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
)
101 * Check for deleted inodes and recycle immediately on the last
102 * release. Be sure to destroy any left-over buffer cache buffers
103 * so we do not waste time trying to flush them.
105 * Note that deleting the file block chains under the inode chain
106 * would just be a waste of energy, so don't do it.
108 * WARNING: nvtruncbuf() can only be safely called without the inode
109 * lock held due to the way our write thread works.
111 if (ip
->flags
& HAMMER2_INODE_ISUNLINKED
) {
116 * Detect updates to the embedded data which may be
117 * synchronized by the strategy code. Simply mark the
118 * inode modified so it gets picked up by our normal flush.
120 nblksize
= hammer2_calc_logical(ip
, 0, &lbase
, NULL
);
121 nvtruncbuf(vp
, 0, nblksize
, 0, 0);
128 * Reclaim a vnode so that it can be reused; after the inode is
129 * disassociated, the filesystem must manage it alone.
133 hammer2_vop_reclaim(struct vop_reclaim_args
*ap
)
147 * The final close of a deleted file or directory marks it for
148 * destruction. The DELETED flag allows the flusher to shortcut
149 * any modified blocks still unflushed (that is, just ignore them).
151 * HAMMER2 usually does not try to optimize the freemap by returning
152 * deleted blocks to it as it does not usually know how many snapshots
153 * might be referencing portions of the file/dir.
159 * NOTE! We do not attempt to flush chains here, flushing is
160 * really fragile and could also deadlock.
165 * A modified inode may require chain synchronization. This
166 * synchronization is usually handled by VOP_SNYC / VOP_FSYNC
167 * when vfsync() is called. However, that requires a vnode.
169 * When the vnode is disassociated we must keep track of any modified
170 * inode via the sideq so that it is properly flushed. We cannot
171 * safely synchronize the inode from inside the reclaim due to
172 * potentially deep locks held as-of when the reclaim occurs.
173 * Interactions and potential deadlocks abound.
175 if ((ip
->flags
& (HAMMER2_INODE_ISUNLINKED
|
176 HAMMER2_INODE_MODIFIED
|
177 HAMMER2_INODE_RESIZED
)) &&
178 (ip
->flags
& HAMMER2_INODE_ISDELETED
) == 0) {
179 hammer2_inode_sideq_t
*ipul
;
181 ipul
= kmalloc(sizeof(*ipul
), pmp
->minode
, M_WAITOK
| M_ZERO
);
184 hammer2_spin_ex(&pmp
->list_spin
);
185 if ((ip
->flags
& HAMMER2_INODE_ONSIDEQ
) == 0) {
187 atomic_set_int(&ip
->flags
, HAMMER2_INODE_ONSIDEQ
);
188 TAILQ_INSERT_TAIL(&pmp
->sideq
, ipul
, entry
);
190 hammer2_spin_unex(&pmp
->list_spin
);
192 hammer2_spin_unex(&pmp
->list_spin
);
193 kfree(ipul
, pmp
->minode
);
194 hammer2_inode_drop(ip
); /* vp ref */
196 /* retain ref from vp for ipul */
198 hammer2_inode_drop(ip
); /* vp ref */
202 * XXX handle background sync when ip dirty, kernel will no longer
203 * notify us regarding this inode because there is no longer a
204 * vnode attached to it.
212 hammer2_vop_fsync(struct vop_fsync_args
*ap
)
221 /* XXX can't do this yet */
222 hammer2_trans_init(ip
->pmp
, HAMMER2_TRANS_ISFLUSH
);
223 vfsync(vp
, ap
->a_waitfor
, 1, NULL
, NULL
);
225 hammer2_trans_init(ip
->pmp
, 0);
226 vfsync(vp
, ap
->a_waitfor
, 1, NULL
, NULL
);
229 * Calling chain_flush here creates a lot of duplicative
230 * COW operations due to non-optimal vnode ordering.
232 * Only do it for an actual fsync() syscall. The other forms
233 * which call this function will eventually call chain_flush
234 * on the volume root as a catch-all, which is far more optimal.
