2 * Copyright (c) 2011-2015 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@dragonflybsd.org>
6 * by Venkatesh Srinivas <vsrinivas@dragonflybsd.org>
7 * by Daniel Flores (GSOC 2013 - mentored by Matthew Dillon, compression)
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in
17 * the documentation and/or other materials provided with the
19 * 3. Neither the name of The DragonFly Project nor the names of its
20 * contributors may be used to endorse or promote products derived
21 * from this software without specific, prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
26 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
27 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
28 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
29 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
30 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
31 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
32 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
33 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * Kernel Filesystem interface
39 * NOTE! local ipdata pointers must be reloaded on any modifying operation
40 * to the inode as its underlying chain may have changed.
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/fcntl.h>
49 #include <sys/namei.h>
50 #include <sys/mount.h>
51 #include <sys/vnode.h>
52 #include <sys/mountctl.h>
53 #include <sys/dirent.h>
55 #include <sys/objcache.h>
56 #include <sys/event.h>
58 #include <vfs/fifofs/fifo.h>
62 static int hammer2_read_file(hammer2_inode_t
*ip
, struct uio
*uio
,
64 static int hammer2_write_file(hammer2_inode_t
*ip
, struct uio
*uio
,
65 int ioflag
, int seqcount
);
66 static void hammer2_extend_file(hammer2_inode_t
*ip
, hammer2_key_t nsize
);
67 static void hammer2_truncate_file(hammer2_inode_t
*ip
, hammer2_key_t nsize
);
69 struct objcache
*cache_xops
;
73 hammer2_knote(struct vnode
*vp
, int flags
)
76 KNOTE(&vp
->v_pollinfo
.vpi_kqinfo
.ki_note
, flags
);
80 * Last reference to a vnode is going away but it is still cached.
84 hammer2_vop_inactive(struct vop_inactive_args
*ap
)
103 * Check for deleted inodes and recycle immediately on the last
104 * release. Be sure to destroy any left-over buffer cache buffers
105 * so we do not waste time trying to flush them.
107 * Note that deleting the file block chains under the inode chain
108 * would just be a waste of energy, so don't do it.
110 * WARNING: nvtruncbuf() can only be safely called without the inode
111 * lock held due to the way our write thread works.
113 if (ip
->flags
& HAMMER2_INODE_ISUNLINKED
) {
118 * Detect updates to the embedded data which may be
119 * synchronized by the strategy code. Simply mark the
120 * inode modified so it gets picked up by our normal flush.
122 nblksize
= hammer2_calc_logical(ip
, 0, &lbase
, NULL
);
123 nvtruncbuf(vp
, 0, nblksize
, 0, 0);
131 * Reclaim a vnode so that it can be reused; after the inode is
132 * disassociated, the filesystem must manage it alone.
136 hammer2_vop_reclaim(struct vop_reclaim_args
*ap
)
152 * The final close of a deleted file or directory marks it for
153 * destruction. The DELETED flag allows the flusher to shortcut
154 * any modified blocks still unflushed (that is, just ignore them).
156 * HAMMER2 usually does not try to optimize the freemap by returning
157 * deleted blocks to it as it does not usually know how many snapshots
158 * might be referencing portions of the file/dir.
164 * NOTE! We do not attempt to flush chains here, flushing is
165 * really fragile and could also deadlock.
170 * This occurs if the inode was unlinked while open. Reclamation of
171 * these inodes requires processing we cannot safely do here so add
172 * the inode to the sideq in that situation.
174 * A modified inode may require chain synchronization which will no
175 * longer be driven by a sync or fsync without the vnode, also use
176 * the sideq for that.
178 * A reclaim can occur at any time so we cannot safely start a
179 * transaction to handle reclamation of unlinked files. Instead,
180 * the ip is left with a reference and placed on a linked list and
184 if ((ip
->flags
& (HAMMER2_INODE_ISUNLINKED
|
185 HAMMER2_INODE_MODIFIED
|
186 HAMMER2_INODE_RESIZED
)) &&
187 (ip
->flags
& HAMMER2_INODE_ISDELETED
) == 0) {
188 hammer2_inode_sideq_t
*ipul
;
190 ipul
= kmalloc(sizeof(*ipul
), pmp
->minode
, M_WAITOK
| M_ZERO
);
193 hammer2_spin_ex(&pmp
->list_spin
);
194 if ((ip
->flags
& HAMMER2_INODE_ONSIDEQ
) == 0) {
196 atomic_set_int(&ip
->flags
, HAMMER2_INODE_ONSIDEQ
);
197 TAILQ_INSERT_TAIL(&pmp
->sideq
, ipul
, entry
);
198 hammer2_spin_unex(&pmp
->list_spin
);
200 hammer2_spin_unex(&pmp
->list_spin
);
201 kfree(ipul
, pmp
->minode
);
202 hammer2_inode_drop(ip
); /* vp ref */
204 /* retain ref from vp for ipul */
206 hammer2_inode_drop(ip
); /* vp ref */
210 * XXX handle background sync when ip dirty, kernel will no longer
211 * notify us regarding this inode because there is no longer a
212 * vnode attached to it.
221 hammer2_vop_fsync(struct vop_fsync_args
*ap
)
231 /* XXX can't do this yet */
232 hammer2_trans_init(ip
->pmp
, HAMMER2_TRANS_ISFLUSH
);
233 vfsync(vp
, ap
->a_waitfor
, 1, NULL
, NULL
);
235 hammer2_trans_init(ip
->pmp
, 0);
236 vfsync(vp
, ap
->a_waitfor
, 1, NULL
, NULL
);
239 * Calling chain_flush here creates a lot of duplicative
240 * COW operations due to non-optimal vnode ordering.
242 * Only do it for an actual fsync() syscall. The other forms
243 * which call this function will eventually call chain_flush
244 * on the volume root as a catch-all, which is far more optimal.
246 hammer2_inode_lock(ip
, 0);
247 if (ip
->flags
& HAMMER2_INODE_MODIFIED
)
248 hammer2_inode_chain_sync(ip
);
249 hammer2_inode_unlock(ip
);
250 hammer2_trans_done(ip
->pmp
);
258 hammer2_vop_access(struct vop_access_args
*ap
)
260 hammer2_inode_t
*ip
= VTOI(ap
->a_vp
);
266 hammer2_inode_lock(ip
, HAMMER2_RESOLVE_SHARED
);
267 uid
= hammer2_to_unix_xid(&ip
->meta
.uid
);
268 gid
= hammer2_to_unix_xid(&ip
->meta
.gid
);
269 error
= vop_helper_access(ap
, uid
, gid
, ip
->meta
.mode
, ip
->meta
.uflags
);
270 hammer2_inode_unlock(ip
);
278 hammer2_vop_getattr(struct vop_getattr_args
*ap
)
284 hammer2_chain_t
*chain
;
294 hammer2_inode_lock(ip
, HAMMER2_RESOLVE_SHARED
);
296 vap
->va_fsid
= pmp
->mp
->mnt_stat
.f_fsid
.val
[0];
297 vap
->va_fileid
= ip
->meta
.inum
;
298 vap
->va_mode
= ip
->meta
.mode
;
299 vap
->va_nlink
= ip
->meta
.nlinks
;
300 vap
->va_uid
= hammer2_to_unix_xid(&ip
->meta
.uid
);
301 vap
->va_gid
= hammer2_to_unix_xid(&ip
->meta
.gid
);
304 vap
->va_size
= ip
->meta
.size
; /* protected by shared lock */
305 vap
->va_blocksize
= HAMMER2_PBUFSIZE
;
306 vap
->va_flags
= ip
->meta
.uflags
;
307 hammer2_time_to_timespec(ip
->meta
.ctime
, &vap
->va_ctime
);
308 hammer2_time_to_timespec(ip
->meta
.mtime
, &vap
->va_mtime
);
309 hammer2_time_to_timespec(ip
->meta
.mtime
, &vap
->va_atime
);
312 if (ip
->meta
.type
== HAMMER2_OBJTYPE_DIRECTORY
) {
314 * Can't really calculate directory use sans the files under
315 * it, just assume one block for now.
