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 1.78 2008/06/28 18:10:55 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_bmap(struct vop_bmap_args
*ap
);
79 static int hammer_vop_nsymlink(struct vop_nsymlink_args
*);
80 static int hammer_vop_nwhiteout(struct vop_nwhiteout_args
*);
81 static int hammer_vop_ioctl(struct vop_ioctl_args
*);
82 static int hammer_vop_mountctl(struct vop_mountctl_args
*);
84 static int hammer_vop_fifoclose (struct vop_close_args
*);
85 static int hammer_vop_fiforead (struct vop_read_args
*);
86 static int hammer_vop_fifowrite (struct vop_write_args
*);
88 static int hammer_vop_specclose (struct vop_close_args
*);
89 static int hammer_vop_specread (struct vop_read_args
*);
90 static int hammer_vop_specwrite (struct vop_write_args
*);
92 struct vop_ops hammer_vnode_vops
= {
93 .vop_default
= vop_defaultop
,
94 .vop_fsync
= hammer_vop_fsync
,
95 .vop_getpages
= vop_stdgetpages
,
96 .vop_putpages
= vop_stdputpages
,
97 .vop_read
= hammer_vop_read
,
98 .vop_write
= hammer_vop_write
,
99 .vop_access
= hammer_vop_access
,
100 .vop_advlock
= hammer_vop_advlock
,
101 .vop_close
= hammer_vop_close
,
102 .vop_ncreate
= hammer_vop_ncreate
,
103 .vop_getattr
= hammer_vop_getattr
,
104 .vop_inactive
= hammer_vop_inactive
,
105 .vop_reclaim
= hammer_vop_reclaim
,
106 .vop_nresolve
= hammer_vop_nresolve
,
107 .vop_nlookupdotdot
= hammer_vop_nlookupdotdot
,
108 .vop_nlink
= hammer_vop_nlink
,
109 .vop_nmkdir
= hammer_vop_nmkdir
,
110 .vop_nmknod
= hammer_vop_nmknod
,
111 .vop_open
= hammer_vop_open
,
112 .vop_pathconf
= hammer_vop_pathconf
,
113 .vop_print
= hammer_vop_print
,
114 .vop_readdir
= hammer_vop_readdir
,
115 .vop_readlink
= hammer_vop_readlink
,
116 .vop_nremove
= hammer_vop_nremove
,
117 .vop_nrename
= hammer_vop_nrename
,
118 .vop_nrmdir
= hammer_vop_nrmdir
,
119 .vop_setattr
= hammer_vop_setattr
,
120 .vop_bmap
= hammer_vop_bmap
,
121 .vop_strategy
= hammer_vop_strategy
,
122 .vop_nsymlink
= hammer_vop_nsymlink
,
123 .vop_nwhiteout
= hammer_vop_nwhiteout
,
124 .vop_ioctl
= hammer_vop_ioctl
,
125 .vop_mountctl
= hammer_vop_mountctl
128 struct vop_ops hammer_spec_vops
= {
129 .vop_default
= spec_vnoperate
,
130 .vop_fsync
= hammer_vop_fsync
,
131 .vop_read
= hammer_vop_specread
,
132 .vop_write
= hammer_vop_specwrite
,
133 .vop_access
= hammer_vop_access
,
134 .vop_close
= hammer_vop_specclose
,
135 .vop_getattr
= hammer_vop_getattr
,
136 .vop_inactive
= hammer_vop_inactive
,
137 .vop_reclaim
= hammer_vop_reclaim
,
138 .vop_setattr
= hammer_vop_setattr
141 struct vop_ops hammer_fifo_vops
= {
142 .vop_default
= fifo_vnoperate
,
143 .vop_fsync
= hammer_vop_fsync
,
144 .vop_read
= hammer_vop_fiforead
,
145 .vop_write
= hammer_vop_fifowrite
,
146 .vop_access
= hammer_vop_access
,
147 .vop_close
= hammer_vop_fifoclose
,
148 .vop_getattr
= hammer_vop_getattr
,
149 .vop_inactive
= hammer_vop_inactive
,
150 .vop_reclaim
= hammer_vop_reclaim
,
151 .vop_setattr
= hammer_vop_setattr
154 #ifdef DEBUG_TRUNCATE
155 struct hammer_inode
*HammerTruncIp
;
158 static int hammer_dounlink(hammer_transaction_t trans
, struct nchandle
*nch
,
159 struct vnode
*dvp
, struct ucred
*cred
, int flags
);
160 static int hammer_vop_strategy_read(struct vop_strategy_args
*ap
);
161 static int hammer_vop_strategy_write(struct vop_strategy_args
*ap
);
162 static void hammer_cleanup_write_io(hammer_inode_t ip
);
163 static void hammer_update_rsv_databufs(hammer_inode_t ip
);
168 hammer_vop_vnoperate(struct vop_generic_args
*)
170 return (VOCALL(&hammer_vnode_vops
, ap
));
175 * hammer_vop_fsync { vp, waitfor }
177 * fsync() an inode to disk and wait for it to be completely committed
178 * such that the information would not be undone if a crash occured after
183 hammer_vop_fsync(struct vop_fsync_args
*ap
)
185 hammer_inode_t ip
= VTOI(ap
->a_vp
);
187 vfsync(ap
->a_vp
, ap
->a_waitfor
, 1, NULL
, NULL
);
188 hammer_flush_inode(ip
, HAMMER_FLUSH_SIGNAL
);
189 if (ap
->a_waitfor
== MNT_WAIT
)
190 hammer_wait_inode(ip
);
195 * hammer_vop_read { vp, uio, ioflag, cred }
199 hammer_vop_read(struct vop_read_args
*ap
)
201 struct hammer_transaction trans
;
212 if (ap
->a_vp
->v_type
!= VREG
)
219 * Allow the UIO's size to override the sequential heuristic.
221 blksize
= hammer_blocksize(uio
->uio_offset
);
222 seqcount
= (uio
->uio_resid
+ (blksize
- 1)) / blksize
;
223 ioseqcount
= ap
->a_ioflag
>> 16;
224 if (seqcount
< ioseqcount
)
225 seqcount
= ioseqcount
;
227 hammer_start_transaction(&trans
, ip
->hmp
);
230 * Access the data typically in HAMMER_BUFSIZE blocks via the
231 * buffer cache, but HAMMER may use a variable block size based
234 while (uio
->uio_resid
> 0 && uio
->uio_offset
< ip
->ino_data
.size
) {
238 blksize
= hammer_blocksize(uio
->uio_offset
);
239 offset
= (int)uio
->uio_offset
& (blksize
- 1);
240 base_offset
= uio
->uio_offset
- offset
;
242 if (hammer_debug_cluster_enable
) {
244 * Use file_limit to prevent cluster_read() from
245 * creating buffers of the wrong block size past
248 file_limit
= ip
->ino_data
.size
;
249 if (base_offset
< HAMMER_XDEMARC
&&
250 file_limit
> HAMMER_XDEMARC
) {
251 file_limit
= HAMMER_XDEMARC
;
253 error
= cluster_read(ap
->a_vp
,
254 file_limit
, base_offset
,
258 error
= bread(ap
->a_vp
, base_offset
, blksize
, &bp
);
261 kprintf("error %d\n", error
);
266 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
267 n
= blksize
- offset
;
268 if (n
> uio
->uio_resid
)
270 if (n
> ip
->ino_data
.size
- uio
->uio_offset
)
271 n
= (int)(ip
->ino_data
.size
- uio
->uio_offset
);
272 error
= uiomove((char *)bp
->b_data
+ offset
, n
, uio
);
274 /* data has a lower priority then meta-data */
275 bp
->b_flags
|= B_AGE
;
280 if ((ip
->flags
& HAMMER_INODE_RO
) == 0 &&
281 (ip
->hmp
->mp
->mnt_flag
& MNT_NOATIME
) == 0) {
282 ip
->ino_data
.atime
= trans
.time
;
283 hammer_modify_inode(ip
, HAMMER_INODE_ATIME
);
285 hammer_done_transaction(&trans
);
290 * hammer_vop_write { vp, uio, ioflag, cred }
294 hammer_vop_write(struct vop_write_args
*ap
)
296 struct hammer_transaction trans
;
297 struct hammer_inode
*ip
;
310 if (ap
->a_vp
->v_type
!= VREG
)
315 seqcount
= ap
->a_ioflag
>> 16;
317 if (ip
->flags
& HAMMER_INODE_RO
)
321 * Create a transaction to cover the operations we perform.
323 hammer_start_transaction(&trans
, hmp
);
329 if (ap
->a_ioflag
& IO_APPEND
)
330 uio
->uio_offset
= ip
->ino_data
.size
;
333 * Check for illegal write offsets. Valid range is 0...2^63-1.
335 * NOTE: the base_off assignment is required to work around what
336 * I consider to be a GCC-4 optimization bug.
338 if (uio
->uio_offset
< 0) {
339 hammer_done_transaction(&trans
);
342 base_offset
= uio
->uio_offset
+ uio
->uio_resid
; /* work around gcc-4 */
343 if (uio
->uio_resid
> 0 && base_offset
<= 0) {
344 hammer_done_transaction(&trans
);
349 * Access the data typically in HAMMER_BUFSIZE blocks via the
350 * buffer cache, but HAMMER may use a variable block size based
354 while (uio
->uio_resid
> 0) {
359 if ((error
= hammer_checkspace(hmp
)) != 0)
362 blksize
= hammer_blocksize(uio
->uio_offset
);
365 * Do not allow HAMMER to blow out the buffer cache. Very
366 * large UIOs can lockout other processes due to bwillwrite()
369 * Do not allow HAMMER to blow out system memory by
370 * accumulating too many records. Records are so well
371 * decoupled from the buffer cache that it is possible
372 * for userland to push data out to the media via
373 * direct-write, but build up the records queued to the
374 * backend faster then the backend can flush them out.
375 * HAMMER has hit its write limit but the frontend has
376 * no pushback to slow it down.
378 * Always check at the beginning so separate writes are
379 * not able to bypass this code (count++).
381 * WARNING: Cannot unlock vp when doing a NOCOPY write as
382 * part of a putpages operation. Doing so could cause us
383 * to deadlock against the VM system when we try to re-lock.
