminated.
1590 record
= hammer_alloc_mem_record(nip
);
1591 bytes
= strlen(ap
->a_target
);
1593 record
->rec
.base
.base
.key
= HAMMER_FIXKEY_SYMLINK
;
1594 record
->rec
.base
.base
.rec_type
= HAMMER_RECTYPE_FIX
;
1595 record
->rec
.base
.data_len
= bytes
;
1596 record
->data
= (void *)ap
->a_target
;
1597 /* will be reallocated by routine below */
1598 error
= hammer_ip_add_record(&trans
, record
);
1601 * Set the file size to the length of the link.
1604 nip
->ino_rec
.ino_size
= bytes
;
1605 hammer_modify_inode(&trans
, nip
, HAMMER_INODE_RDIRTY
);
1609 error
= hammer_ip_add_directory(&trans
, dip
, nch
->ncp
, nip
);
1610 hammer_unlock(&dip
->lock
);
1611 hammer_unlock(&nip
->lock
);
1617 hammer_rel_inode(nip
, 0);
1620 error
= hammer_get_vnode(nip
, LK_EXCLUSIVE
, ap
->a_vpp
);
1621 hammer_rel_inode(nip
, 0);
1623 cache_setunresolved(ap
->a_nch
);
1624 cache_setvp(ap
->a_nch
, *ap
->a_vpp
);
1627 hammer_done_transaction(&trans
);
1632 * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1636 hammer_vop_nwhiteout(struct vop_nwhiteout_args
*ap
)
1638 struct hammer_transaction trans
;
1641 hammer_start_transaction(&trans
, VTOI(ap
->a_dvp
)->hmp
);
1642 error
= hammer_dounlink(&trans
, ap
->a_nch
, ap
->a_dvp
,
1643 ap
->a_cred
, ap
->a_flags
);
1644 hammer_done_transaction(&trans
);
1650 * hammer_vop_ioctl { vp, command, data, fflag, cred }
1654 hammer_vop_ioctl(struct vop_ioctl_args
*ap
)
1656 struct hammer_inode
*ip
= ap
->a_vp
->v_data
;
1658 return(hammer_ioctl(ip
, ap
->a_command
, ap
->a_data
,
1659 ap
->a_fflag
, ap
->a_cred
));
1664 hammer_vop_mountctl(struct vop_mountctl_args
*ap
)
1669 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
1672 case MOUNTCTL_SET_EXPORT
:
1673 if (ap
->a_ctllen
!= sizeof(struct export_args
))
1675 error
= hammer_vfs_export(mp
, ap
->a_op
,
1676 (const struct export_args
*)ap
->a_ctl
);
1679 error
= journal_mountctl(ap
);
1686 * hammer_vop_strategy { vp, bio }
1688 * Strategy call, used for regular file read & write only. Note that the
1689 * bp may represent a cluster.
1691 * To simplify operation and allow better optimizations in the future,
1692 * this code does not make any assumptions with regards to buffer alignment
1697 hammer_vop_strategy(struct vop_strategy_args
*ap
)
1702 bp
= ap
->a_bio
->bio_buf
;
1706 error
= hammer_vop_strategy_read(ap
);
1709 error
= hammer_vop_strategy_write(ap
);
1712 bp
->b_error
= error
= EINVAL
;
1713 bp
->b_flags
|= B_ERROR
;
1721 * Read from a regular file. Iterate the related records and fill in the
1722 * BIO/BUF. Gaps are zero-filled.
1724 * The support code in hammer_object.c should be used to deal with mixed
1725 * in-memory and on-disk records.
1731 hammer_vop_strategy_read(struct vop_strategy_args
*ap
)
1733 struct hammer_transaction trans
;
1734 struct hammer_inode
*ip
;
1735 struct hammer_cursor cursor
;
1736 hammer_record_ondisk_t rec
;
1737 hammer_base_elm_t base
;
1750 ip
= ap
->a_vp
->v_data
;
1752 hammer_simple_transaction(&trans
, ip
->hmp
);
1753 hammer_init_cursor(&trans
, &cursor
, &ip
->cache
[0]);
1756 * Key range (begin and end inclusive) to scan. Note that the key's
1757 * stored in the actual records represent BASE+LEN, not BASE. The
1758 * first record containing bio_offset will have a key > bio_offset.