317 vap
->va_bytes
+= HAMMER2_INODE_BYTES
;
319 for (i
= 0; i
< ip
->cluster
.nchains
; ++i
) {
320 if ((chain
= ip
->cluster
.array
[i
].chain
) != NULL
) {
322 chain
->bref
.embed
.stats
.data_count
) {
324 chain
->bref
.embed
.stats
.data_count
;
329 vap
->va_type
= hammer2_get_vtype(ip
->meta
.type
);
331 vap
->va_uid_uuid
= ip
->meta
.uid
;
332 vap
->va_gid_uuid
= ip
->meta
.gid
;
333 vap
->va_vaflags
= VA_UID_UUID_VALID
| VA_GID_UUID_VALID
|
336 hammer2_inode_unlock(ip
);
344 hammer2_vop_setattr(struct vop_setattr_args
*ap
)
356 hammer2_update_time(&ctime
);
360 if (ip
->pmp
->ronly
) {
365 hammer2_pfs_memory_wait(ip
->pmp
);
366 hammer2_trans_init(ip
->pmp
, 0);
367 hammer2_inode_lock(ip
, 0);
370 if (vap
->va_flags
!= VNOVAL
) {
373 flags
= ip
->meta
.uflags
;
374 error
= vop_helper_setattr_flags(&flags
, vap
->va_flags
,
375 hammer2_to_unix_xid(&ip
->meta
.uid
),
378 if (ip
->meta
.uflags
!= flags
) {
379 hammer2_inode_modify(ip
);
380 ip
->meta
.uflags
= flags
;
381 ip
->meta
.ctime
= ctime
;
382 kflags
|= NOTE_ATTRIB
;
384 if (ip
->meta
.uflags
& (IMMUTABLE
| APPEND
)) {
391 if (ip
->meta
.uflags
& (IMMUTABLE
| APPEND
)) {
395 if (vap
->va_uid
!= (uid_t
)VNOVAL
|| vap
->va_gid
!= (gid_t
)VNOVAL
) {
396 mode_t cur_mode
= ip
->meta
.mode
;
397 uid_t cur_uid
= hammer2_to_unix_xid(&ip
->meta
.uid
);
398 gid_t cur_gid
= hammer2_to_unix_xid(&ip
->meta
.gid
);
402 error
= vop_helper_chown(ap
->a_vp
, vap
->va_uid
, vap
->va_gid
,
404 &cur_uid
, &cur_gid
, &cur_mode
);
406 hammer2_guid_to_uuid(&uuid_uid
, cur_uid
);
407 hammer2_guid_to_uuid(&uuid_gid
, cur_gid
);
408 if (bcmp(&uuid_uid
, &ip
->meta
.uid
, sizeof(uuid_uid
)) ||
409 bcmp(&uuid_gid
, &ip
->meta
.gid
, sizeof(uuid_gid
)) ||
410 ip
->meta
.mode
!= cur_mode
412 hammer2_inode_modify(ip
);
413 ip
->meta
.uid
= uuid_uid
;
414 ip
->meta
.gid
= uuid_gid
;
415 ip
->meta
.mode
= cur_mode
;
416 ip
->meta
.ctime
= ctime
;
418 kflags
|= NOTE_ATTRIB
;
425 if (vap
->va_size
!= VNOVAL
&& ip
->meta
.size
!= vap
->va_size
) {
428 if (vap
->va_size
== ip
->meta
.size
)
430 if (vap
->va_size
< ip
->meta
.size
) {
431 hammer2_mtx_ex(&ip
->truncate_lock
);
432 hammer2_truncate_file(ip
, vap
->va_size
);
433 hammer2_mtx_unlock(&ip
->truncate_lock
);
435 hammer2_extend_file(ip
, vap
->va_size
);
437 hammer2_inode_modify(ip
);
438 ip
->meta
.mtime
= ctime
;
446 /* atime not supported */
447 if (vap
->va_atime
.tv_sec
!= VNOVAL
) {
448 hammer2_inode_modify(ip
);
449 ip
->meta
.atime
= hammer2_timespec_to_time(&vap
->va_atime
);
450 kflags
|= NOTE_ATTRIB
;
453 if (vap
->va_mode
!= (mode_t
)VNOVAL
) {
454 mode_t cur_mode
= ip
->meta
.mode
;
455 uid_t cur_uid
= hammer2_to_unix_xid(&ip
->meta
.uid
);
456 gid_t cur_gid
= hammer2_to_unix_xid(&ip
->meta
.gid
);
458 error
= vop_helper_chmod(ap
->a_vp
, vap
->va_mode
, ap
->a_cred
,
459 cur_uid
, cur_gid
, &cur_mode
);
460 if (error
== 0 && ip
->meta
.mode
!= cur_mode
) {
461 hammer2_inode_modify(ip
);
462 ip
->meta
.mode
= cur_mode
;
463 ip
->meta
.ctime
= ctime
;
464 kflags
|= NOTE_ATTRIB
;
468 if (vap
->va_mtime
.tv_sec
!= VNOVAL
) {
469 hammer2_inode_modify(ip
);
470 ip
->meta
.mtime
= hammer2_timespec_to_time(&vap
->va_mtime
);
471 kflags
|= NOTE_ATTRIB
;
476 * If a truncation occurred we must call inode_fsync() now in order
477 * to trim the related data chains, otherwise a later expansion can
480 * If an extend occured that changed the DIRECTDATA state, we must
481 * call inode_fsync now in order to prepare the inode's indirect
484 if (ip
->flags
& HAMMER2_INODE_RESIZED
)
485 hammer2_inode_chain_sync(ip
);
490 hammer2_inode_unlock(ip
);
491 hammer2_trans_done(ip
->pmp
);
492 hammer2_knote(ip
->vp
, kflags
);
500 hammer2_vop_readdir(struct vop_readdir_args
*ap
)
502 hammer2_xop_readdir_t
*xop
;
503 hammer2_blockref_t bref
;
520 saveoff
= uio
->uio_offset
;
525 * Setup cookies directory entry cookies if requested
527 if (ap
->a_ncookies
) {
528 ncookies
= uio
->uio_resid
/ 16 + 1;
531 cookies
= kmalloc(ncookies
* sizeof(off_t
), M_TEMP
, M_WAITOK
);
538 hammer2_inode_lock(ip
, HAMMER2_RESOLVE_SHARED
);
541 * Handle artificial entries. To ensure that only positive 64 bit
542 * quantities are returned to userland we always strip off bit 63.
543 * The hash code is designed such that codes 0x0000-0x7FFF are not
544 * used, allowing us to use these codes for articial entries.
546 * Entry 0 is used for '.' and entry 1 is used for '..'. Do not
547 * allow '..' to cross the mount point into (e.g.) the super-root.
550 inum
= ip
->meta
.inum
& HAMMER2_DIRHASH_USERMSK
;
551 r
= vop_write_dirent(&error
, uio
, inum
, DT_DIR
, 1, ".");
555 cookies
[cookie_index
] = saveoff
;
558 if (cookie_index
== ncookies
)
564 * Be careful with lockorder when accessing ".."
566 * (ip is the current dir. xip is the parent dir).
568 inum
= ip
->meta
.inum
& HAMMER2_DIRHASH_USERMSK
;
569 if (ip
!= ip
->pmp
->iroot
)
570 inum
= ip
->meta
.iparent
& HAMMER2_DIRHASH_USERMSK
;
571 r
= vop_write_dirent(&error
, uio
, inum
, DT_DIR
, 2, "..");
575 cookies
[cookie_index
] = saveoff
;
578 if (cookie_index
== ncookies
)
582 lkey
= saveoff
| HAMMER2_DIRHASH_VISIBLE
;
583 if (hammer2_debug
& 0x0020)
584 kprintf("readdir: lkey %016jx\n", lkey
);
589 * Use XOP for cluster scan.
591 * parent is the inode cluster, already locked for us. Don't
592 * double lock shared locks as this will screw up upgrades.
594 xop
= hammer2_xop_alloc(ip
, 0);
596 hammer2_xop_start(&xop
->head
, hammer2_xop_readdir
);
599 const hammer2_inode_data_t
*ripdata
;
601 error
= hammer2_xop_collect(&xop
->head
, 0);
604 if (cookie_index
== ncookies
)
606 if (hammer2_debug
& 0x0020)
607 kprintf("cluster chain %p %p\n",
608 xop
->head
.cluster
.focus
,
609 (xop
->head
.cluster
.focus
?