385 if ((count
++ & 15) == 0 || count
> 64) {
389 if (uio
->uio_segflg
!= UIO_NOCOPY
) {
391 if ((ap
->a_ioflag
& IO_NOBWILL
) == 0)
396 * Pending record flush check.
398 if (hmp
->rsv_recs
> hammer_limit_recs
/ 2) {
400 * Get the inode on the flush list
402 if (ip
->rsv_recs
>= 64)
403 hammer_flush_inode(ip
, HAMMER_FLUSH_SIGNAL
);
404 else if (ip
->rsv_recs
>= 16)
405 hammer_flush_inode(ip
, 0);
408 * Keep the flusher going if the system keeps
411 delta
= hmp
->count_newrecords
-
412 hmp
->last_newrecords
;
413 if (delta
< 0 || delta
> hammer_limit_recs
/ 2) {
414 hmp
->last_newrecords
= hmp
->count_newrecords
;
415 hammer_sync_hmp(hmp
, MNT_NOWAIT
);
419 * If we have gotten behind start slowing
422 delta
= (hmp
->rsv_recs
- hammer_limit_recs
) *
423 hz
/ hammer_limit_recs
;
425 tsleep(&trans
, 0, "hmrslo", delta
);
428 if (uio
->uio_segflg
!= UIO_NOCOPY
)
429 vn_lock(ap
->a_vp
, LK_EXCLUSIVE
|LK_RETRY
);
433 * Calculate the blocksize at the current offset and figure
434 * out how much we can actually write.
436 blkmask
= blksize
- 1;
437 offset
= (int)uio
->uio_offset
& blkmask
;
438 base_offset
= uio
->uio_offset
& ~(int64_t)blkmask
;
439 n
= blksize
- offset
;
440 if (n
> uio
->uio_resid
)
442 if (uio
->uio_offset
+ n
> ip
->ino_data
.size
) {
443 vnode_pager_setsize(ap
->a_vp
, uio
->uio_offset
+ n
);
447 if (uio
->uio_segflg
== UIO_NOCOPY
) {
449 * Issuing a write with the same data backing the
450 * buffer. Instantiate the buffer to collect the
451 * backing vm pages, then read-in any missing bits.
453 * This case is used by vop_stdputpages().
455 bp
= getblk(ap
->a_vp
, base_offset
,
456 blksize
, GETBLK_BHEAVY
, 0);
457 if ((bp
->b_flags
& B_CACHE
) == 0) {
459 error
= bread(ap
->a_vp
, base_offset
,
462 } else if (offset
== 0 && uio
->uio_resid
>= blksize
) {
464 * Even though we are entirely overwriting the buffer
465 * we may still have to zero it out to avoid a
466 * mmap/write visibility issue.
468 bp
= getblk(ap
->a_vp
, base_offset
, blksize
, GETBLK_BHEAVY
, 0);
469 if ((bp
->b_flags
& B_CACHE
) == 0)
471 } else if (base_offset
>= ip
->ino_data
.size
) {
473 * If the base offset of the buffer is beyond the
474 * file EOF, we don't have to issue a read.
476 bp
= getblk(ap
->a_vp
, base_offset
,
477 blksize
, GETBLK_BHEAVY
, 0);
481 * Partial overwrite, read in any missing bits then
482 * replace the portion being written.
484 error
= bread(ap
->a_vp
, base_offset
, blksize
, &bp
);
489 error
= uiomove((char *)bp
->b_data
+ offset
,
494 * If we screwed up we have to undo any VM size changes we
500 vtruncbuf(ap
->a_vp
, ip
->ino_data
.size
,
501 hammer_blocksize(ip
->ino_data
.size
));
505 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
506 if (ip
->ino_data
.size
< uio
->uio_offset
) {
507 ip
->ino_data
.size
= uio
->uio_offset
;
508 flags
= HAMMER_INODE_DDIRTY
;
509 vnode_pager_setsize(ap
->a_vp
, ip
->ino_data
.size
);
513 ip
->ino_data
.mtime
= trans
.time
;
514 flags
|= HAMMER_INODE_MTIME
| HAMMER_INODE_BUFS
;
515 hammer_modify_inode(ip
, flags
);
518 * Try to keep track of cached dirty data.
520 if ((bp
->b_flags
& B_DIRTY
) == 0) {
526 * Final buffer disposition.
529 bp
->b_flags
|= B_AGE
;
530 if (ap
->a_ioflag
& IO_SYNC
) {
532 } else if (ap
->a_ioflag
& IO_DIRECT
) {
538 hammer_done_transaction(&trans
);
543 * hammer_vop_access { vp, mode, cred }
547 hammer_vop_access(struct vop_access_args
*ap
)
549 struct hammer_inode
*ip
= VTOI(ap
->a_vp
);
554 uid
= hammer_to_unix_xid(&ip
->ino_data
.uid
);
555 gid
= hammer_to_unix_xid(&ip
->ino_data
.gid
);
557 error
= vop_helper_access(ap
, uid
, gid
, ip
->ino_data
.mode
,
558 ip
->ino_data
.uflags
);
563 * hammer_vop_advlock { vp, id, op, fl, flags }
567 hammer_vop_advlock(struct vop_advlock_args
*ap
)
569 hammer_inode_t ip
= VTOI(ap
->a_vp
);
571 return (lf_advlock(ap
, &ip
->advlock
, ip
->ino_data
.size
));
575 * hammer_vop_close { vp, fflag }
579 hammer_vop_close(struct vop_close_args
*ap
)
581 hammer_inode_t ip
= VTOI(ap
->a_vp
);
583 if ((ip
->flags
| ip
->sync_flags
) & HAMMER_INODE_MODMASK
)
584 hammer_inode_waitreclaims(ip
->hmp
);
585 return (vop_stdclose(ap
));
589 * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
591 * The operating system has already ensured that the directory entry
592 * does not exist and done all appropriate namespace locking.
596 hammer_vop_ncreate(struct vop_ncreate_args
*ap
)
598 struct hammer_transaction trans
;
599 struct hammer_inode
*dip
;
600 struct hammer_inode
*nip
;
601 struct nchandle
*nch
;
605 dip
= VTOI(ap
->a_dvp
);
607 if (dip
->flags
& HAMMER_INODE_RO
)
609 if ((error
= hammer_checkspace(dip
->hmp
)) != 0)
613 * Create a transaction to cover the operations we perform.
615 hammer_start_transaction(&trans
, dip
->hmp
);
618 * Create a new filesystem object of the requested type. The
619 * returned inode will be referenced and shared-locked to prevent
620 * it from being moved to the flusher.
623 error
= hammer_create_inode(&trans
, ap
->a_vap
, ap
->a_cred
,
626 hkprintf("hammer_create_inode error %d\n", error
);
627 hammer_done_transaction(&trans
);
633 * Add the new filesystem object to the directory. This will also
634 * bump the inode's link count.
636 error
= hammer_ip_add_directory(&trans
, dip
,
637 nch
->ncp
->nc_name
, nch
->ncp
->nc_nlen
,
640 hkprintf("hammer_ip_add_directory error %d\n", error
);
646 hammer_rel_inode(nip
, 0);
647 hammer_done_transaction(&trans
);
650 error
= hammer_get_vnode(nip
, ap
->a_vpp
);
651 hammer_done_transaction(&trans
);
652 hammer_rel_inode(nip
, 0);
654 cache_setunresolved(ap
->a_nch
);
655 cache_setvp(ap
->a_nch
, *ap
->a_vpp
);
662 * hammer_vop_getattr { vp, vap }
664 * Retrieve an inode's attribute information. When accessing inodes
665 * historically we fake the atime field to ensure consistent results.
666 * The atime field is stored in the B-Tree element and allowed to be
667 * updated without cycling the element.
671 hammer_vop_getattr(struct vop_getattr_args
*ap
)
673 struct hammer_inode
*ip
= VTOI(ap
->a_vp
);
674 struct vattr
*vap
= ap
->a_vap
;
676 vap
->va_fsid
= ip
->hmp
->fsid_udev
;
678 * XXX munge the device if we are in a pseudo-fs, so user utilities
679 * do not think its the same 'filesystem'.
681 if (ip
->obj_localization
)
682 vap
->va_fsid
+= ip
->obj_localization
;
683 vap
->va_fileid
= ip
->ino_leaf
.base
.obj_id
;
684 vap
->va_mode
= ip
->ino_data
.mode
;
685 vap
->va_nlink
= ip
->ino_data
.nlinks
;
686 vap
->va_uid
= hammer_to_unix_xid(&ip
->ino_data
.uid
);
687 vap
->va_gid
= hammer_to_unix_xid(&ip
->ino_data
.gid
);
690 vap
->va_size
= ip
->ino_data
.size
;
693 * We must provide a consistent atime and mtime for snapshots
694 * so people can do a 'tar cf - ... | md5' on them and get
695 * consistent results.
697 if (ip
->flags
& HAMMER_INODE_RO
) {
698 hammer_time_to_timespec(ip
->ino_data
.ctime
, &vap
->va_atime
);
699 hammer_time_to_timespec(ip
->ino_data
.ctime
, &vap
->va_mtime
);
701 hammer_time_to_timespec(ip
->ino_data
.atime
, &vap
->va_atime
);
702 hammer_time_to_timespec(ip
->ino_data
.mtime
, &vap
->va_mtime
);
704 hammer_time_to_timespec(ip
->ino_data
.ctime
, &vap
->va_ctime
);
705 vap
->va_flags
= ip
->ino_data
.uflags
;
706 vap
->va_gen
= 1; /* hammer inums are unique for all time */
707 vap
->va_blocksize
= HAMMER_BUFSIZE
;
708 if (ip
->ino_data
.size
>= HAMMER_XDEMARC
) {
709 vap
->va_bytes
= (ip
->ino_data
.size
+ HAMMER_XBUFMASK64
) &
711 } else if (ip
->ino_data
.size
> HAMMER_BUFSIZE
/ 2) {
712 vap
->va_bytes
= (ip
->ino_data
.size
+ HAMMER_BUFMASK64
) &
715 vap
->va_bytes
= (ip
->ino_data
.size
+ 15) & ~15;
717 vap
->va_type
= hammer_get_vnode_type(ip
->ino_data
.obj_type
);
718 vap
->va_filerev
= 0; /* XXX */
719 /* mtime uniquely identifies any adjustments made to the file XXX */
720 vap
->va_fsmid
= ip
->ino_data
.mtime
;
721 vap
->va_uid_uuid
= ip
->ino_data
.uid
;
722 vap
->va_gid_uuid
= ip
->ino_data
.gid
;
723 vap
->va_fsid_uuid
= ip
->hmp
->fsid
;
724 vap
->va_vaflags
= VA_UID_UUID_VALID
| VA_GID_UUID_VALID
|
727 switch (ip
->ino_data
.obj_type
) {
728 case HAMMER_OBJTYPE_CDEV
:
729 case HAMMER_OBJTYPE_BDEV
:
730 vap
->va_rmajor
= ip
->ino_data
.rmajor
;
731 vap
->va_rminor
= ip
->ino_data
.rminor
;
741 * hammer_vop_nresolve { nch, dvp, cred }
743 * Locate the requested directory entry.