1760 cursor
.key_beg
.obj_id
= ip
->obj_id
;
1761 cursor
.key_beg
.create_tid
= 0;
1762 cursor
.key_beg
.delete_tid
= 0;
1763 cursor
.key_beg
.obj_type
= 0;
1764 cursor
.key_beg
.key
= bio
->bio_offset
+ 1;
1765 cursor
.asof
= ip
->obj_asof
;
1766 cursor
.flags
|= HAMMER_CURSOR_ASOF
| HAMMER_CURSOR_DATAEXTOK
;
1768 cursor
.key_end
= cursor
.key_beg
;
1769 KKASSERT(ip
->ino_rec
.base
.base
.obj_type
== HAMMER_OBJTYPE_REGFILE
);
1771 if (ip
->ino_rec
.base
.base
.obj_type
== HAMMER_OBJTYPE_DBFILE
) {
1772 cursor
.key_beg
.rec_type
= HAMMER_RECTYPE_DB
;
1773 cursor
.key_end
.rec_type
= HAMMER_RECTYPE_DB
;
1774 cursor
.key_end
.key
= 0x7FFFFFFFFFFFFFFFLL
;
1778 ran_end
= bio
->bio_offset
+ bp
->b_bufsize
;
1779 cursor
.key_beg
.rec_type
= HAMMER_RECTYPE_DATA
;
1780 cursor
.key_end
.rec_type
= HAMMER_RECTYPE_DATA
;
1781 tmp64
= ran_end
+ MAXPHYS
+ 1; /* work-around GCC-4 bug */
1782 if (tmp64
< ran_end
)
1783 cursor
.key_end
.key
= 0x7FFFFFFFFFFFFFFFLL
;
1785 cursor
.key_end
.key
= ran_end
+ MAXPHYS
+ 1;
1787 cursor
.flags
|= HAMMER_CURSOR_END_INCLUSIVE
;
1789 error
= hammer_ip_first(&cursor
, ip
);
1792 while (error
== 0) {
1793 error
= hammer_ip_resolve_data(&cursor
);
1796 rec
= cursor
.record
;
1797 base
= &rec
->base
.base
;
1799 rec_offset
= base
->key
- rec
->data
.base
.data_len
;
1802 * Calculate the gap, if any, and zero-fill it.
1804 n
= (int)(rec_offset
- (bio
->bio_offset
+ boff
));
1806 if (n
> bp
->b_bufsize
- boff
)
1807 n
= bp
->b_bufsize
- boff
;
1808 bzero((char *)bp
->b_data
+ boff
, n
);
1814 * Calculate the data offset in the record and the number
1815 * of bytes we can copy.
1817 * Note there is a degenerate case here where boff may
1818 * already be at bp->b_bufsize.
1822 n
= rec
->data
.base
.data_len
- roff
;
1824 if (n
> bp
->b_bufsize
- boff
)
1825 n
= bp
->b_bufsize
- boff
;
1828 * If we cached a truncation point on our front-end the
1829 * on-disk version may still have physical records beyond
1830 * that point. Truncate visibility.
1832 if (ip
->trunc_off
<= rec_offset
)
1834 else if (ip
->trunc_off
< rec_offset
+ n
)
1835 n
= (int)(ip
->trunc_off
- rec_offset
);
1841 bcopy((char *)cursor
.data
+ roff
,
1842 (char *)bp
->b_data
+ boff
, n
);
1845 if (boff
== bp
->b_bufsize
)
1847 error
= hammer_ip_next(&cursor
);
1849 hammer_done_cursor(&cursor
);
1850 hammer_done_transaction(&trans
);
1853 * There may have been a gap after the last record
1855 if (error
== ENOENT
)
1857 if (error
== 0 && boff
!= bp
->b_bufsize
) {
1858 KKASSERT(boff
< bp
->b_bufsize
);
1859 bzero((char *)bp
->b_data
+ boff
, bp
->b_bufsize
- boff
);
1860 /* boff = bp->b_bufsize; */
1863 bp
->b_error
= error
;
1865 bp
->b_flags
|= B_ERROR
;
1871 * Write to a regular file. Because this is a strategy call the OS is
1872 * trying to actually sync data to the media. HAMMER can only flush
1873 * the entire inode (so the TID remains properly synchronized).
1875 * Basically all we do here is place the bio on the inode's flush queue
1876 * and activate the flusher.