610 xop
->head
.cluster
.focus
->data
: (void *)-1));
611 ripdata
= &hammer2_cluster_rdata(&xop
->head
.cluster
)->ipdata
;
612 hammer2_cluster_bref(&xop
->head
.cluster
, &bref
);
613 if (bref
.type
== HAMMER2_BREF_TYPE_INODE
) {
614 dtype
= hammer2_get_dtype(ripdata
);
615 saveoff
= bref
.key
& HAMMER2_DIRHASH_USERMSK
;
616 r
= vop_write_dirent(&error
, uio
,
618 HAMMER2_DIRHASH_USERMSK
,
620 ripdata
->meta
.name_len
,
625 cookies
[cookie_index
] = saveoff
;
628 /* XXX chain error */
629 kprintf("bad chain type readdir %d\n", bref
.type
);
632 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
633 if (error
== ENOENT
) {
636 saveoff
= (hammer2_key_t
)-1;
638 saveoff
= bref
.key
& HAMMER2_DIRHASH_USERMSK
;
641 hammer2_inode_unlock(ip
);
643 *ap
->a_eofflag
= eofflag
;
644 if (hammer2_debug
& 0x0020)
645 kprintf("readdir: done at %016jx\n", saveoff
);
646 uio
->uio_offset
= saveoff
& ~HAMMER2_DIRHASH_VISIBLE
;
647 if (error
&& cookie_index
== 0) {
649 kfree(cookies
, M_TEMP
);
651 *ap
->a_cookies
= NULL
;
655 *ap
->a_ncookies
= cookie_index
;
656 *ap
->a_cookies
= cookies
;
664 * hammer2_vop_readlink { vp, uio, cred }
668 hammer2_vop_readlink(struct vop_readlink_args
*ap
)
675 if (vp
->v_type
!= VLNK
)
679 error
= hammer2_read_file(ip
, ap
->a_uio
, 0);
685 hammer2_vop_read(struct vop_read_args
*ap
)
695 * Read operations supported on this vnode?
698 if (vp
->v_type
!= VREG
)
708 seqcount
= ap
->a_ioflag
>> 16;
709 bigread
= (uio
->uio_resid
> 100 * 1024 * 1024);
711 error
= hammer2_read_file(ip
, uio
, seqcount
);
717 hammer2_vop_write(struct vop_write_args
*ap
)
727 * Read operations supported on this vnode?
730 if (vp
->v_type
!= VREG
)
739 if (ip
->pmp
->ronly
) {
743 seqcount
= ap
->a_ioflag
>> 16;
746 * Check resource limit
748 if (uio
->uio_resid
> 0 && (td
= uio
->uio_td
) != NULL
&& td
->td_proc
&&
749 uio
->uio_offset
+ uio
->uio_resid
>
750 td
->td_proc
->p_rlimit
[RLIMIT_FSIZE
].rlim_cur
) {
751 lwpsignal(td
->td_proc
, td
->td_lwp
, SIGXFSZ
);
756 * The transaction interlocks against flush initiations
757 * (note: but will run concurrently with the actual flush).
759 * To avoid deadlocking against the VM system, we must flag any
760 * transaction related to the buffer cache or other direct
761 * VM page manipulation.
763 if (uio
->uio_segflg
== UIO_NOCOPY
)
764 hammer2_trans_init(ip
->pmp
, HAMMER2_TRANS_BUFCACHE
);
766 hammer2_trans_init(ip
->pmp
, 0);
767 error
= hammer2_write_file(ip
, uio
, ap
->a_ioflag
, seqcount
);
768 hammer2_trans_done(ip
->pmp
);
774 * Perform read operations on a file or symlink given an UNLOCKED
777 * The passed ip is not locked.
781 hammer2_read_file(hammer2_inode_t
*ip
, struct uio
*uio
, int seqcount
)
792 * WARNING! Assumes that the kernel interlocks size changes at the
795 hammer2_mtx_sh(&ip
->lock
);
796 hammer2_mtx_sh(&ip
->truncate_lock
);
797 size
= ip
->meta
.size
;
798 hammer2_mtx_unlock(&ip
->lock
);
800 while (uio
->uio_resid
> 0 && uio
->uio_offset
< size
) {
807 lblksize
= hammer2_calc_logical(ip
, uio
->uio_offset
,
811 error
= cluster_read(ip
->vp
, leof
, lbase
, lblksize
,
812 uio
->uio_resid
, seqcount
* MAXBSIZE
,
815 if (uio
->uio_segflg
== UIO_NOCOPY
) {
816 bp
= getblk(ip
->vp
, lbase
, lblksize
, GETBLK_BHEAVY
, 0);
817 if (bp
->b_flags
& B_CACHE
) {
820 if (bp
->b_xio
.xio_npages
!= 16)
821 kprintf("NPAGES BAD\n");
822 for (i
= 0; i
< bp
->b_xio
.xio_npages
; ++i
) {
824 m
= bp
->b_xio
.xio_pages
[i
];
825 if (m
== NULL
|| m
->valid
== 0) {
826 kprintf("bp %016jx %016jx pg %d inv",
829 kprintf("m->object %p/%p", m
->object
, ip
->vp
->v_object
);
835 kprintf("b_flags %08x, b_error %d\n", bp
->b_flags
, bp
->b_error
);
839 error
= bread(ip
->vp
, lbase
, lblksize
, &bp
);
845 loff
= (int)(uio
->uio_offset
- lbase
);
847 if (n
> uio
->uio_resid
)
849 if (n
> size
- uio
->uio_offset
)
850 n
= (int)(size
- uio
->uio_offset
);
851 bp
->b_flags
|= B_AGE
;
852 uiomovebp(bp
, (char *)bp
->b_data
+ loff
, n
, uio
);
855 hammer2_mtx_unlock(&ip
->truncate_lock
);
861 * Write to the file represented by the inode via the logical buffer cache.
862 * The inode may represent a regular file or a symlink.
864 * The inode must not be locked.
868 hammer2_write_file(hammer2_inode_t
*ip
, struct uio
*uio
,
869 int ioflag
, int seqcount
)
871 hammer2_key_t old_eof
;
872 hammer2_key_t new_eof
;
881 * WARNING! Assumes that the kernel interlocks size changes at the
884 hammer2_mtx_ex(&ip
->lock
);
885 hammer2_mtx_sh(&ip
->truncate_lock
);
886 if (ioflag
& IO_APPEND
)
887 uio
->uio_offset
= ip
->meta
.size
;
888 old_eof
= ip
->meta
.size
;
891 * Extend the file if necessary. If the write fails at some point
892 * we will truncate it back down to cover as much as we were able
895 * Doing this now makes it easier to calculate buffer sizes in
902 if (uio
->uio_offset
+ uio
->uio_resid
> old_eof
) {
903 new_eof
= uio
->uio_offset
+ uio
->uio_resid
;
905 hammer2_extend_file(ip
, new_eof
);
906 kflags
|= NOTE_EXTEND
;
910 hammer2_mtx_unlock(&ip
->lock
);
915 while (uio
->uio_resid
> 0) {
924 * Don't allow the buffer build to blow out the buffer
927 if ((ioflag
& IO_RECURSE
) == 0)
928 bwillwrite(HAMMER2_PBUFSIZE
);
931 * This nominally tells us how much we can cluster and
932 * what the logical buffer size needs to be. Currently
933 * we don't try to cluster the write and just handle one
936 lblksize
= hammer2_calc_logical(ip
, uio
->uio_offset
,
938 loff
= (int)(uio
->uio_offset
- lbase
);
940 KKASSERT(lblksize
<= 65536);
943 * Calculate bytes to copy this transfer and whether the
944 * copy completely covers the buffer or not.
948 if (n
> uio
->uio_resid
) {
950 if (loff
== lbase
&& uio
->uio_offset
+ n
== new_eof
)
958 if (lbase
>= new_eof
)
964 if (uio
->uio_segflg
== UIO_NOCOPY
) {
966 * Issuing a write with the same data backing the
967 * buffer. Instantiate the buffer to collect the
968 * backing vm pages, then read-in any missing bits.