747 hammer_vop_nresolve(struct vop_nresolve_args
*ap
)
749 struct hammer_transaction trans
;
750 struct namecache
*ncp
;
754 struct hammer_cursor cursor
;
762 u_int32_t localization
;
765 * Misc initialization, plus handle as-of name extensions. Look for
766 * the '@@' extension. Note that as-of files and directories cannot
769 dip
= VTOI(ap
->a_dvp
);
770 ncp
= ap
->a_nch
->ncp
;
771 asof
= dip
->obj_asof
;
775 hammer_simple_transaction(&trans
, dip
->hmp
);
777 for (i
= 0; i
< nlen
; ++i
) {
778 if (ncp
->nc_name
[i
] == '@' && ncp
->nc_name
[i
+1] == '@') {
779 asof
= hammer_str_to_tid(ncp
->nc_name
+ i
+ 2);
780 flags
|= HAMMER_INODE_RO
;
787 * If there is no path component the time extension is relative to
791 ip
= hammer_get_inode(&trans
, dip
, dip
->obj_id
,
792 asof
, dip
->obj_localization
,
795 error
= hammer_get_vnode(ip
, &vp
);
796 hammer_rel_inode(ip
, 0);
802 cache_setvp(ap
->a_nch
, vp
);
809 * Calculate the namekey and setup the key range for the scan. This
810 * works kinda like a chained hash table where the lower 32 bits
811 * of the namekey synthesize the chain.
813 * The key range is inclusive of both key_beg and key_end.
815 namekey
= hammer_directory_namekey(ncp
->nc_name
, nlen
);
817 error
= hammer_init_cursor(&trans
, &cursor
, &dip
->cache
[1], dip
);
818 cursor
.key_beg
.localization
= dip
->obj_localization
+
819 HAMMER_LOCALIZE_MISC
;
820 cursor
.key_beg
.obj_id
= dip
->obj_id
;
821 cursor
.key_beg
.key
= namekey
;
822 cursor
.key_beg
.create_tid
= 0;
823 cursor
.key_beg
.delete_tid
= 0;
824 cursor
.key_beg
.rec_type
= HAMMER_RECTYPE_DIRENTRY
;
825 cursor
.key_beg
.obj_type
= 0;
827 cursor
.key_end
= cursor
.key_beg
;
828 cursor
.key_end
.key
|= 0xFFFFFFFFULL
;
830 cursor
.flags
|= HAMMER_CURSOR_END_INCLUSIVE
| HAMMER_CURSOR_ASOF
;
833 * Scan all matching records (the chain), locate the one matching
834 * the requested path component.
836 * The hammer_ip_*() functions merge in-memory records with on-disk
837 * records for the purposes of the search.
840 localization
= HAMMER_DEF_LOCALIZATION
;
843 error
= hammer_ip_first(&cursor
);
845 error
= hammer_ip_resolve_data(&cursor
);
848 if (nlen
== cursor
.leaf
->data_len
- HAMMER_ENTRY_NAME_OFF
&&
849 bcmp(ncp
->nc_name
, cursor
.data
->entry
.name
, nlen
) == 0) {
850 obj_id
= cursor
.data
->entry
.obj_id
;
851 localization
= cursor
.data
->entry
.localization
;
854 error
= hammer_ip_next(&cursor
);
857 hammer_done_cursor(&cursor
);
859 ip
= hammer_get_inode(&trans
, dip
, obj_id
,
863 error
= hammer_get_vnode(ip
, &vp
);
864 hammer_rel_inode(ip
, 0);
870 cache_setvp(ap
->a_nch
, vp
);
873 } else if (error
== ENOENT
) {
874 cache_setvp(ap
->a_nch
, NULL
);
877 hammer_done_transaction(&trans
);
882 * hammer_vop_nlookupdotdot { dvp, vpp, cred }
884 * Locate the parent directory of a directory vnode.
886 * dvp is referenced but not locked. *vpp must be returned referenced and
887 * locked. A parent_obj_id of 0 does not necessarily indicate that we are
888 * at the root, instead it could indicate that the directory we were in was
891 * NOTE: as-of sequences are not linked into the directory structure. If
892 * we are at the root with a different asof then the mount point, reload
893 * the same directory with the mount point's asof. I'm not sure what this
894 * will do to NFS. We encode ASOF stamps in NFS file handles so it might not
895 * get confused, but it hasn't been tested.
899 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args
*ap
)
901 struct hammer_transaction trans
;
902 struct hammer_inode
*dip
;
903 struct hammer_inode
*ip
;
904 int64_t parent_obj_id
;
905 u_int32_t parent_obj_localization
;
909 dip
= VTOI(ap
->a_dvp
);
910 asof
= dip
->obj_asof
;
913 * Whos are parent? This could be the root of a pseudo-filesystem
914 * whos parent is in another localization domain.
916 parent_obj_id
= dip
->ino_data
.parent_obj_id
;
917 if (dip
->obj_id
== HAMMER_OBJID_ROOT
)
918 parent_obj_localization
= dip
->ino_data
.ext
.obj
.parent_obj_localization
;
920 parent_obj_localization
= dip
->obj_localization
;
922 if (parent_obj_id
== 0) {
923 if (dip
->obj_id
== HAMMER_OBJID_ROOT
&&
924 asof
!= dip
->hmp
->asof
) {
925 parent_obj_id
= dip
->obj_id
;
926 asof
= dip
->hmp
->asof
;
927 *ap
->a_fakename
= kmalloc(19, M_TEMP
, M_WAITOK
);
928 ksnprintf(*ap
->a_fakename
, 19, "0x%016llx",
936 hammer_simple_transaction(&trans
, dip
->hmp
);
938 ip
= hammer_get_inode(&trans
, dip
, parent_obj_id
,
939 asof
, parent_obj_localization
,
942 error
= hammer_get_vnode(ip
, ap
->a_vpp
);
943 hammer_rel_inode(ip
, 0);
947 hammer_done_transaction(&trans
);
952 * hammer_vop_nlink { nch, dvp, vp, cred }
956 hammer_vop_nlink(struct vop_nlink_args
*ap
)
958 struct hammer_transaction trans
;
959 struct hammer_inode
*dip
;
960 struct hammer_inode
*ip
;
961 struct nchandle
*nch
;
965 dip
= VTOI(ap
->a_dvp
);
968 if (dip
->flags
& HAMMER_INODE_RO
)
970 if (ip
->flags
& HAMMER_INODE_RO
)
972 if ((error
= hammer_checkspace(dip
->hmp
)) != 0)
976 * Create a transaction to cover the operations we perform.
978 hammer_start_transaction(&trans
, dip
->hmp
);
981 * Add the filesystem object to the directory. Note that neither
982 * dip nor ip are referenced or locked, but their vnodes are
983 * referenced. This function will bump the inode's link count.
985 error
= hammer_ip_add_directory(&trans
, dip
,
986 nch
->ncp
->nc_name
, nch
->ncp
->nc_nlen
,
993 cache_setunresolved(nch
);
994 cache_setvp(nch
, ap
->a_vp
);
996 hammer_done_transaction(&trans
);
1001 * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
1003 * The operating system has already ensured that the directory entry
1004 * does not exist and done all appropriate namespace locking.
1008 hammer_vop_nmkdir(struct vop_nmkdir_args
*ap
)
1010 struct hammer_transaction trans
;
1011 struct hammer_inode
*dip
;
1012 struct hammer_inode
*nip
;
1013 struct nchandle
*nch
;
1017 dip
= VTOI(ap
->a_dvp
);
1019 if (dip
->flags
& HAMMER_INODE_RO
)
1021 if ((error
= hammer_checkspace(dip
->hmp
)) != 0)
1025 * Create a transaction to cover the operations we perform.
1027 hammer_start_transaction(&trans
, dip
->hmp
);
1030 * Create a new filesystem object of the requested type. The
1031 * returned inode will be referenced but not locked.
1033 error
= hammer_create_inode(&trans
, ap
->a_vap
, ap
->a_cred
,
1036 hkprintf("hammer_mkdir error %d\n", error
);
1037 hammer_done_transaction(&trans
);
1042 * Add the new filesystem object to the directory. This will also
1043 * bump the inode's link count.
1045 error
= hammer_ip_add_directory(&trans
, dip
,
1046 nch
->ncp
->nc_name
, nch
->ncp
->nc_nlen
,
1049 hkprintf("hammer_mkdir (add) error %d\n", error
);
1055 hammer_rel_inode(nip
, 0);
1058 error
= hammer_get_vnode(nip
, ap
->a_vpp
);
1059 hammer_rel_inode(nip
, 0);
1061 cache_setunresolved(ap
->a_nch
);
1062 cache_setvp(ap
->a_nch
, *ap
->a_vpp
);
1065 hammer_done_transaction(&trans
);
1070 * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
1072 * The operating system has already ensured that the directory entry
1073 * does not exist and done all appropriate namespace locking.
1077 hammer_vop_nmknod(struct vop_nmknod_args
*ap
)
1079 struct hammer_transaction trans
;
1080 struct hammer_inode
*dip
;
1081 struct hammer_inode
*nip
;
1082 struct nchandle
*nch
;
1087 dip
= VTOI(ap
->a_dvp
);
1089 if (dip
->flags
& HAMMER_INODE_RO
)
1091 if ((error
= hammer_checkspace(dip
->hmp
)) != 0)
1095 * Create a transaction to cover the operations we perform.
1097 hammer_start_transaction(&trans
, dip
->hmp
);
1100 * Create a new filesystem object of the requested type. The
1101 * returned inode will be referenced but not locked.