236 hammer2_inode_lock(ip
, 0);
237 if (ip
->flags
& HAMMER2_INODE_MODIFIED
)
238 hammer2_inode_chain_sync(ip
);
239 hammer2_inode_unlock(ip
);
240 hammer2_trans_done(ip
->pmp
);
247 hammer2_vop_access(struct vop_access_args
*ap
)
249 hammer2_inode_t
*ip
= VTOI(ap
->a_vp
);
254 hammer2_inode_lock(ip
, HAMMER2_RESOLVE_SHARED
);
255 uid
= hammer2_to_unix_xid(&ip
->meta
.uid
);
256 gid
= hammer2_to_unix_xid(&ip
->meta
.gid
);
257 error
= vop_helper_access(ap
, uid
, gid
, ip
->meta
.mode
, ip
->meta
.uflags
);
258 hammer2_inode_unlock(ip
);
265 hammer2_vop_getattr(struct vop_getattr_args
*ap
)
271 hammer2_chain_t
*chain
;
280 hammer2_inode_lock(ip
, HAMMER2_RESOLVE_SHARED
);
282 vap
->va_fsid
= pmp
->mp
->mnt_stat
.f_fsid
.val
[0];
283 vap
->va_fileid
= ip
->meta
.inum
;
284 vap
->va_mode
= ip
->meta
.mode
;
285 vap
->va_nlink
= ip
->meta
.nlinks
;
286 vap
->va_uid
= hammer2_to_unix_xid(&ip
->meta
.uid
);
287 vap
->va_gid
= hammer2_to_unix_xid(&ip
->meta
.gid
);
290 vap
->va_size
= ip
->meta
.size
; /* protected by shared lock */
291 vap
->va_blocksize
= HAMMER2_PBUFSIZE
;
292 vap
->va_flags
= ip
->meta
.uflags
;
293 hammer2_time_to_timespec(ip
->meta
.ctime
, &vap
->va_ctime
);
294 hammer2_time_to_timespec(ip
->meta
.mtime
, &vap
->va_mtime
);
295 hammer2_time_to_timespec(ip
->meta
.mtime
, &vap
->va_atime
);
298 if (ip
->meta
.type
== HAMMER2_OBJTYPE_DIRECTORY
) {
300 * Can't really calculate directory use sans the files under
301 * it, just assume one block for now.
303 vap
->va_bytes
+= HAMMER2_INODE_BYTES
;
305 for (i
= 0; i
< ip
->cluster
.nchains
; ++i
) {
306 if ((chain
= ip
->cluster
.array
[i
].chain
) != NULL
) {
308 chain
->bref
.embed
.stats
.data_count
) {
310 chain
->bref
.embed
.stats
.data_count
;
315 vap
->va_type
= hammer2_get_vtype(ip
->meta
.type
);
317 vap
->va_uid_uuid
= ip
->meta
.uid
;
318 vap
->va_gid_uuid
= ip
->meta
.gid
;
319 vap
->va_vaflags
= VA_UID_UUID_VALID
| VA_GID_UUID_VALID
|
322 hammer2_inode_unlock(ip
);
329 hammer2_vop_setattr(struct vop_setattr_args
*ap
)
340 hammer2_update_time(&ctime
);
346 if (hammer2_vfs_enospace(ip
, 0, ap
->a_cred
) > 1)
349 hammer2_pfs_memory_wait(ip
->pmp
);
350 hammer2_trans_init(ip
->pmp
, 0);
351 hammer2_inode_lock(ip
, 0);
354 if (vap
->va_flags
!= VNOVAL
) {
357 flags
= ip
->meta
.uflags
;
358 error
= vop_helper_setattr_flags(&flags
, vap
->va_flags
,
359 hammer2_to_unix_xid(&ip
->meta
.uid
),
362 if (ip
->meta
.uflags
!= flags
) {
363 hammer2_inode_modify(ip
);
364 ip
->meta
.uflags
= flags
;
365 ip
->meta
.ctime
= ctime
;
366 kflags
|= NOTE_ATTRIB
;
368 if (ip
->meta
.uflags
& (IMMUTABLE
| APPEND
)) {
375 if (ip
->meta
.uflags
& (IMMUTABLE
| APPEND
)) {
379 if (vap
->va_uid
!= (uid_t