970 * This case is used by vop_stdputpages().
972 bp
= getblk(ip
->vp
, lbase
, lblksize
, GETBLK_BHEAVY
, 0);
973 if ((bp
->b_flags
& B_CACHE
) == 0) {
975 error
= bread(ip
->vp
, lbase
, lblksize
, &bp
);
977 } else if (trivial
) {
979 * Even though we are entirely overwriting the buffer
980 * we may still have to zero it out to avoid a
981 * mmap/write visibility issue.
983 bp
= getblk(ip
->vp
, lbase
, lblksize
, GETBLK_BHEAVY
, 0);
984 if ((bp
->b_flags
& B_CACHE
) == 0)
988 * Partial overwrite, read in any missing bits then
989 * replace the portion being written.
991 * (The strategy code will detect zero-fill physical
992 * blocks for this case).
994 error
= bread(ip
->vp
, lbase
, lblksize
, &bp
);
1005 * Ok, copy the data in
1007 error
= uiomovebp(bp
, bp
->b_data
+ loff
, n
, uio
);
1008 kflags
|= NOTE_WRITE
;
1016 * WARNING: Pageout daemon will issue UIO_NOCOPY writes
1017 * with IO_SYNC or IO_ASYNC set. These writes
1018 * must be handled as the pageout daemon expects.
1020 * NOTE! H2 relies on cluster_write() here because it
1021 * cannot preallocate disk blocks at the logical
1022 * level due to not knowing what the compression
1023 * size will be at this time.
1025 * We must use cluster_write() here and we depend
1026 * on the write-behind feature to flush buffers
1027 * appropriately. If we let the buffer daemons do
1028 * it the block allocations will be all over the
1031 if (ioflag
& IO_SYNC
) {
1033 } else if ((ioflag
& IO_DIRECT
) && endofblk
) {
1035 } else if (ioflag
& IO_ASYNC
) {
1037 } else if (ip
->vp
->v_mount
->mnt_flag
& MNT_NOCLUSTERW
) {
1041 bp
->b_flags
|= B_CLUSTEROK
;
1042 cluster_write(bp
, new_eof
, lblksize
, seqcount
);
1044 bp
->b_flags
|= B_CLUSTEROK
;
1051 * Cleanup. If we extended the file EOF but failed to write through
1052 * the entire write is a failure and we have to back-up.
1054 if (error
&& new_eof
!= old_eof
) {
1055 hammer2_mtx_unlock(&ip
->truncate_lock
);
1056 hammer2_mtx_ex(&ip
->lock
);
1057 hammer2_mtx_ex(&ip
->truncate_lock
);
1058 hammer2_truncate_file(ip
, old_eof
);
1059 if (ip
->flags
& HAMMER2_INODE_MODIFIED
)
1060 hammer2_inode_chain_sync(ip
);
1061 hammer2_mtx_unlock(&ip
->lock
);
1062 } else if (modified
) {
1063 hammer2_mtx_ex(&ip
->lock
);
1064 hammer2_inode_modify(ip
);
1065 hammer2_update_time(&ip
->meta
.mtime
);
1066 if (ip
->flags
& HAMMER2_INODE_MODIFIED
)
1067 hammer2_inode_chain_sync(ip
);
1068 hammer2_mtx_unlock(&ip
->lock
);
1069 hammer2_knote(ip
->vp
, kflags
);
1071 hammer2_trans_assert_strategy(ip
->pmp
);
1072 hammer2_mtx_unlock(&ip
->truncate_lock
);
1078 * Truncate the size of a file. The inode must not be locked.
1080 * We must unconditionally set HAMMER2_INODE_RESIZED to properly
1081 * ensure that any on-media data beyond the new file EOF has been destroyed.
1083 * WARNING: nvtruncbuf() can only be safely called without the inode lock
1084 * held due to the way our write thread works. If the truncation
1085 * occurs in the middle of a buffer, nvtruncbuf() is responsible
1086 * for dirtying that buffer and zeroing out trailing bytes.
1088 * WARNING! Assumes that the kernel interlocks size changes at the
1091 * WARNING! Caller assumes responsibility for removing dead blocks
1092 * if INODE_RESIZED is set.
1096 hammer2_truncate_file(hammer2_inode_t
*ip
, hammer2_key_t nsize
)
1098 hammer2_key_t lbase
;
1102 hammer2_mtx_unlock(&ip
->lock
);
1104 nblksize
= hammer2_calc_logical(ip
, nsize
, &lbase
, NULL
);
1105 nvtruncbuf(ip
->vp
, nsize
,
1106 nblksize
, (int)nsize
& (nblksize
- 1),
1109 hammer2_mtx_ex(&ip
->lock
);
1110 KKASSERT((ip
->flags
& HAMMER2_INODE_RESIZED
) == 0);
1111 ip
->osize
= ip
->meta
.size
;
1112 ip
->meta
.size
= nsize
;
1113 atomic_set_int(&ip
->flags
, HAMMER2_INODE_RESIZED
);
1114 hammer2_inode_modify(ip
);
1119 * Extend the size of a file. The inode must not be locked.
1121 * Even though the file size is changing, we do not have to set the
1122 * INODE_RESIZED bit unless the file size crosses the EMBEDDED_BYTES
1123 * boundary. When this occurs a hammer2_inode_chain_sync() is required
1124 * to prepare the inode cluster's indirect block table, otherwise
1125 * async execution of the strategy code will implode on us.
1127 * WARNING! Assumes that the kernel interlocks size changes at the
1130 * WARNING! Caller assumes responsibility for transitioning out
1131 * of the inode DIRECTDATA mode if INODE_RESIZED is set.
1135 hammer2_extend_file(hammer2_inode_t
*ip
, hammer2_key_t nsize
)
1137 hammer2_key_t lbase
;
1138 hammer2_key_t osize
;
1144 KKASSERT((ip
->flags
& HAMMER2_INODE_RESIZED
) == 0);
1145 hammer2_inode_modify(ip
);
1146 osize
= ip
->meta
.size
;
1148 ip
->meta
.size
= nsize
;
1150 if (osize
<= HAMMER2_EMBEDDED_BYTES
&& nsize
> HAMMER2_EMBEDDED_BYTES
) {
1151 atomic_set_int(&ip
->flags
, HAMMER2_INODE_RESIZED
);
1152 hammer2_inode_chain_sync(ip
);
1155 hammer2_mtx_unlock(&ip
->lock
);
1157 oblksize
= hammer2_calc_logical(ip
, osize
, &lbase
, NULL
);
1158 nblksize
= hammer2_calc_logical(ip
, nsize
, &lbase
, NULL
);
1164 hammer2_mtx_ex(&ip
->lock
);
1171 hammer2_vop_nresolve(struct vop_nresolve_args
*ap
)
1173 hammer2_xop_nresolve_t
*xop
;
1174 hammer2_inode_t
*ip
;
1175 hammer2_inode_t
*dip
;
1176 struct namecache
*ncp
;
1181 dip
= VTOI(ap
->a_dvp
);
1182 xop
= hammer2_xop_alloc(dip
, 0);
1184 ncp
= ap
->a_nch
->ncp
;
1185 hammer2_xop_setname(&xop
->head
, ncp
->nc_name
, ncp
->nc_nlen
);
1188 * Note: In DragonFly the kernel handles '.' and '..'.
1190 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1191 hammer2_xop_start(&xop
->head
, hammer2_xop_nresolve
);
1193 error
= hammer2_xop_collect(&xop
->head
, 0);
1197 ip
= hammer2_inode_get(dip
->pmp
, dip
, &xop
->head
.cluster
, -1);
1199 hammer2_inode_unlock(dip
);
1202 * Acquire the related vnode
1204 * NOTE: For error processing, only ENOENT resolves the namecache
1205 * entry to NULL, otherwise we just return the error and
1206 * leave the namecache unresolved.
1208 * NOTE: multiple hammer2_inode structures can be aliased to the
1209 * same chain element, for example for hardlinks. This
1210 * use case does not 'reattach' inode associations that
1211 * might already exist, but always allocates a new one.