1103 * If mknod specifies a directory a pseudo-fs is created.
1105 pseudofs
= (ap
->a_vap
->va_type
== VDIR
);
1106 error
= hammer_create_inode(&trans
, ap
->a_vap
, ap
->a_cred
,
1107 dip
, pseudofs
, &nip
);
1109 hammer_done_transaction(&trans
);
1115 * Add the new filesystem object to the directory. This will also
1116 * bump the inode's link count.
1118 error
= hammer_ip_add_directory(&trans
, dip
,
1119 nch
->ncp
->nc_name
, nch
->ncp
->nc_nlen
,
1126 hammer_rel_inode(nip
, 0);
1129 error
= hammer_get_vnode(nip
, ap
->a_vpp
);
1130 hammer_rel_inode(nip
, 0);
1132 cache_setunresolved(ap
->a_nch
);
1133 cache_setvp(ap
->a_nch
, *ap
->a_vpp
);
1136 hammer_done_transaction(&trans
);
1141 * hammer_vop_open { vp, mode, cred, fp }
1145 hammer_vop_open(struct vop_open_args
*ap
)
1149 ip
= VTOI(ap
->a_vp
);
1151 if ((ap
->a_mode
& FWRITE
) && (ip
->flags
& HAMMER_INODE_RO
))
1153 return(vop_stdopen(ap
));
1157 * hammer_vop_pathconf { vp, name, retval }
1161 hammer_vop_pathconf(struct vop_pathconf_args
*ap
)
1167 * hammer_vop_print { vp }
1171 hammer_vop_print(struct vop_print_args
*ap
)
1177 * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
1181 hammer_vop_readdir(struct vop_readdir_args
*ap
)
1183 struct hammer_transaction trans
;
1184 struct hammer_cursor cursor
;
1185 struct hammer_inode
*ip
;
1187 hammer_base_elm_t base
;
1195 ip
= VTOI(ap
->a_vp
);
1197 saveoff
= uio
->uio_offset
;
1199 if (ap
->a_ncookies
) {
1200 ncookies
= uio
->uio_resid
/ 16 + 1;
1201 if (ncookies
> 1024)
1203 cookies
= kmalloc(ncookies
* sizeof(off_t
), M_TEMP
, M_WAITOK
);
1211 hammer_simple_transaction(&trans
, ip
->hmp
);
1214 * Handle artificial entries
1218 r
= vop_write_dirent(&error
, uio
, ip
->obj_id
, DT_DIR
, 1, ".");
1222 cookies
[cookie_index
] = saveoff
;
1225 if (cookie_index
== ncookies
)
1229 if (ip
->ino_data
.parent_obj_id
) {
1230 r
= vop_write_dirent(&error
, uio
,
1231 ip
->ino_data
.parent_obj_id
,
1234 r
= vop_write_dirent(&error
, uio
,
1235 ip
->obj_id
, DT_DIR
, 2, "..");
1240 cookies
[cookie_index
] = saveoff
;
1243 if (cookie_index
== ncookies
)
1248 * Key range (begin and end inclusive) to scan. Directory keys
1249 * directly translate to a 64 bit 'seek' position.
1251 hammer_init_cursor(&trans
, &cursor
, &ip
->cache
[1], ip
);
1252 cursor
.key_beg
.localization
= ip
->obj_localization
+
1253 HAMMER_LOCALIZE_MISC
;
1254 cursor
.key_beg
.obj_id
= ip
->obj_id
;
1255 cursor
.key_beg
.create_tid
= 0;
1256 cursor
.key_beg
.delete_tid
= 0;
1257 cursor
.key_beg
.rec_type
= HAMMER_RECTYPE_DIRENTRY
;
1258 cursor
.key_beg
.obj_type
= 0;
1259 cursor
.key_beg
.key
= saveoff
;
1261 cursor
.key_end
= cursor
.key_beg
;
1262 cursor
.key_end
.key
= HAMMER_MAX_KEY
;
1263 cursor
.asof
= ip
->obj_asof
;
1264 cursor
.flags
|= HAMMER_CURSOR_END_INCLUSIVE
| HAMMER_CURSOR_ASOF
;
1266 error
= hammer_ip_first(&cursor
);
1268 while (error
== 0) {
1269 error
= hammer_ip_resolve_data(&cursor
);
1272 base
= &cursor
.leaf
->base
;
1273 saveoff
= base
->key
;
1274 KKASSERT(cursor
.leaf
->data_len
> HAMMER_ENTRY_NAME_OFF
);
1276 if (base
->obj_id
!= ip
->obj_id
)
1277 panic("readdir: bad record at %p", cursor
.node
);
1279 r
= vop_write_dirent(
1280 &error
, uio
, cursor
.data
->entry
.obj_id
,
1281 hammer_get_dtype(cursor
.leaf
->base
.obj_type
),
1282 cursor
.leaf
->data_len
- HAMMER_ENTRY_NAME_OFF
,
1283 (void *)cursor
.data
->entry
.name
);
1288 cookies
[cookie_index
] = base
->key
;
1290 if (cookie_index
== ncookies
)
1292 error
= hammer_ip_next(&cursor
);
1294 hammer_done_cursor(&cursor
);
1297 hammer_done_transaction(&trans
);
1300 *ap
->a_eofflag
= (error
== ENOENT
);
1301 uio
->uio_offset
= saveoff
;
1302 if (error
&& cookie_index
== 0) {
1303 if (error
== ENOENT
)
1306 kfree(cookies
, M_TEMP
);
1307 *ap
->a_ncookies
= 0;
1308 *ap
->a_cookies
= NULL
;
1311 if (error
== ENOENT
)
1314 *ap
->a_ncookies
= cookie_index
;
1315 *ap
->a_cookies
= cookies
;
1322 * hammer_vop_readlink { vp, uio, cred }
1326 hammer_vop_readlink(struct vop_readlink_args
*ap
)
1328 struct hammer_transaction trans
;
1329 struct hammer_cursor cursor
;
1330 struct hammer_inode
*ip
;
1333 ip
= VTOI(ap
->a_vp
);
1336 * Shortcut if the symlink data was stuffed into ino_data.
1338 if (ip
->ino_data
.size
<= HAMMER_INODE_BASESYMLEN
) {
1339 error
= uiomove(ip
->ino_data
.ext
.symlink
,
1340 ip
->ino_data
.size
, ap
->a_uio
);
1347 hammer_simple_transaction(&trans
, ip
->hmp
);
1348 hammer_init_cursor(&trans
, &cursor
, &ip
->cache
[1], ip
);
1351 * Key range (begin and end inclusive) to scan. Directory keys
1352 * directly translate to a 64 bit 'seek' position.
1354 cursor
.key_beg
.localization
= ip
->obj_localization
+
1355 HAMMER_LOCALIZE_MISC
;
1356 cursor
.key_beg
.obj_id
= ip
->obj_id
;
1357 cursor
.key_beg
.create_tid
= 0;
1358 cursor
.key_beg
.delete_tid
= 0;
1359 cursor
.key_beg
.rec_type
= HAMMER_RECTYPE_FIX
;
1360 cursor
.key_beg
.obj_type
= 0;
1361 cursor
.key_beg
.key
= HAMMER_FIXKEY_SYMLINK
;
1362 cursor
.asof
= ip
->obj_asof
;
1363 cursor
.flags
|= HAMMER_CURSOR_ASOF
;
1365 error
= hammer_ip_lookup(&cursor
);
1367 error
= hammer_ip_resolve_data(&cursor
);
1369 KKASSERT(cursor
.leaf
->data_len
>=
1370 HAMMER_SYMLINK_NAME_OFF
);
1371 error
= uiomove(cursor
.data
->symlink
.name
,
1372 cursor
.leaf
->data_len
-
1373 HAMMER_SYMLINK_NAME_OFF
,
1377 hammer_done_cursor(&cursor
);
1378 hammer_done_transaction(&trans
);
1383 * hammer_vop_nremove { nch, dvp, cred }
1387 hammer_vop_nremove(struct vop_nremove_args
*ap
)
1389 struct hammer_transaction trans
;
1390 struct hammer_inode
*dip
;
1393 dip
= VTOI(ap
->a_dvp
);
1395 if (hammer_nohistory(dip
) == 0 &&
1396 (error
= hammer_checkspace(dip
->hmp
)) != 0) {
1400 hammer_start_transaction(&trans
, dip
->hmp
);
1401 error
= hammer_dounlink(&trans
, ap
->a_nch
, ap
->a_dvp
, ap
->a_cred
, 0);
1402 hammer_done_transaction(&trans
);
1408 * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1412 hammer_vop_nrename(struct vop_nrename_args
*ap
)
1414 struct hammer_transaction trans
;
1415 struct namecache
*fncp
;
1416 struct namecache
*tncp
;
1417 struct hammer_inode
*fdip
;
1418 struct hammer_inode
*tdip
;
1419 struct hammer_inode
*ip
;
1420 struct hammer_cursor cursor
;
1424 fdip
= VTOI(ap
->a_fdvp
);
1425 tdip
= VTOI(ap
->a_tdvp
);
1426 fncp
= ap
->a_fnch
->ncp
;
1427 tncp
= ap
->a_tnch
->ncp
;
1428 ip
= VTOI(fncp
->nc_vp
);
1429 KKASSERT(ip
!= NULL
);
1431 if (fdip
->flags
& HAMMER_INODE_RO
)
1433 if (tdip
->flags
& HAMMER_INODE_RO
)
1435 if (ip
->flags
& HAMMER_INODE_RO
)
1437 if ((error
= hammer_checkspace(fdip
->hmp
)) != 0)
1440 hammer_start_transaction(&trans
, fdip
->hmp
);
1443 * Remove tncp from the target directory and then link ip as
1444 * tncp. XXX pass trans to dounlink
1446 * Force the inode sync-time to match the transaction so it is
1447 * in-sync with the creation of the target directory entry.
1449 error
= hammer_dounlink(&trans
, ap
->a_tnch
, ap
->a_tdvp
, ap
->a_cred
, 0);
1450 if (error
== 0 || error
== ENOENT
) {
1451 error
= hammer_ip_add_directory(&trans
, tdip
,
1452 tncp
->nc_name
, tncp
->nc_nlen
,
1455 ip
->ino_data
.parent_obj_id
= tdip
->obj_id
;
1456 hammer_modify_inode(ip
, HAMMER_INODE_DDIRTY
);
1460 goto failed
; /* XXX */
1463 * Locate the record in the originating directory and remove it.