1880 hammer_vop_strategy_write(struct vop_strategy_args
*ap
)
1888 ip
= ap
->a_vp
->v_data
;
1890 if (ip
->flags
& HAMMER_INODE_RO
) {
1891 bp
->b_error
= EROFS
;
1892 bp
->b_flags
|= B_ERROR
;
1898 * If the inode is being flushed we cannot re-queue buffers
1899 * it may have already flushed, or it could result in duplicate
1900 * records in the database.
1903 if (ip
->flags
& HAMMER_INODE_WRITE_ALT
)
1904 TAILQ_INSERT_TAIL(&ip
->bio_alt_list
, bio
, bio_act
);
1906 TAILQ_INSERT_TAIL(&ip
->bio_list
, bio
, bio_act
);
1908 hammer_modify_inode(NULL
, ip
, HAMMER_INODE_BUFS
);
1909 hammer_flush_inode(ip
, HAMMER_FLUSH_FORCE
|HAMMER_FLUSH_SIGNAL
);
1914 * Backend code which actually performs the write to the media. This
1915 * routine is typically called from the flusher. The bio will be disposed
1916 * of (biodone'd) by this routine.
1918 * Iterate the related records and mark for deletion. If existing edge
1919 * records (left and right side) overlap our write they have to be marked
1920 * deleted and new records created, usually referencing a portion of the
1921 * original data. Then add a record to represent the buffer.
1924 hammer_dowrite(hammer_transaction_t trans
, hammer_inode_t ip
, struct bio
*bio
)
1926 struct buf
*bp
= bio
->bio_buf
;
1929 KKASSERT(ip
->flush_state
== HAMMER_FST_FLUSH
);
1932 * If the inode is going or gone, just throw away any frontend
1935 if (ip
->flags
& HAMMER_INODE_DELETED
) {
1941 * Delete any records overlapping our range. This function will
1942 * (eventually) properly truncate partial overlaps.
1944 if (ip
->sync_ino_rec
.base
.base
.obj_type
== HAMMER_OBJTYPE_DBFILE
) {
1945 error
= hammer_ip_delete_range(trans
, ip
, bio
->bio_offset
,
1948 error
= hammer_ip_delete_range(trans
, ip
, bio
->bio_offset
,
1954 * Add a single record to cover the write. We can write a record
1955 * with only the actual file data - for example, a small 200 byte
1956 * file does not have to write out a 16K record.
1958 * While the data size does not have to be aligned, we still do it
1959 * to reduce fragmentation in a future allocation model.
1964 if (ip
->sync_ino_rec
.ino_size
- bio
->bio_offset
>
1966 limit_size
= bp
->b_bufsize
;
1968 limit_size
= (int)(ip
->sync_ino_rec
.ino_size
-
1970 KKASSERT(limit_size
>= 0);
1971 limit_size
= (limit_size
+ 63) & ~63;
1974 error
= hammer_ip_sync_data(trans
, ip
, bio
->bio_offset
,
1975 bp
->b_data
, limit_size
);
1979 Debugger("hammer_dowrite: error");
1982 bp
->b_resid
= bp
->b_bufsize
;
1983 bp
->b_error
= error
;
1984 bp
->b_flags
|= B_ERROR
;
1994 * dounlink - disconnect a directory entry
1996 * XXX whiteout support not really in yet
1999 hammer_dounlink(hammer_transaction_t trans
, struct nchandle
*nch
,
2000 struct vnode
*dvp
, struct ucred
*cred
, int flags
)
2002 struct namecache
*ncp
;
2005 hammer_record_ondisk_t rec
;
2006 struct hammer_cursor cursor
;
2011 * Calculate the namekey and setup the key range for the scan. This
2012 * works kinda like a chained hash table where the lower 32 bits
2013 * of the namekey synthesize the chain.
2015 * The key range is inclusive of both key_beg and key_end.