1213 * WARNING: inode structure is locked exclusively via inode_get
1214 * but chain was locked shared. inode_unlock()
1215 * will handle it properly.
1218 vp
= hammer2_igetv(ip
, &error
);
1221 cache_setvp(ap
->a_nch
, vp
);
1222 } else if (error
== ENOENT
) {
1223 cache_setvp(ap
->a_nch
, NULL
);
1225 hammer2_inode_unlock(ip
);
1228 * The vp should not be released until after we've disposed
1229 * of our locks, because it might cause vop_inactive() to
1236 cache_setvp(ap
->a_nch
, NULL
);
1238 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1239 KASSERT(error
|| ap
->a_nch
->ncp
->nc_vp
!= NULL
,
1240 ("resolve error %d/%p ap %p\n",
1241 error
, ap
->a_nch
->ncp
->nc_vp
, ap
));
1249 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args
*ap
)
1251 hammer2_inode_t
*dip
;
1256 dip
= VTOI(ap
->a_dvp
);
1257 inum
= dip
->meta
.iparent
;
1261 error
= hammer2_vfs_vget(ap
->a_dvp
->v_mount
, NULL
,
1272 hammer2_vop_nmkdir(struct vop_nmkdir_args
*ap
)
1274 hammer2_inode_t
*dip
;
1275 hammer2_inode_t
*nip
;
1276 struct namecache
*ncp
;
1277 const uint8_t *name
;
1283 dip
= VTOI(ap
->a_dvp
);
1284 if (dip
->pmp
->ronly
) {
1289 ncp
= ap
->a_nch
->ncp
;
1290 name
= ncp
->nc_name
;
1291 name_len
= ncp
->nc_nlen
;
1293 hammer2_pfs_memory_wait(dip
->pmp
);
1294 hammer2_trans_init(dip
->pmp
, 0);
1296 inum
= hammer2_trans_newinum(dip
->pmp
);
1299 * Create the actual inode as a hidden file in the iroot, then
1300 * create the directory entry as a hardlink to it. The creation
1301 * of the actual inode sets its nlinks to 1 which is the value
1304 nip
= hammer2_inode_create(dip
->pmp
->iroot
, dip
, ap
->a_vap
, ap
->a_cred
,
1309 hammer2_inode_create(dip
, dip
, NULL
, NULL
,
1312 HAMMER2_OBJTYPE_HARDLINK
, nip
->meta
.type
,
1317 KKASSERT(nip
== NULL
);
1320 *ap
->a_vpp
= hammer2_igetv(nip
, &error
);
1321 hammer2_inode_unlock(nip
);
1325 * Update dip's mtime
1330 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1331 hammer2_update_time(&mtime
);
1332 hammer2_inode_modify(dip
);
1333 dip
->meta
.mtime
= mtime
;
1334 hammer2_inode_unlock(dip
);
1337 hammer2_trans_done(dip
->pmp
);
1340 cache_setunresolved(ap
->a_nch
);
1341 cache_setvp(ap
->a_nch
, *ap
->a_vpp
);
1349 hammer2_vop_open(struct vop_open_args
*ap
)
1351 return vop_stdopen(ap
);
1355 * hammer2_vop_advlock { vp, id, op, fl, flags }
1359 hammer2_vop_advlock(struct vop_advlock_args
*ap
)
1361 hammer2_inode_t
*ip
= VTOI(ap
->a_vp
);
1364 size
= ip
->meta
.size
;
1365 return (lf_advlock(ap
, &ip
->advlock
, size
));
1370 hammer2_vop_close(struct vop_close_args
*ap
)
1372 return vop_stdclose(ap
);
1376 * hammer2_vop_nlink { nch, dvp, vp, cred }
1378 * Create a hardlink from (vp) to {dvp, nch}.
1382 hammer2_vop_nlink(struct vop_nlink_args
*ap
)
1384 hammer2_inode_t
*tdip
; /* target directory to create link in */
1385 hammer2_inode_t
*ip
; /* inode we are hardlinking to */
1386 struct namecache
*ncp
;
1387 const uint8_t *name
;
1391 if (ap
->a_dvp
->v_mount
!= ap
->a_vp
->v_mount
)
1395 tdip
= VTOI(ap
->a_dvp
);
1396 if (tdip
->pmp
->ronly
) {
1401 ncp
= ap
->a_nch
->ncp
;
1402 name
= ncp
->nc_name
;
1403 name_len
= ncp
->nc_nlen
;
1406 * ip represents the file being hardlinked. The file could be a
1407 * normal file or a hardlink target if it has already been hardlinked.
1408 * (with the new semantics, it will almost always be a hardlink
1411 * Bump nlinks and potentially also create or move the hardlink
1412 * target in the parent directory common to (ip) and (tdip). The
1413 * consolidation code can modify ip->cluster. The returned cluster
1416 ip
= VTOI(ap
->a_vp
);
1417 KASSERT(ip
->pmp
, ("ip->pmp is NULL %p %p", ip
, ip
->pmp
));
1418 hammer2_pfs_memory_wait(ip
->pmp
);
1419 hammer2_trans_init(ip
->pmp
, 0);
1422 * Target should be an indexed inode or there's no way we will ever
1423 * be able to find it!
1425 KKASSERT((ip
->meta
.name_key
& HAMMER2_DIRHASH_VISIBLE
) == 0);
1430 * Can return NULL and error == EXDEV if the common parent
1431 * crosses a directory with the xlink flag set.
1433 hammer2_inode_lock(tdip
, 0);
1434 hammer2_inode_lock(ip
, 0);
1437 * Create the hardlink target and bump nlinks.
1440 hammer2_inode_create(tdip
, tdip
, NULL
, NULL
,
1443 HAMMER2_OBJTYPE_HARDLINK
, ip
->meta
.type
,
1445 hammer2_inode_modify(ip
);
1450 * Update dip's mtime
1454 hammer2_update_time(&mtime
);
1455 hammer2_inode_modify(tdip
);
1456 tdip
->meta
.mtime
= mtime
;
1458 cache_setunresolved(ap
->a_nch
);
1459 cache_setvp(ap
->a_nch
, ap
->a_vp
);
1461 hammer2_inode_unlock(ip
);
1462 hammer2_inode_unlock(tdip
);
1464 hammer2_trans_done(ip
->pmp
);
1471 * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1473 * The operating system has already ensured that the directory entry
1474 * does not exist and done all appropriate namespace locking.
1478 hammer2_vop_ncreate(struct vop_ncreate_args
*ap
)
1480 hammer2_inode_t
*dip
;
1481 hammer2_inode_t
*nip
;
1482 struct namecache
*ncp
;
1483 const uint8_t *name
;
1489 dip
= VTOI(ap
->a_dvp
);
1490 if (dip
->pmp
->ronly
) {
1495 ncp
= ap
->a_nch
->ncp
;
1496 name
= ncp
->nc_name
;
1497 name_len
= ncp
->nc_nlen
;
1498 hammer2_pfs_memory_wait(dip
->pmp
);
1499 hammer2_trans_init(dip
->pmp
, 0);
1501 inum
= hammer2_trans_newinum(dip
->pmp
);
1504 * Create the actual inode as a hidden file in the iroot, then
1505 * create the directory entry as a hardlink to it. The creation
1506 * of the actual inode sets its nlinks to 1 which is the value
1509 nip
= hammer2_inode_create(dip
->pmp
->iroot
, dip
, ap
->a_vap
, ap
->a_cred
,
1515 hammer2_inode_create(dip
, dip
, NULL
, NULL
,
1518 HAMMER2_OBJTYPE_HARDLINK
, nip
->meta
.type
,
1522 KKASSERT(nip
== NULL
);
1525 *ap
->a_vpp
= hammer2_igetv(nip
, &error
);
1526 hammer2_inode_unlock(nip
);
1530 * Update dip's mtime
1535 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1536 hammer2_update_time(&mtime
);
1537 hammer2_inode_modify(dip
);
1538 dip
->meta
.mtime
= mtime
;
1539 hammer2_inode_unlock(dip
);
1542 hammer2_trans_done(dip
->pmp
);
1545 cache_setunresolved(ap
->a_nch
);
1546 cache_setvp(ap
->a_nch
, *ap
->a_vpp
);
1553 * Make a device node (typically a fifo)
1557 hammer2_vop_nmknod(struct vop_nmknod_args
*ap
)
1559 hammer2_inode_t
*dip
;
1560 hammer2_inode_t
*nip
;
1561 struct namecache
*ncp
;
1562 const uint8_t *name
;
1568 dip
= VTOI(ap
->a_dvp
);
1569 if (dip
->pmp
->ronly
) {
1574 ncp
= ap
->a_nch
->ncp
;
1575 name
= ncp
->nc_name
;
1576 name_len
= ncp
->nc_nlen
;
1577 hammer2_pfs_memory_wait(dip
->pmp
);
1578 hammer2_trans_init(dip
->pmp
, 0);
1581 * The device node is entered as the directory entry itself and not
1582 * as a hardlink to an inode. Since one cannot obtain a
1583 * file handle on the filesystem entry representing the device, we
1584 * do not have to worry about indexing its inode.