1465 * Calculate the namekey and setup the key range for the scan. This
1466 * works kinda like a chained hash table where the lower 32 bits
1467 * of the namekey synthesize the chain.
1469 * The key range is inclusive of both key_beg and key_end.
1471 namekey
= hammer_directory_namekey(fncp
->nc_name
, fncp
->nc_nlen
);
1473 hammer_init_cursor(&trans
, &cursor
, &fdip
->cache
[1], fdip
);
1474 cursor
.key_beg
.localization
= fdip
->obj_localization
+
1475 HAMMER_LOCALIZE_MISC
;
1476 cursor
.key_beg
.obj_id
= fdip
->obj_id
;
1477 cursor
.key_beg
.key
= namekey
;
1478 cursor
.key_beg
.create_tid
= 0;
1479 cursor
.key_beg
.delete_tid
= 0;
1480 cursor
.key_beg
.rec_type
= HAMMER_RECTYPE_DIRENTRY
;
1481 cursor
.key_beg
.obj_type
= 0;
1483 cursor
.key_end
= cursor
.key_beg
;
1484 cursor
.key_end
.key
|= 0xFFFFFFFFULL
;
1485 cursor
.asof
= fdip
->obj_asof
;
1486 cursor
.flags
|= HAMMER_CURSOR_END_INCLUSIVE
| HAMMER_CURSOR_ASOF
;
1489 * Scan all matching records (the chain), locate the one matching
1490 * the requested path component.
1492 * The hammer_ip_*() functions merge in-memory records with on-disk
1493 * records for the purposes of the search.
1495 error
= hammer_ip_first(&cursor
);
1496 while (error
== 0) {
1497 if (hammer_ip_resolve_data(&cursor
) != 0)
1499 nlen
= cursor
.leaf
->data_len
- HAMMER_ENTRY_NAME_OFF
;
1501 if (fncp
->nc_nlen
== nlen
&&
1502 bcmp(fncp
->nc_name
, cursor
.data
->entry
.name
, nlen
) == 0) {
1505 error
= hammer_ip_next(&cursor
);
1509 * If all is ok we have to get the inode so we can adjust nlinks.
1511 * WARNING: hammer_ip_del_directory() may have to terminate the
1512 * cursor to avoid a recursion. It's ok to call hammer_done_cursor()
1516 error
= hammer_ip_del_directory(&trans
, &cursor
, fdip
, ip
);
1519 * XXX A deadlock here will break rename's atomicy for the purposes
1520 * of crash recovery.
1522 if (error
== EDEADLK
) {
1523 hammer_done_cursor(&cursor
);
1528 * Cleanup and tell the kernel that the rename succeeded.
1530 hammer_done_cursor(&cursor
);
1532 cache_rename(ap
->a_fnch
, ap
->a_tnch
);
1535 hammer_done_transaction(&trans
);
1540 * hammer_vop_nrmdir { nch, dvp, cred }
1544 hammer_vop_nrmdir(struct vop_nrmdir_args
*ap
)
1546 struct hammer_transaction trans
;
1547 struct hammer_inode
*dip
;
1550 dip
= VTOI(ap
->a_dvp
);
1552 if (hammer_nohistory(dip
) == 0 &&
1553 (error
= hammer_checkspace(dip
->hmp
)) != 0) {
1557 hammer_start_transaction(&trans
, dip
->hmp
);
1558 error
= hammer_dounlink(&trans
, ap
->a_nch
, ap
->a_dvp
, ap
->a_cred
, 0);
1559 hammer_done_transaction(&trans
);
1565 * hammer_vop_setattr { vp, vap, cred }
1569 hammer_vop_setattr(struct vop_setattr_args
*ap
)
1571 struct hammer_transaction trans
;
1573 struct hammer_inode
*ip
;
1578 int64_t aligned_size
;
1582 ip
= ap
->a_vp
->v_data
;
1585 if (ap
->a_vp
->v_mount
->mnt_flag
& MNT_RDONLY
)
1587 if (ip
->flags
& HAMMER_INODE_RO
)
1589 if (hammer_nohistory(ip
) == 0 &&
1590 (error
= hammer_checkspace(ip
->hmp
)) != 0) {
1594 hammer_start_transaction(&trans
, ip
->hmp
);
1597 if (vap
->va_flags
!= VNOVAL
) {
1598 flags
= ip
->ino_data
.uflags
;
1599 error
= vop_helper_setattr_flags(&flags
, vap
->va_flags
,
1600 hammer_to_unix_xid(&ip
->ino_data
.uid
),
1603 if (ip
->ino_data
.uflags
!= flags
) {
1604 ip
->ino_data
.uflags
= flags
;
1605 modflags
|= HAMMER_INODE_DDIRTY
;
1607 if (ip
->ino_data
.uflags
& (IMMUTABLE
| APPEND
)) {
1614 if (ip
->ino_data
.uflags
& (IMMUTABLE
| APPEND
)) {
1618 if (vap
->va_uid
!= (uid_t
)VNOVAL
|| vap
->va_gid
!= (gid_t
)VNOVAL
) {
1619 mode_t cur_mode
= ip
->ino_data
.mode
;
1620 uid_t cur_uid
= hammer_to_unix_xid(&ip
->ino_data
.uid
);
1621 gid_t cur_gid
= hammer_to_unix_xid(&ip
->ino_data
.gid
);
1625 error
= vop_helper_chown(ap
->a_vp
, vap
->va_uid
, vap
->va_gid
,
1627 &cur_uid
, &cur_gid
, &cur_mode
);
1629 hammer_guid_to_uuid(&uuid_uid
, cur_uid
);
1630 hammer_guid_to_uuid(&uuid_gid
, cur_gid
);
1631 if (bcmp(&uuid_uid
, &ip
->ino_data
.uid
,
1632 sizeof(uuid_uid
)) ||
1633 bcmp(&uuid_gid
, &ip
->ino_data
.gid
,
1634 sizeof(uuid_gid
)) ||
1635 ip
->ino_data
.mode
!= cur_mode
1637 ip
->ino_data
.uid
= uuid_uid
;
1638 ip
->ino_data
.gid
= uuid_gid
;
1639 ip
->ino_data
.mode
= cur_mode
;
1641 modflags
|= HAMMER_INODE_DDIRTY
;
1644 while (vap
->va_size
!= VNOVAL
&& ip
->ino_data
.size
!= vap
->va_size
) {
1645 switch(ap
->a_vp
->v_type
) {
1647 if (vap
->va_size
== ip
->ino_data
.size
)
1650 * XXX break atomicy, we can deadlock the backend
1651 * if we do not release the lock. Probably not a
1654 blksize
= hammer_blocksize(vap
->va_size
);
1655 if (vap
->va_size
< ip
->ino_data
.size
) {
1656 vtruncbuf(ap
->a_vp
, vap
->va_size
, blksize
);
1659 vnode_pager_setsize(ap
->a_vp
, vap
->va_size
);
1662 ip
->ino_data
.size
= vap
->va_size
;
1663 modflags
|= HAMMER_INODE_DDIRTY
;
1666 * on-media truncation is cached in the inode until
1667 * the inode is synchronized.
1670 hammer_ip_frontend_trunc(ip
, vap
->va_size
);
1671 hammer_update_rsv_databufs(ip
);
1672 #ifdef DEBUG_TRUNCATE
1673 if (HammerTruncIp
== NULL
)
1676 if ((ip
->flags
& HAMMER_INODE_TRUNCATED
) == 0) {
1677 ip
->flags
|= HAMMER_INODE_TRUNCATED
;
1678 ip
->trunc_off
= vap
->va_size
;
1679 #ifdef DEBUG_TRUNCATE
1680 if (ip
== HammerTruncIp
)
1681 kprintf("truncate1 %016llx\n", ip
->trunc_off
);
1683 } else if (ip
->trunc_off
> vap
->va_size
) {
1684 ip
->trunc_off
= vap
->va_size
;
1685 #ifdef DEBUG_TRUNCATE
1686 if (ip
== HammerTruncIp
)
1687 kprintf("truncate2 %016llx\n", ip
->trunc_off
);
1690 #ifdef DEBUG_TRUNCATE
1691 if (ip
== HammerTruncIp
)
1692 kprintf("truncate3 %016llx (ignored)\n", vap
->va_size
);
1698 * If truncating we have to clean out a portion of
1699 * the last block on-disk. We do this in the
1700 * front-end buffer cache.
1702 aligned_size
= (vap
->va_size
+ (blksize
- 1)) &
1703 ~(int64_t)(blksize
- 1);
1704 if (truncating
&& vap
->va_size
< aligned_size
) {
1708 aligned_size
-= blksize
;
1710 offset
= (int)vap
->va_size
& (blksize
- 1);
1711 error
= bread(ap
->a_vp
, aligned_size
,
1713 hammer_ip_frontend_trunc(ip
, aligned_size
);
1715 bzero(bp
->b_data
+ offset
,
1719 kprintf("ERROR %d\n", error
);
1725 if ((ip
->flags
& HAMMER_INODE_TRUNCATED
) == 0) {
1726 ip
->flags
|= HAMMER_INODE_TRUNCATED
;
1727 ip
->trunc_off
= vap
->va_size
;
1728 } else if (ip
->trunc_off
> vap
->va_size
) {
1729 ip
->trunc_off
= vap
->va_size
;
1731 hammer_ip_frontend_trunc(ip
, vap
->va_size
);
1732 ip
->ino_data
.size
= vap
->va_size
;
1733 modflags
|= HAMMER_INODE_DDIRTY
;
1741 if (vap
->va_atime
.tv_sec
!= VNOVAL
) {
1742 ip
->ino_data
.atime
=
1743 hammer_timespec_to_time(&vap
->va_atime
);
1744 modflags
|= HAMMER_INODE_ATIME
;
1746 if (vap
->va_mtime
.tv_sec
!= VNOVAL
) {
1747 ip
->ino_data
.mtime
=
1748 hammer_timespec_to_time(&vap
->va_mtime
);
1749 modflags
|= HAMMER_INODE_MTIME
;
1751 if (vap
->va_mode
!= (mode_t
)VNOVAL
) {
1752 mode_t cur_mode
= ip
->ino_data
.mode
;
1753 uid_t cur_uid
= hammer_to_unix_xid(&ip
->ino_data
.uid
);
1754 gid_t cur_gid
= hammer_to_unix_xid(&ip
->ino_data
.gid
);
1756 error
= vop_helper_chmod(ap
->a_vp
, vap
->va_mode
, ap
->a_cred
,
1757 cur_uid
, cur_gid
, &cur_mode
);
1758 if (error
== 0 && ip
->ino_data
.mode
!= cur_mode
) {
1759 ip
->ino_data
.mode
= cur_mode
;
1760 modflags
|= HAMMER_INODE_DDIRTY
;
1765 hammer_modify_inode(ip
, modflags
);
1766 hammer_done_transaction(&trans
);
1771 * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1775 hammer_vop_nsymlink(struct vop_nsymlink_args
*ap
)
1777 struct hammer_transaction trans
;
1778 struct hammer_inode
*dip
;
1779 struct hammer_inode
*nip
;
1780 struct nchandle
*nch
;
1781 hammer_record_t record
;
1785 ap
->a_vap
->va_type
= VLNK
;
1788 dip
= VTOI(ap
->a_dvp
);
1790 if (dip
->flags
& HAMMER_INODE_RO
)
1792 if ((error
= hammer_checkspace(dip
->hmp
)) != 0)
1796 * Create a transaction to cover the operations we perform.