2020 if (dip
->flags
& HAMMER_INODE_RO
)
2023 namekey
= hammer_directory_namekey(ncp
->nc_name
, ncp
->nc_nlen
);
2025 hammer_init_cursor(trans
, &cursor
, &dip
->cache
[0]);
2026 cursor
.key_beg
.obj_id
= dip
->obj_id
;
2027 cursor
.key_beg
.key
= namekey
;
2028 cursor
.key_beg
.create_tid
= 0;
2029 cursor
.key_beg
.delete_tid
= 0;
2030 cursor
.key_beg
.rec_type
= HAMMER_RECTYPE_DIRENTRY
;
2031 cursor
.key_beg
.obj_type
= 0;
2033 cursor
.key_end
= cursor
.key_beg
;
2034 cursor
.key_end
.key
|= 0xFFFFFFFFULL
;
2035 cursor
.asof
= dip
->obj_asof
;
2036 cursor
.flags
|= HAMMER_CURSOR_END_INCLUSIVE
| HAMMER_CURSOR_ASOF
;
2039 * Scan all matching records (the chain), locate the one matching
2040 * the requested path component. info->last_error contains the
2041 * error code on search termination and could be 0, ENOENT, or
2044 * The hammer_ip_*() functions merge in-memory records with on-disk
2045 * records for the purposes of the search.
2047 error
= hammer_ip_first(&cursor
, dip
);
2048 while (error
== 0) {
2049 error
= hammer_ip_resolve_data(&cursor
);
2052 rec
= cursor
.record
;
2053 if (ncp
->nc_nlen
== rec
->entry
.base
.data_len
&&
2054 bcmp(ncp
->nc_name
, cursor
.data
, ncp
->nc_nlen
) == 0) {
2057 error
= hammer_ip_next(&cursor
);
2061 * If all is ok we have to get the inode so we can adjust nlinks.
2063 * If the target is a directory, it must be empty.
2066 ip
= hammer_get_inode(trans
, &dip
->cache
[1],
2068 dip
->hmp
->asof
, 0, &error
);
2069 if (error
== ENOENT
) {
2070 kprintf("obj_id %016llx\n", rec
->entry
.obj_id
);
2071 Debugger("ENOENT unlinking object that should exist");
2075 * If we are trying to remove a directory the directory must
2078 * WARNING: hammer_ip_check_directory_empty() may have to
2079 * terminate the cursor to avoid a deadlock. It is ok to
2080 * call hammer_done_cursor() twice.
2082 if (error
== 0 && ip
->ino_rec
.base
.base
.obj_type
==
2083 HAMMER_OBJTYPE_DIRECTORY
) {
2084 error
= hammer_ip_check_directory_empty(trans
, &cursor
,
2089 * Delete the directory entry.
2091 * WARNING: hammer_ip_del_directory() may have to terminate
2092 * the cursor to avoid a deadlock. It is ok to call
2093 * hammer_done_cursor() twice.
2096 hammer_lock_sh(&ip
->lock
);
2097 hammer_lock_sh(&dip
->lock
);
2098 error
= hammer_ip_del_directory(trans
, &cursor
,
2100 hammer_unlock(&dip
->lock
);
2101 hammer_unlock(&ip
->lock
);
2104 cache_setunresolved(nch
);
2105 cache_setvp(nch
, NULL
);
2108 cache_inval_vp(ip
->vp
, CINV_DESTROY
);
2110 hammer_rel_inode(ip
, 0);
2112 hammer_done_cursor(&cursor
);
2113 if (error
== EDEADLK
)
2119 /************************************************************************
2120 * FIFO AND SPECFS OPS *
2121 ************************************************************************
2126 hammer_vop_fifoclose (struct vop_close_args
*ap
)
2128 /* XXX update itimes */
2129 return (VOCALL(&fifo_vnode_vops
, &ap
->a_head
));
2133 hammer_vop_fiforead (struct vop_read_args
*ap
)
2137 error
= VOCALL(&fifo_vnode_vops
, &ap
->a_head
);
2138 /* XXX update access time */
2143 hammer_vop_fifowrite (struct vop_write_args
*ap
)
2147 error
= VOCALL(&fifo_vnode_vops
, &ap
->a_head
);
2148 /* XXX update access time */
2153 hammer_vop_specclose (struct vop_close_args
*ap
)
2155 /* XXX update itimes */
2156 return (VOCALL(&spec_vnode_vops
, &ap
->a_head
));
2160 hammer_vop_specread (struct vop_read_args
*ap
)
2162 /* XXX update access time */
2163 return (VOCALL(&spec_vnode_vops
, &ap
->a_head
));
2167 hammer_vop_specwrite (struct vop_write_args
*ap
)
2169 /* XXX update last change time */
2170 return (VOCALL(&spec_vnode_vops
, &ap
->a_head
));