1586 inum
= hammer2_trans_newinum(dip
->pmp
);
1587 nip
= hammer2_inode_create(dip
->pmp
->iroot
, dip
, ap
->a_vap
, ap
->a_cred
,
1592 hammer2_inode_create(dip
, dip
, NULL
, NULL
,
1595 HAMMER2_OBJTYPE_HARDLINK
, nip
->meta
.type
,
1601 KKASSERT(nip
== NULL
);
1604 *ap
->a_vpp
= hammer2_igetv(nip
, &error
);
1605 hammer2_inode_unlock(nip
);
1609 * Update dip's mtime
1614 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1615 hammer2_update_time(&mtime
);
1616 hammer2_inode_modify(dip
);
1617 dip
->meta
.mtime
= mtime
;
1618 hammer2_inode_unlock(dip
);
1621 hammer2_trans_done(dip
->pmp
);
1624 cache_setunresolved(ap
->a_nch
);
1625 cache_setvp(ap
->a_nch
, *ap
->a_vpp
);
1632 * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1636 hammer2_vop_nsymlink(struct vop_nsymlink_args
*ap
)
1638 hammer2_inode_t
*dip
;
1639 hammer2_inode_t
*nip
;
1640 struct namecache
*ncp
;
1641 const uint8_t *name
;
1646 dip
= VTOI(ap
->a_dvp
);
1647 if (dip
->pmp
->ronly
)
1650 ncp
= ap
->a_nch
->ncp
;
1651 name
= ncp
->nc_name
;
1652 name_len
= ncp
->nc_nlen
;
1653 hammer2_pfs_memory_wait(dip
->pmp
);
1654 hammer2_trans_init(dip
->pmp
, 0);
1656 ap
->a_vap
->va_type
= VLNK
; /* enforce type */
1659 * The softlink is entered into the directory itself and not
1660 * as a hardlink to an inode. Since one cannot obtain a
1661 * file handle on the softlink itself we do not have to worry
1662 * about indexing its inode.
1664 inum
= hammer2_trans_newinum(dip
->pmp
);
1666 nip
= hammer2_inode_create(dip
->pmp
->iroot
, dip
, ap
->a_vap
, ap
->a_cred
,
1671 hammer2_inode_create(dip
, dip
, NULL
, NULL
,
1674 HAMMER2_OBJTYPE_HARDLINK
, nip
->meta
.type
,
1680 KKASSERT(nip
== NULL
);
1682 hammer2_trans_done(dip
->pmp
);
1685 *ap
->a_vpp
= hammer2_igetv(nip
, &error
);
1688 * Build the softlink (~like file data) and finalize the namecache.
1695 bytes
= strlen(ap
->a_target
);
1697 hammer2_inode_unlock(nip
);
1698 bzero(&auio
, sizeof(auio
));
1699 bzero(&aiov
, sizeof(aiov
));
1700 auio
.uio_iov
= &aiov
;
1701 auio
.uio_segflg
= UIO_SYSSPACE
;
1702 auio
.uio_rw
= UIO_WRITE
;
1703 auio
.uio_resid
= bytes
;
1704 auio
.uio_iovcnt
= 1;
1705 auio
.uio_td
= curthread
;
1706 aiov
.iov_base
= ap
->a_target
;
1707 aiov
.iov_len
= bytes
;
1708 error
= hammer2_write_file(nip
, &auio
, IO_APPEND
, 0);
1709 /* XXX handle error */
1712 hammer2_inode_unlock(nip
);
1716 * Update dip's mtime
1721 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1722 hammer2_update_time(&mtime
);
1723 hammer2_inode_modify(dip
);
1724 dip
->meta
.mtime
= mtime
;
1725 hammer2_inode_unlock(dip
);
1728 hammer2_trans_done(dip
->pmp
);
1731 * Finalize namecache
1734 cache_setunresolved(ap
->a_nch
);
1735 cache_setvp(ap
->a_nch
, *ap
->a_vpp
);
1736 /* hammer2_knote(ap->a_dvp, NOTE_WRITE); */
1742 * hammer2_vop_nremove { nch, dvp, cred }
1746 hammer2_vop_nremove(struct vop_nremove_args
*ap
)
1748 hammer2_xop_unlink_t
*xop
;
1749 hammer2_inode_t
*dip
;
1750 hammer2_inode_t
*ip
;
1751 struct namecache
*ncp
;
1756 dip
= VTOI(ap
->a_dvp
);
1757 if (dip
->pmp
->ronly
) {
1762 ncp
= ap
->a_nch
->ncp
;
1764 hammer2_pfs_memory_wait(dip
->pmp
);
1765 hammer2_trans_init(dip
->pmp
, 0);
1766 hammer2_inode_lock(dip
, 0);
1769 * The unlink XOP unlinks the path from the directory and
1770 * locates and returns the cluster associated with the real inode.
1771 * We have to handle nlinks here on the frontend.
1773 xop
= hammer2_xop_alloc(dip
, HAMMER2_XOP_MODIFYING
);
1774 hammer2_xop_setname(&xop
->head
, ncp
->nc_name
, ncp
->nc_nlen
);
1775 isopen
= cache_isopen(ap
->a_nch
);
1777 xop
->dopermanent
= 0;
1778 hammer2_xop_start(&xop
->head
, hammer2_xop_unlink
);
1781 * Collect the real inode and adjust nlinks, destroy the real
1782 * inode if nlinks transitions to 0 and it was the real inode
1783 * (else it has already been removed).
1785 error
= hammer2_xop_collect(&xop
->head
, 0);
1786 hammer2_inode_unlock(dip
);
1789 ip
= hammer2_inode_get(dip
->pmp
, dip
, &xop
->head
.cluster
, -1);
1790 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1792 hammer2_inode_unlink_finisher(ip
, isopen
);
1793 hammer2_inode_unlock(ip
);
1796 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1800 * Update dip's mtime
1805 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1806 hammer2_update_time(&mtime
);
1807 hammer2_inode_modify(dip
);
1808 dip
->meta
.mtime
= mtime
;
1809 hammer2_inode_unlock(dip
);
1812 hammer2_inode_run_sideq(dip
->pmp
);
1813 hammer2_trans_done(dip
->pmp
);
1815 cache_unlink(ap
->a_nch
);
1821 * hammer2_vop_nrmdir { nch, dvp, cred }
1825 hammer2_vop_nrmdir(struct vop_nrmdir_args
*ap
)
1827 hammer2_xop_unlink_t
*xop
;
1828 hammer2_inode_t
*dip
;
1829 hammer2_inode_t
*ip
;
1830 struct namecache
*ncp
;
1835 dip
= VTOI(ap
->a_dvp
);
1836 if (dip
->pmp
->ronly
) {
1841 hammer2_pfs_memory_wait(dip
->pmp
);
1842 hammer2_trans_init(dip
->pmp
, 0);
1843 hammer2_inode_lock(dip
, 0);
1845 xop
= hammer2_xop_alloc(dip
, HAMMER2_XOP_MODIFYING
);
1847 ncp
= ap
->a_nch
->ncp
;
1848 hammer2_xop_setname(&xop
->head
, ncp
->nc_name
, ncp
->nc_nlen
);
1849 isopen
= cache_isopen(ap
->a_nch
);
1851 xop
->dopermanent
= 0;
1852 hammer2_xop_start(&xop
->head
, hammer2_xop_unlink
);
1855 * Collect the real inode and adjust nlinks, destroy the real
1856 * inode if nlinks transitions to 0 and it was the real inode
1857 * (else it has already been removed).