1798 hammer_start_transaction(&trans
, dip
->hmp
);
1801 * Create a new filesystem object of the requested type. The
1802 * returned inode will be referenced but not locked.
1805 error
= hammer_create_inode(&trans
, ap
->a_vap
, ap
->a_cred
,
1808 hammer_done_transaction(&trans
);
1814 * Add a record representing the symlink. symlink stores the link
1815 * as pure data, not a string, and is no \0 terminated.
1818 bytes
= strlen(ap
->a_target
);
1820 if (bytes
<= HAMMER_INODE_BASESYMLEN
) {
1821 bcopy(ap
->a_target
, nip
->ino_data
.ext
.symlink
, bytes
);
1823 record
= hammer_alloc_mem_record(nip
, bytes
);
1824 record
->type
= HAMMER_MEM_RECORD_GENERAL
;
1826 record
->leaf
.base
.localization
= nip
->obj_localization
+
1827 HAMMER_LOCALIZE_MISC
;
1828 record
->leaf
.base
.key
= HAMMER_FIXKEY_SYMLINK
;
1829 record
->leaf
.base
.rec_type
= HAMMER_RECTYPE_FIX
;
1830 record
->leaf
.data_len
= bytes
;
1831 KKASSERT(HAMMER_SYMLINK_NAME_OFF
== 0);
1832 bcopy(ap
->a_target
, record
->data
->symlink
.name
, bytes
);
1833 error
= hammer_ip_add_record(&trans
, record
);
1837 * Set the file size to the length of the link.
1840 nip
->ino_data
.size
= bytes
;
1841 hammer_modify_inode(nip
, HAMMER_INODE_DDIRTY
);
1845 error
= hammer_ip_add_directory(&trans
, dip
, nch
->ncp
->nc_name
,
1846 nch
->ncp
->nc_nlen
, nip
);
1852 hammer_rel_inode(nip
, 0);
1855 error
= hammer_get_vnode(nip
, ap
->a_vpp
);
1856 hammer_rel_inode(nip
, 0);
1858 cache_setunresolved(ap
->a_nch
);
1859 cache_setvp(ap
->a_nch
, *ap
->a_vpp
);
1862 hammer_done_transaction(&trans
);
1867 * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1871 hammer_vop_nwhiteout(struct vop_nwhiteout_args
*ap
)
1873 struct hammer_transaction trans
;
1874 struct hammer_inode
*dip
;
1877 dip
= VTOI(ap
->a_dvp
);
1879 if (hammer_nohistory(dip
) == 0 &&
1880 (error
= hammer_checkspace(dip
->hmp
)) != 0) {
1884 hammer_start_transaction(&trans
, dip
->hmp
);
1885 error
= hammer_dounlink(&trans
, ap
->a_nch
, ap
->a_dvp
,
1886 ap
->a_cred
, ap
->a_flags
);
1887 hammer_done_transaction(&trans
);
1893 * hammer_vop_ioctl { vp, command, data, fflag, cred }
1897 hammer_vop_ioctl(struct vop_ioctl_args
*ap
)
1899 struct hammer_inode
*ip
= ap
->a_vp
->v_data
;
1901 return(hammer_ioctl(ip
, ap
->a_command
, ap
->a_data
,
1902 ap
->a_fflag
, ap
->a_cred
));
1907 hammer_vop_mountctl(struct vop_mountctl_args
*ap
)
1912 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
1915 case MOUNTCTL_SET_EXPORT
:
1916 if (ap
->a_ctllen
!= sizeof(struct export_args
))
1918 error
= hammer_vfs_export(mp
, ap
->a_op
,
1919 (const struct export_args
*)ap
->a_ctl
);
1922 error
= journal_mountctl(ap
);
1929 * hammer_vop_strategy { vp, bio }
1931 * Strategy call, used for regular file read & write only. Note that the
1932 * bp may represent a cluster.
1934 * To simplify operation and allow better optimizations in the future,
1935 * this code does not make any assumptions with regards to buffer alignment
1940 hammer_vop_strategy(struct vop_strategy_args
*ap
)
1945 bp
= ap
->a_bio
->bio_buf
;
1949 error
= hammer_vop_strategy_read(ap
);
1952 error
= hammer_vop_strategy_write(ap
);
1955 bp
->b_error
= error
= EINVAL
;
1956 bp
->b_flags
|= B_ERROR
;
1964 * Read from a regular file. Iterate the related records and fill in the
1965 * BIO/BUF. Gaps are zero-filled.
1967 * The support code in hammer_object.c should be used to deal with mixed
1968 * in-memory and on-disk records.
1970 * NOTE: Can be called from the cluster code with an oversized buf.
1976 hammer_vop_strategy_read(struct vop_strategy_args
*ap
)
1978 struct hammer_transaction trans
;
1979 struct hammer_inode
*ip
;
1980 struct hammer_cursor cursor
;
1981 hammer_base_elm_t base
;
1982 hammer_off_t disk_offset
;
1996 ip
= ap
->a_vp
->v_data
;
1999 * The zone-2 disk offset may have been set by the cluster code via
2000 * a BMAP operation, or else should be NOOFFSET.
2002 * Checking the high bits for a match against zone-2 should suffice.
2004 nbio
= push_bio(bio
);
2005 if ((nbio
->bio_offset
& HAMMER_OFF_ZONE_MASK
) ==
2006 HAMMER_ZONE_RAW_BUFFER
) {
2007 error
= hammer_io_direct_read(ip
->hmp
, nbio
);
2012 * Well, that sucked. Do it the hard way. If all the stars are
2013 * aligned we may still be able to issue a direct-read.
2015 hammer_simple_transaction(&trans
, ip
->hmp
);
2016 hammer_init_cursor(&trans
, &cursor
, &ip
->cache
[1], ip
);
2019 * Key range (begin and end inclusive) to scan. Note that the key's
2020 * stored in the actual records represent BASE+LEN, not BASE. The
2021 * first record containing bio_offset will have a key > bio_offset.
2023 cursor
.key_beg
.localization
= ip
->obj_localization
+
2024 HAMMER_LOCALIZE_MISC
;
2025 cursor
.key_beg
.obj_id
= ip
->obj_id
;
2026 cursor
.key_beg
.create_tid
= 0;
2027 cursor
.key_beg
.delete_tid
= 0;
2028 cursor
.key_beg
.obj_type
= 0;
2029 cursor
.key_beg
.key
= bio
->bio_offset
+ 1;
2030 cursor
.asof
= ip
->obj_asof
;
2031 cursor
.flags
|= HAMMER_CURSOR_ASOF
;
2033 cursor
.key_end
= cursor
.key_beg
;
2034 KKASSERT(ip
->ino_data
.obj_type
== HAMMER_OBJTYPE_REGFILE
);
2036 if (ip
->ino_data
.obj_type
== HAMMER_OBJTYPE_DBFILE
) {
2037 cursor
.key_beg
.rec_type
= HAMMER_RECTYPE_DB
;
2038 cursor
.key_end
.rec_type
= HAMMER_RECTYPE_DB
;
2039 cursor
.key_end
.key
= 0x7FFFFFFFFFFFFFFFLL
;
2043 ran_end
= bio
->bio_offset
+ bp
->b_bufsize
;
2044 cursor
.key_beg
.rec_type
= HAMMER_RECTYPE_DATA
;
2045 cursor
.key_end
.rec_type
= HAMMER_RECTYPE_DATA
;
2046 tmp64
= ran_end
+ MAXPHYS
+ 1; /* work-around GCC-4 bug */
2047 if (tmp64
< ran_end
)
2048 cursor
.key_end
.key
= 0x7FFFFFFFFFFFFFFFLL
;
2050 cursor
.key_end
.key
= ran_end
+ MAXPHYS
+ 1;
2052 cursor
.flags
|= HAMMER_CURSOR_END_INCLUSIVE
;
2054 error
= hammer_ip_first(&cursor
);
2057 while (error
== 0) {
2059 * Get the base file offset of the record. The key for
2060 * data records is (base + bytes) rather then (base).
2062 base
= &cursor
.leaf
->base
;
2063 rec_offset
= base
->key
- cursor
.leaf
->data_len
;
2066 * Calculate the gap, if any, and zero-fill it.
2068 * n is the offset of the start of the record verses our
2069 * current seek offset in the bio.
2071 n
= (int)(rec_offset
- (bio
->bio_offset
+ boff
));
2073 if (n
> bp
->b_bufsize
- boff
)
2074 n
= bp
->b_bufsize
- boff
;
2075 bzero((char *)bp
->b_data
+ boff
, n
);
2081 * Calculate the data offset in the record and the number
2082 * of bytes we can copy.
2084 * There are two degenerate cases. First, boff may already
2085 * be at bp->b_bufsize. Secondly, the data offset within
2086 * the record may exceed the record's size.
2090 n
= cursor
.leaf
->data_len
- roff
;
2092 kprintf("strategy_read: bad n=%d roff=%d\n", n
, roff
);
2094 } else if (n
> bp
->b_bufsize
- boff
) {
2095 n
= bp
->b_bufsize
- boff
;
2099 * Deal with cached truncations. This cool bit of code
2100 * allows truncate()/ftruncate() to avoid having to sync
2103 * If the frontend is truncated then all backend records are
2104 * subject to the frontend's truncation.
2106 * If the backend is truncated then backend records on-disk
2107 * (but not in-memory) are subject to the backend's
2108 * truncation. In-memory records owned by the backend
2109 * represent data written after the truncation point on the
2110 * backend and must not be truncated.
2112 * Truncate operations deal with frontend buffer cache
2113 * buffers and frontend-owned in-memory records synchronously.
2115 if (ip
->flags
& HAMMER_INODE_TRUNCATED
) {
2116 if (hammer_cursor_ondisk(&cursor
) ||
2117 cursor
.iprec
->flush_state
== HAMMER_FST_FLUSH
) {
2118 if (ip
->trunc_off
<= rec_offset
)
2120 else if (ip
->trunc_off
< rec_offset
+ n
)
2121 n
= (int)(ip
->trunc_off
- rec_offset
);
2124 if (ip
->sync_flags
& HAMMER_INODE_TRUNCATED
) {
2125 if (hammer_cursor_ondisk(&cursor
)) {
2126 if (ip
->sync_trunc_off
<= rec_offset
)
2128 else if (ip
->sync_trunc_off
< rec_offset
+ n
)
2129 n
= (int)(ip
->sync_trunc_off
- rec_offset
);
2134 * Try to issue a direct read into our bio if possible,
2135 * otherwise resolve the element data into a hammer_buffer
2138 * The buffer on-disk should be zerod past any real
2139 * truncation point, but may not be for any synthesized
2140 * truncation point from above.
2142 if (boff
== 0 && n
== bp
->b_bufsize
&&
2143 ((cursor
.leaf
->data_offset
+ roff
) & HAMMER_BUFMASK
) == 0) {
2144 disk_offset
= hammer_blockmap_lookup(
2146 cursor
.leaf
->data_offset
+ roff
,
2150 nbio
->bio_offset
= disk_offset
;
2151 error
= hammer_io_direct_read(trans
.hmp
, nbio
);
2154 error
= hammer_ip_resolve_data(&cursor
);
2156 bcopy((char *)cursor
.data
+ roff
,
2157 (char *)bp
->b_data
+ boff
, n
);
2164 * Iterate until we have filled the request.
2167 if (boff
== bp
->b_bufsize
)
2169 error
= hammer_ip_next(&cursor
);
2173 * There may have been a gap after the last record
2175 if (error
== ENOENT
)
2177 if (error
== 0 && boff
!= bp
->b_bufsize
) {
2178 KKASSERT(boff
< bp
->b_bufsize
);
2179 bzero((char *)bp
->b_data
+ boff
, bp
->b_bufsize
- boff
);
2180 /* boff = bp->b_bufsize; */
2183 bp
->b_error
= error
;
2185 bp
->b_flags
|= B_ERROR
;
2190 hammer_cache_node(&ip
->cache
[1], cursor
.node
);
2191 hammer_done_cursor(&cursor
);
2192 hammer_done_transaction(&trans
);
2197 * BMAP operation - used to support cluster_read() only.
2199 * (struct vnode *vp, off_t loffset, off_t *doffsetp, int *runp, int *runb)
2201 * This routine may return EOPNOTSUPP if the opration is not supported for
2202 * the specified offset. The contents of the pointer arguments do not
2203 * need to be initialized in that case.
2205 * If a disk address is available and properly aligned return 0 with
2206 * *doffsetp set to the zone-2 address, and *runp / *runb set appropriately
2207 * to the run-length relative to that offset. Callers may assume that
2208 * *doffsetp is valid if 0 is returned, even if *runp is not sufficiently
2209 * large, so return EOPNOTSUPP if it is not sufficiently large.
2213 hammer_vop_bmap(struct vop_bmap_args
*ap
)
2215 struct hammer_transaction trans
;
2216 struct hammer_inode
*ip
;
2217 struct hammer_cursor cursor
;
2218 hammer_base_elm_t base
;
2222 int64_t base_offset
;
2223 int64_t base_disk_offset
;
2224 int64_t last_offset
;
2225 hammer_off_t last_disk_offset
;
2226 hammer_off_t disk_offset
;
2231 ip
= ap
->a_vp
->v_data
;
2234 * We can only BMAP regular files. We can't BMAP database files,
2237 if (ip
->ino_data
.obj_type
!= HAMMER_OBJTYPE_REGFILE
)
2241 * bmap is typically called with runp/runb both NULL when used
2242 * for writing. We do not support BMAP for writing atm.
2244 if (ap
->a_cmd
!= BUF_CMD_READ
)
2248 * Scan the B-Tree to acquire blockmap addresses, then translate
2251 hammer_simple_transaction(&trans
, ip
->hmp
);
2253 kprintf("bmap_beg %016llx ip->cache %p\n", ap
->a_loffset
, ip
->cache
[1]);
2255 hammer_init_cursor(&trans
, &cursor
, &ip
->cache
[1], ip
);
2258 * Key range (begin and end inclusive) to scan. Note that the key's
2259 * stored in the actual records represent BASE+LEN, not BASE. The
2260 * first record containing bio_offset will have a key > bio_offset.
2262 cursor
.key_beg
.localization
= ip
->obj_localization
+
2263 HAMMER_LOCALIZE_MISC
;
2264 cursor
.key_beg
.obj_id
= ip
->obj_id
;
2265 cursor
.key_beg
.create_tid
= 0;
2266 cursor
.key_beg
.delete_tid
= 0;
2267 cursor
.key_beg
.obj_type
= 0;
2269 cursor
.key_beg
.key
= ap
->a_loffset
- MAXPHYS
+ 1;
2271 cursor
.key_beg
.key
= ap
->a_loffset
+ 1;
2272 if (cursor
.key_beg
.key
< 0)
2273 cursor
.key_beg
.key
= 0;
2274 cursor
.asof
= ip
->obj_asof
;
2275 cursor
.flags
|= HAMMER_CURSOR_ASOF
;
2277 cursor
.key_end
= cursor
.key_beg
;
2278 KKASSERT(ip
->ino_data
.obj_type
== HAMMER_OBJTYPE_REGFILE
);
2280 ran_end
= ap
->a_loffset
+ MAXPHYS
;
2281 cursor
.key_beg
.rec_type
= HAMMER_RECTYPE_DATA
;
2282 cursor
.key_end
.rec_type
= HAMMER_RECTYPE_DATA
;
2283 tmp64
= ran_end
+ MAXPHYS
+ 1; /* work-around GCC-4 bug */
2284 if (tmp64
< ran_end
)
2285 cursor
.key_end
.key
= 0x7FFFFFFFFFFFFFFFLL
;
2287 cursor
.key_end
.key
= ran_end
+ MAXPHYS
+ 1;
2289 cursor
.flags
|= HAMMER_CURSOR_END_INCLUSIVE
;
2291 error
= hammer_ip_first(&cursor
);
2292 base_offset
= last_offset
= 0;
2293 base_disk_offset
= last_disk_offset
= 0;
2295 while (error
== 0) {
2297 * Get the base file offset of the record. The key for
2298 * data records is (base + bytes) rather then (base).
2300 * NOTE: rec_offset + rec_len may exceed the end-of-file.
2301 * The extra bytes should be zero on-disk and the BMAP op
2302 * should still be ok.
2304 base
= &cursor
.leaf
->base
;
2305 rec_offset
= base
->key
- cursor
.leaf
->data_len
;
2306 rec_len
= cursor
.leaf
->data_len
;
2309 * Incorporate any cached truncation.
2311 * NOTE: Modifications to rec_len based on synthesized
2312 * truncation points remove the guarantee that any extended
2313 * data on disk is zero (since the truncations may not have
2314 * taken place on-media yet).
2316 if (ip
->flags
& HAMMER_INODE_TRUNCATED
) {
2317 if (hammer_cursor_ondisk(&cursor
) ||
2318 cursor
.iprec
->flush_state
== HAMMER_FST_FLUSH
) {
2319 if (ip
->trunc_off
<= rec_offset
)
2321 else if (ip
->trunc_off
< rec_offset
+ rec_len
)
2322 rec_len
= (int)(ip
->trunc_off
- rec_offset
);
2325 if (ip
->sync_flags
& HAMMER_INODE_TRUNCATED
) {
2326 if (hammer_cursor_ondisk(&cursor
)) {
2327 if (ip
->sync_trunc_off
<= rec_offset
)
2329 else if (ip
->sync_trunc_off
< rec_offset
+ rec_len
)
2330 rec_len
= (int)(ip
->sync_trunc_off
- rec_offset
);
2335 * Accumulate information. If we have hit a discontiguous
2336 * block reset base_offset unless we are already beyond the
2337 * requested offset. If we are, that's it, we stop.
2339 disk_offset
= hammer_blockmap_lookup(trans
.hmp
,
2340 cursor
.leaf
->data_offset
,
2344 if (rec_offset
!= last_offset
||
2345 disk_offset
!= last_disk_offset
) {
2346 if (rec_offset
> ap
->a_loffset
)
2348 base_offset
= rec_offset
;
2349 base_disk_offset
= disk_offset
;
2351 last_offset
= rec_offset
+ rec_len
;
2352 last_disk_offset
= disk_offset
+ rec_len
;
2354 error
= hammer_ip_next(&cursor
);
2358 kprintf("BMAP %016llx: %016llx - %016llx\n",
2359 ap
->a_loffset
, base_offset
, last_offset
);
2360 kprintf("BMAP %16s: %016llx - %016llx\n",
2361 "", base_disk_offset
, last_disk_offset
);
2365 hammer_cache_node(&ip
->cache
[1], cursor
.node
);
2367 kprintf("bmap_end2 %016llx ip->cache %p\n", ap
->a_loffset
, ip
->cache
[1]);
2370 hammer_done_cursor(&cursor
);
2371 hammer_done_transaction(&trans
);
2374 * If we couldn't find any records or the records we did find were
2375 * all behind the requested offset, return failure. A forward
2376 * truncation can leave a hole w/ no on-disk records.
2378 if (last_offset
== 0 || last_offset
< ap
->a_loffset
)
2379 return (EOPNOTSUPP
);
2382 * Figure out the block size at the requested offset and adjust
2383 * our limits so the cluster_read() does not create inappropriately
2384 * sized buffer cache buffers.