1859 error
= hammer2_xop_collect(&xop
->head
, 0);
1860 hammer2_inode_unlock(dip
);
1863 ip
= hammer2_inode_get(dip
->pmp
, dip
, &xop
->head
.cluster
, -1);
1864 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1866 hammer2_inode_unlink_finisher(ip
, isopen
);
1867 hammer2_inode_unlock(ip
);
1870 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1874 * Update dip's mtime
1879 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1880 hammer2_update_time(&mtime
);
1881 hammer2_inode_modify(dip
);
1882 dip
->meta
.mtime
= mtime
;
1883 hammer2_inode_unlock(dip
);
1886 hammer2_inode_run_sideq(dip
->pmp
);
1887 hammer2_trans_done(dip
->pmp
);
1889 cache_unlink(ap
->a_nch
);
1895 * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1899 hammer2_vop_nrename(struct vop_nrename_args
*ap
)
1901 struct namecache
*fncp
;
1902 struct namecache
*tncp
;
1903 hammer2_inode_t
*fdip
;
1904 hammer2_inode_t
*tdip
;
1905 hammer2_inode_t
*ip
;
1906 const uint8_t *fname
;
1908 const uint8_t *tname
;
1916 if (ap
->a_fdvp
->v_mount
!= ap
->a_tdvp
->v_mount
)
1918 if (ap
->a_fdvp
->v_mount
!= ap
->a_fnch
->ncp
->nc_vp
->v_mount
)
1921 fdip
= VTOI(ap
->a_fdvp
); /* source directory */
1922 tdip
= VTOI(ap
->a_tdvp
); /* target directory */
1924 if (fdip
->pmp
->ronly
)
1928 fncp
= ap
->a_fnch
->ncp
; /* entry name in source */
1929 fname
= fncp
->nc_name
;
1930 fname_len
= fncp
->nc_nlen
;
1932 tncp
= ap
->a_tnch
->ncp
; /* entry name in target */
1933 tname
= tncp
->nc_name
;
1934 tname_len
= tncp
->nc_nlen
;
1936 hammer2_pfs_memory_wait(tdip
->pmp
);
1937 hammer2_trans_init(tdip
->pmp
, 0);
1943 * ip is the inode being renamed. If this is a hardlink then
1944 * ip represents the actual file and not the hardlink marker.
1946 ip
= VTOI(fncp
->nc_vp
);
1948 KKASSERT((ip
->meta
.name_key
& HAMMER2_DIRHASH_VISIBLE
) == 0);
1951 * Can return NULL and error == EXDEV if the common parent
1952 * crosses a directory with the xlink flag set.
1955 hammer2_inode_lock(fdip
, 0);
1956 hammer2_inode_lock(tdip
, 0);
1957 hammer2_inode_ref(ip
); /* extra ref */
1959 hammer2_inode_lock(ip
, 0);
1962 * Delete the target namespace.
1965 hammer2_xop_unlink_t
*xop2
;
1966 hammer2_inode_t
*tip
;
1970 * The unlink XOP unlinks the path from the directory and
1971 * locates and returns the cluster associated with the real
1972 * inode. We have to handle nlinks here on the frontend.
1974 xop2
= hammer2_xop_alloc(tdip
, HAMMER2_XOP_MODIFYING
);
1975 hammer2_xop_setname(&xop2
->head
, tname
, tname_len
);
1976 isopen
= cache_isopen(ap
->a_tnch
);
1978 xop2
->dopermanent
= 0;
1979 hammer2_xop_start(&xop2
->head
, hammer2_xop_unlink
);
1982 * Collect the real inode and adjust nlinks, destroy the real
1983 * inode if nlinks transitions to 0 and it was the real inode
1984 * (else it has already been removed).
1986 tnch_error
= hammer2_xop_collect(&xop2
->head
, 0);
1987 /* hammer2_inode_unlock(tdip); */
1989 if (tnch_error
== 0) {
1990 tip
= hammer2_inode_get(tdip
->pmp
, NULL
,
1991 &xop2
->head
.cluster
, -1);
1992 hammer2_xop_retire(&xop2
->head
, HAMMER2_XOPMASK_VOP
);
1994 hammer2_inode_unlink_finisher(tip
, isopen
);
1995 hammer2_inode_unlock(tip
);
1998 hammer2_xop_retire(&xop2
->head
, HAMMER2_XOPMASK_VOP
);
2000 /* hammer2_inode_lock(tdip, 0); */
2002 if (tnch_error
&& tnch_error
!= ENOENT
) {
2010 * Resolve the collision space for (tdip, tname, tname_len)
2012 * tdip must be held exclusively locked to prevent races.
2015 hammer2_xop_scanlhc_t
*sxop
;
2016 hammer2_tid_t lhcbase
;
2018 tlhc
= hammer2_dirhash(tname
, tname_len
);
2020 sxop
= hammer2_xop_alloc(tdip
, HAMMER2_XOP_MODIFYING
);
2022 hammer2_xop_start(&sxop
->head
, hammer2_xop_scanlhc
);
2023 while ((error
= hammer2_xop_collect(&sxop
->head
, 0)) == 0) {
2024 if (tlhc
!= sxop
->head
.cluster
.focus
->bref
.key
)
2028 hammer2_xop_retire(&sxop
->head
, HAMMER2_XOPMASK_VOP
);
2031 if (error
!= ENOENT
)
2036 if ((lhcbase
^ tlhc
) & ~HAMMER2_DIRHASH_LOMASK
) {
2043 * Everything is setup, do the rename.
2045 * We have to synchronize ip->meta to the underlying operation.
2047 * NOTE: To avoid deadlocks we cannot lock (ip) while we are
2048 * unlinking elements from their directories. Locking
2049 * the nlinks field does not lock the whole inode.
2051 /* hammer2_inode_lock(ip, 0); */
2053 hammer2_xop_nrename_t
*xop4
;
2055 xop4
= hammer2_xop_alloc(fdip
, HAMMER2_XOP_MODIFYING
);
2057 xop4
->ip_key
= ip
->meta
.name_key
;
2058 hammer2_xop_setip2(&xop4
->head
, ip
);
2059 hammer2_xop_setip3(&xop4
->head
, tdip
);
2060 hammer2_xop_setname(&xop4
->head
, fname
, fname_len
);
2061 hammer2_xop_setname2(&xop4
->head
, tname
, tname_len
);
2062 hammer2_xop_start(&xop4
->head
, hammer2_xop_nrename
);
2064 error
= hammer2_xop_collect(&xop4
->head
, 0);
2065 hammer2_xop_retire(&xop4
->head
, HAMMER2_XOPMASK_VOP
);
2067 if (error
== ENOENT
)
2070 (ip
->meta
.name_key
& HAMMER2_DIRHASH_VISIBLE
)) {
2071 hammer2_inode_modify(ip
);
2072 ip
->meta
.name_len
= tname_len
;
2073 ip
->meta
.name_key
= tlhc
;
2082 * Update directory mtimes to represent the something changed.
2084 if (update_fdip
|| update_tdip
) {
2087 hammer2_update_time(&mtime
);
2089 hammer2_inode_modify(fdip
);
2090 fdip
->meta
.mtime
= mtime
;
2093 hammer2_inode_modify(tdip
);
2094 tdip
->meta
.mtime
= mtime
;
2097 hammer2_inode_unlock(ip
);
2098 hammer2_inode_unlock(tdip
);
2099 hammer2_inode_unlock(fdip
);
2100 hammer2_inode_drop(ip
);
2101 hammer2_inode_run_sideq(fdip
->pmp
);
2103 hammer2_trans_done(tdip
->pmp
);
2106 * Issue the namecache update after unlocking all the internal
2107 * hammer structures, otherwise we might deadlock.