2386 blksize
= hammer_blocksize(ap
->a_loffset
);
2387 if (hammer_blocksize(base_offset
) != blksize
) {
2388 base_offset
= hammer_blockdemarc(base_offset
, ap
->a_loffset
);
2390 if (last_offset
!= ap
->a_loffset
&&
2391 hammer_blocksize(last_offset
- 1) != blksize
) {
2392 last_offset
= hammer_blockdemarc(ap
->a_loffset
,
2397 * Returning EOPNOTSUPP simply prevents the direct-IO optimization
2400 disk_offset
= base_disk_offset
+ (ap
->a_loffset
- base_offset
);
2403 * If doffsetp is not aligned or the forward run size does
2404 * not cover a whole buffer, disallow the direct I/O.
2406 if ((disk_offset
& HAMMER_BUFMASK
) ||
2407 (last_offset
- ap
->a_loffset
) < blksize
) {
2410 *ap
->a_doffsetp
= disk_offset
;
2412 *ap
->a_runb
= ap
->a_loffset
- base_offset
;
2413 KKASSERT(*ap
->a_runb
>= 0);
2416 *ap
->a_runp
= last_offset
- ap
->a_loffset
;
2417 KKASSERT(*ap
->a_runp
>= 0);
2425 * Write to a regular file. Because this is a strategy call the OS is
2426 * trying to actually get data onto the media.
2430 hammer_vop_strategy_write(struct vop_strategy_args
*ap
)
2432 hammer_record_t record
;
2442 ip
= ap
->a_vp
->v_data
;
2445 KKASSERT(bp
->b_bufsize
== hammer_blocksize(bio
->bio_offset
));
2447 if (ip
->flags
& HAMMER_INODE_RO
) {
2448 bp
->b_error
= EROFS
;
2449 bp
->b_flags
|= B_ERROR
;
2451 hammer_cleanup_write_io(ip
);
2456 * Interlock with inode destruction (no in-kernel or directory
2457 * topology visibility). If we queue new IO while trying to
2458 * destroy the inode we can deadlock the vtrunc call in
2459 * hammer_inode_unloadable_check().
2461 if (ip
->flags
& (HAMMER_INODE_DELETING
|HAMMER_INODE_DELETED
)) {
2464 hammer_cleanup_write_io(ip
);
2469 * Reserve space and issue a direct-write from the front-end.
2470 * NOTE: The direct_io code will hammer_bread/bcopy smaller
2473 * An in-memory record will be installed to reference the storage
2474 * until the flusher can get to it.
2476 * Since we own the high level bio the front-end will not try to
2477 * do a direct-read until the write completes.
2479 * NOTE: The only time we do not reserve a full-sized buffers
2480 * worth of data is if the file is small. We do not try to
2481 * allocate a fragment (from the small-data zone) at the end of
2482 * an otherwise large file as this can lead to wildly separated
2485 KKASSERT((bio
->bio_offset
& HAMMER_BUFMASK
) == 0);
2486 KKASSERT(bio
->bio_offset
< ip
->ino_data
.size
);
2487 if (bio
->bio_offset
|| ip
->ino_data
.size
> HAMMER_BUFSIZE
/ 2)
2488 bytes
= bp
->b_bufsize
;
2490 bytes
= ((int)ip
->ino_data
.size
+ 15) & ~15;
2492 record
= hammer_ip_add_bulk(ip
, bio
->bio_offset
, bp
->b_data
,
2495 hammer_io_direct_write(hmp
, &record
->leaf
, bio
);
2496 hammer_rel_mem_record(record
);
2497 if (ip
->rsv_recs
> 1 && hmp
->rsv_recs
> hammer_limit_recs
)
2498 hammer_flush_inode(ip
, 0);
2500 bp
->b_bio2
.bio_offset
= NOOFFSET
;
2501 bp
->b_error
= error
;
2502 bp
->b_flags
|= B_ERROR
;
2505 hammer_cleanup_write_io(ip
);
2510 * Clean-up after disposing of a dirty frontend buffer's data.
2511 * This is somewhat heuristical so try to be robust.
2514 hammer_cleanup_write_io(hammer_inode_t ip
)
2516 if (ip
->rsv_databufs
) {
2518 --ip
->hmp
->rsv_databufs
;
2523 * We can lose track of dirty buffer cache buffers if we truncate, this
2524 * routine will resynchronize the count.
2528 hammer_update_rsv_databufs(hammer_inode_t ip
)
2536 RB_FOREACH(bp
, buf_rb_tree
, &ip
->vp
->v_rbdirty_tree
) {
2542 delta
= n
- ip
->rsv_databufs
;
2543 ip
->rsv_databufs
+= delta
;
2544 ip
->hmp
->rsv_databufs
+= delta
;
2548 * dounlink - disconnect a directory entry
2550 * XXX whiteout support not really in yet
2553 hammer_dounlink(hammer_transaction_t trans
, struct nchandle
*nch
,
2554 struct vnode
*dvp
, struct ucred
*cred
, int flags
)
2556 struct namecache
*ncp
;
2559 struct hammer_cursor cursor
;
2564 * Calculate the namekey and setup the key range for the scan. This
2565 * works kinda like a chained hash table where the lower 32 bits
2566 * of the namekey synthesize the chain.
2568 * The key range is inclusive of both key_beg and key_end.
2573 if (dip
->flags
& HAMMER_INODE_RO
)
2576 namekey
= hammer_directory_namekey(ncp
->nc_name
, ncp
->nc_nlen
);
2578 hammer_init_cursor(trans
, &cursor
, &dip
->cache
[1], dip
);
2579 cursor
.key_beg
.localization
= dip
->obj_localization
+
2580 HAMMER_LOCALIZE_MISC
;
2581 cursor
.key_beg
.obj_id
= dip
->obj_id
;
2582 cursor
.key_beg
.key
= namekey
;
2583 cursor
.key_beg
.create_tid
= 0;
2584 cursor
.key_beg
.delete_tid
= 0;
2585 cursor
.key_beg
.rec_type
= HAMMER_RECTYPE_DIRENTRY
;
2586 cursor
.key_beg
.obj_type
= 0;
2588 cursor
.key_end
= cursor
.key_beg
;
2589 cursor
.key_end
.key
|= 0xFFFFFFFFULL
;
2590 cursor
.asof
= dip
->obj_asof
;
2591 cursor
.flags
|= HAMMER_CURSOR_END_INCLUSIVE
| HAMMER_CURSOR_ASOF
;
2594 * Scan all matching records (the chain), locate the one matching
2595 * the requested path component. info->last_error contains the
2596 * error code on search termination and could be 0, ENOENT, or
2599 * The hammer_ip_*() functions merge in-memory records with on-disk
2600 * records for the purposes of the search.
2602 error
= hammer_ip_first(&cursor
);
2604 while (error
== 0) {
2605 error
= hammer_ip_resolve_data(&cursor
);
2608 nlen
= cursor
.leaf
->data_len
- HAMMER_ENTRY_NAME_OFF
;
2610 if (ncp
->nc_nlen
== nlen
&&
2611 bcmp(ncp
->nc_name
, cursor
.data
->entry
.name
, nlen
) == 0) {
2614 error
= hammer_ip_next(&cursor
);
2618 * If all is ok we have to get the inode so we can adjust nlinks.
2619 * To avoid a deadlock with the flusher we must release the inode
2620 * lock on the directory when acquiring the inode for the entry.
2622 * If the target is a directory, it must be empty.
2625 hammer_unlock(&cursor
.ip
->lock
);
2626 ip
= hammer_get_inode(trans
, dip
, cursor
.data
->entry
.obj_id
,
2628 cursor
.data
->entry
.localization
,
2630 hammer_lock_sh(&cursor
.ip
->lock
);
2631 if (error
== ENOENT
) {
2632 kprintf("obj_id %016llx\n", cursor
.data
->entry
.obj_id
);
2633 Debugger("ENOENT unlinking object that should exist");
2637 * If we are trying to remove a directory the directory must
2640 * WARNING: hammer_ip_check_directory_empty() may have to
2641 * terminate the cursor to avoid a deadlock. It is ok to
2642 * call hammer_done_cursor() twice.
2644 if (error
== 0 && ip
->ino_data
.obj_type
==
2645 HAMMER_OBJTYPE_DIRECTORY
) {
2646 error
= hammer_ip_check_directory_empty(trans
, ip
);
2650 * Delete the directory entry.
2652 * WARNING: hammer_ip_del_directory() may have to terminate
2653 * the cursor to avoid a deadlock. It is ok to call
2654 * hammer_done_cursor() twice.
2657 error
= hammer_ip_del_directory(trans
, &cursor
,
2660 hammer_done_cursor(&cursor
);
2662 cache_setunresolved(nch
);
2663 cache_setvp(nch
, NULL
);
2666 cache_inval_vp(ip
->vp
, CINV_DESTROY
);
2669 hammer_rel_inode(ip
, 0);
2671 hammer_done_cursor(&cursor
);
2673 if (error
== EDEADLK
)
2679 /************************************************************************
2680 * FIFO AND SPECFS OPS *
2681 ************************************************************************
2686 hammer_vop_fifoclose (struct vop_close_args
*ap
)
2688 /* XXX update itimes */
2689 return (VOCALL(&fifo_vnode_vops
, &ap
->a_head
));
2693 hammer_vop_fiforead (struct vop_read_args
*ap
)
2697 error
= VOCALL(&fifo_vnode_vops
, &ap
->a_head
);
2698 /* XXX update access time */
2703 hammer_vop_fifowrite (struct vop_write_args
*ap
)
2707 error
= VOCALL(&fifo_vnode_vops
, &ap
->a_head
);
2708 /* XXX update access time */
2713 hammer_vop_specclose (struct vop_close_args
*ap
)
2715 /* XXX update itimes */
2716 return (VOCALL(&spec_vnode_vops
, &ap
->a_head
));
2720 hammer_vop_specread (struct vop_read_args
*ap
)
2722 /* XXX update access time */
2723 return (VOCALL(&spec_vnode_vops
, &ap
->a_head
));
2727 hammer_vop_specwrite (struct vop_write_args
*ap
)
2729 /* XXX update last change time */
2730 return (VOCALL(&spec_vnode_vops
, &ap
->a_head
));