2109 if (tnch_error
== 0) {
2110 cache_unlink(ap
->a_tnch
);
2111 cache_setunresolved(ap
->a_tnch
);
2114 cache_rename(ap
->a_fnch
, ap
->a_tnch
);
2121 * hammer2_vop_ioctl { vp, command, data, fflag, cred }
2125 hammer2_vop_ioctl(struct vop_ioctl_args
*ap
)
2127 hammer2_inode_t
*ip
;
2131 ip
= VTOI(ap
->a_vp
);
2133 error
= hammer2_ioctl(ip
, ap
->a_command
, (void *)ap
->a_data
,
2134 ap
->a_fflag
, ap
->a_cred
);
2141 hammer2_vop_mountctl(struct vop_mountctl_args
*ap
)
2149 case (MOUNTCTL_SET_EXPORT
):
2150 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
2153 if (ap
->a_ctllen
!= sizeof(struct export_args
))
2156 rc
= vfs_export(mp
, &pmp
->export
,
2157 (const struct export_args
*)ap
->a_ctl
);
2160 rc
= vop_stdmountctl(ap
);
2170 static void filt_hammer2detach(struct knote
*kn
);
2171 static int filt_hammer2read(struct knote
*kn
, long hint
);
2172 static int filt_hammer2write(struct knote
*kn
, long hint
);
2173 static int filt_hammer2vnode(struct knote
*kn
, long hint
);
2175 static struct filterops hammer2read_filtops
=
2176 { FILTEROP_ISFD
| FILTEROP_MPSAFE
,
2177 NULL
, filt_hammer2detach
, filt_hammer2read
};
2178 static struct filterops hammer2write_filtops
=
2179 { FILTEROP_ISFD
| FILTEROP_MPSAFE
,
2180 NULL
, filt_hammer2detach
, filt_hammer2write
};
2181 static struct filterops hammer2vnode_filtops
=
2182 { FILTEROP_ISFD
| FILTEROP_MPSAFE
,
2183 NULL
, filt_hammer2detach
, filt_hammer2vnode
};
2187 hammer2_vop_kqfilter(struct vop_kqfilter_args
*ap
)
2189 struct vnode
*vp
= ap
->a_vp
;
2190 struct knote
*kn
= ap
->a_kn
;
2192 switch (kn
->kn_filter
) {
2194 kn
->kn_fop
= &hammer2read_filtops
;
2197 kn
->kn_fop
= &hammer2write_filtops
;
2200 kn
->kn_fop
= &hammer2vnode_filtops
;
2203 return (EOPNOTSUPP
);
2206 kn
->kn_hook
= (caddr_t
)vp
;
2208 knote_insert(&vp
->v_pollinfo
.vpi_kqinfo
.ki_note
, kn
);
2214 filt_hammer2detach(struct knote
*kn
)
2216 struct vnode
*vp
= (void *)kn
->kn_hook
;
2218 knote_remove(&vp
->v_pollinfo
.vpi_kqinfo
.ki_note
, kn
);
2222 filt_hammer2read(struct knote
*kn
, long hint
)
2224 struct vnode
*vp
= (void *)kn
->kn_hook
;
2225 hammer2_inode_t
*ip
= VTOI(vp
);
2228 if (hint
== NOTE_REVOKE
) {
2229 kn
->kn_flags
|= (EV_EOF
| EV_NODATA
| EV_ONESHOT
);
2232 off
= ip
->meta
.size
- kn
->kn_fp
->f_offset
;
2233 kn
->kn_data
= (off
< INTPTR_MAX
) ? off
: INTPTR_MAX
;
2234 if (kn
->kn_sfflags
& NOTE_OLDAPI
)
2236 return (kn
->kn_data
!= 0);
2241 filt_hammer2write(struct knote
*kn
, long hint
)
2243 if (hint
== NOTE_REVOKE
)
2244 kn
->kn_flags
|= (EV_EOF
| EV_NODATA
| EV_ONESHOT
);
2250 filt_hammer2vnode(struct knote
*kn
, long hint
)
2252 if (kn
->kn_sfflags
& hint
)
2253 kn
->kn_fflags
|= hint
;
2254 if (hint
== NOTE_REVOKE
) {
2255 kn
->kn_flags
|= (EV_EOF
| EV_NODATA
);
2258 return (kn
->kn_fflags
!= 0);
2266 hammer2_vop_markatime(struct vop_markatime_args
*ap
)
2268 hammer2_inode_t
*ip
;
2281 hammer2_vop_fifokqfilter(struct vop_kqfilter_args
*ap
)
2285 error
= VOCALL(&fifo_vnode_vops
, &ap
->a_head
);
2287 error
= hammer2_vop_kqfilter(ap
);
2294 struct vop_ops hammer2_vnode_vops
= {
2295 .vop_default
= vop_defaultop
,
2296 .vop_fsync
= hammer2_vop_fsync
,
2297 .vop_getpages
= vop_stdgetpages
,
2298 .vop_putpages
= vop_stdputpages
,
2299 .vop_access
= hammer2_vop_access
,
2300 .vop_advlock
= hammer2_vop_advlock
,
2301 .vop_close
= hammer2_vop_close
,
2302 .vop_nlink
= hammer2_vop_nlink
,
2303 .vop_ncreate
= hammer2_vop_ncreate
,
2304 .vop_nsymlink
= hammer2_vop_nsymlink
,
2305 .vop_nremove
= hammer2_vop_nremove
,
2306 .vop_nrmdir
= hammer2_vop_nrmdir
,
2307 .vop_nrename
= hammer2_vop_nrename
,
2308 .vop_getattr
= hammer2_vop_getattr
,
2309 .vop_setattr
= hammer2_vop_setattr
,
2310 .vop_readdir
= hammer2_vop_readdir
,
2311 .vop_readlink
= hammer2_vop_readlink
,
2312 .vop_getpages
= vop_stdgetpages
,
2313 .vop_putpages
= vop_stdputpages
,
2314 .vop_read
= hammer2_vop_read
,
2315 .vop_write
= hammer2_vop_write
,
2316 .vop_open
= hammer2_vop_open
,
2317 .vop_inactive
= hammer2_vop_inactive
,
2318 .vop_reclaim
= hammer2_vop_reclaim
,
2319 .vop_nresolve
= hammer2_vop_nresolve
,
2320 .vop_nlookupdotdot
= hammer2_vop_nlookupdotdot
,
2321 .vop_nmkdir
= hammer2_vop_nmkdir
,
2322 .vop_nmknod
= hammer2_vop_nmknod
,
2323 .vop_ioctl
= hammer2_vop_ioctl
,
2324 .vop_mountctl
= hammer2_vop_mountctl
,
2325 .vop_bmap
= hammer2_vop_bmap
,
2326 .vop_strategy
= hammer2_vop_strategy
,
2327 .vop_kqfilter
= hammer2_vop_kqfilter
2330 struct vop_ops hammer2_spec_vops
= {
2331 .vop_default
= vop_defaultop
,
2332 .vop_fsync
= hammer2_vop_fsync
,
2333 .vop_read
= vop_stdnoread
,
2334 .vop_write
= vop_stdnowrite
,
2335 .vop_access
= hammer2_vop_access
,
2336 .vop_close
= hammer2_vop_close
,
2337 .vop_markatime
= hammer2_vop_markatime
,
2338 .vop_getattr
= hammer2_vop_getattr
,
2339 .vop_inactive
= hammer2_vop_inactive
,
2340 .vop_reclaim
= hammer2_vop_reclaim
,
2341 .vop_setattr
= hammer2_vop_setattr
2344 struct vop_ops hammer2_fifo_vops
= {
2345 .vop_default
= fifo_vnoperate
,
2346 .vop_fsync
= hammer2_vop_fsync
,
2348 .vop_read
= hammer2_vop_fiforead
,
2349 .vop_write
= hammer2_vop_fifowrite
,
2351 .vop_access
= hammer2_vop_access
,
2353 .vop_close
= hammer2_vop_fifoclose
,
2355 .vop_markatime
= hammer2_vop_markatime
,
2356 .vop_getattr
= hammer2_vop_getattr
,
2357 .vop_inactive
= hammer2_vop_inactive
,
2358 .vop_reclaim
= hammer2_vop_reclaim
,
2359 .vop_setattr
= hammer2_vop_setattr
,
2360 .vop_kqfilter
= hammer2_vop_fifokqfilter