)VNOVAL
|| vap
->va_gid
!= (gid_t
)VNOVAL
) {
380 mode_t cur_mode
= ip
->meta
.mode
;
381 uid_t cur_uid
= hammer2_to_unix_xid(&ip
->meta
.uid
);
382 gid_t cur_gid
= hammer2_to_unix_xid(&ip
->meta
.gid
);
386 error
= vop_helper_chown(ap
->a_vp
, vap
->va_uid
, vap
->va_gid
,
388 &cur_uid
, &cur_gid
, &cur_mode
);
390 hammer2_guid_to_uuid(&uuid_uid
, cur_uid
);
391 hammer2_guid_to_uuid(&uuid_gid
, cur_gid
);
392 if (bcmp(&uuid_uid
, &ip
->meta
.uid
, sizeof(uuid_uid
)) ||
393 bcmp(&uuid_gid
, &ip
->meta
.gid
, sizeof(uuid_gid
)) ||
394 ip
->meta
.mode
!= cur_mode
396 hammer2_inode_modify(ip
);
397 ip
->meta
.uid
= uuid_uid
;
398 ip
->meta
.gid
= uuid_gid
;
399 ip
->meta
.mode
= cur_mode
;
400 ip
->meta
.ctime
= ctime
;
402 kflags
|= NOTE_ATTRIB
;
409 if (vap
->va_size
!= VNOVAL
&& ip
->meta
.size
!= vap
->va_size
) {
412 if (vap
->va_size
== ip
->meta
.size
)
414 if (vap
->va_size
< ip
->meta
.size
) {
415 hammer2_mtx_ex(&ip
->truncate_lock
);
416 hammer2_truncate_file(ip
, vap
->va_size
);
417 hammer2_mtx_unlock(&ip
->truncate_lock
);
418 kflags
|= NOTE_WRITE
;
420 hammer2_extend_file(ip
, vap
->va_size
);
421 kflags
|= NOTE_WRITE
| NOTE_EXTEND
;
423 hammer2_inode_modify(ip
);
424 ip
->meta
.mtime
= ctime
;
432 /* atime not supported */
433 if (vap
->va_atime
.tv_sec
!= VNOVAL
) {
434 hammer2_inode_modify(ip
);
435 ip
->meta
.atime
= hammer2_timespec_to_time(&vap
->va_atime
);
436 kflags
|= NOTE_ATTRIB
;
439 if (vap
->va_mode
!= (mode_t
)VNOVAL
) {
440 mode_t cur_mode
= ip
->meta
.mode
;
441 uid_t cur_uid
= hammer2_to_unix_xid(&ip
->meta
.uid
);
442 gid_t cur_gid
= hammer2_to_unix_xid(&ip
->meta
.gid
);
444 error
= vop_helper_chmod(ap
->a_vp
, vap
->va_mode
, ap
->a_cred
,
445 cur_uid
, cur_gid
, &cur_mode
);
446 if (error
== 0 && ip
->meta
.mode
!= cur_mode
) {
447 hammer2_inode_modify(ip
);
448 ip
->meta
.mode
= cur_mode
;
449 ip
->meta
.ctime
= ctime
;
450 kflags
|= NOTE_ATTRIB
;
454 if (vap
->va_mtime
.tv_sec
!= VNOVAL
) {
455 hammer2_inode_modify(ip
);
456 ip
->meta
.mtime
= hammer2_timespec_to_time(&vap
->va_mtime
);
457 kflags
|= NOTE_ATTRIB
;
462 * If a truncation occurred we must call inode_fsync() now in order
463 * to trim the related data chains, otherwise a later expansion can
466 * If an extend occured that changed the DIRECTDATA state, we must
467 * call inode_fsync now in order to prepare the inode's indirect
470 if (ip
->flags
& HAMMER2_INODE_RESIZED
)
471 hammer2_inode_chain_sync(ip
);
476 hammer2_inode_unlock(ip
);
477 hammer2_trans_done(ip
->pmp
);
478 hammer2_knote(ip
->vp
, kflags
);
485 hammer2_vop_readdir(struct vop_readdir_args
*ap
)
487 hammer2_xop_readdir_t
*xop
;
488 hammer2_blockref_t bref
;
503 saveoff
= uio
->uio_offset
;
508 * Setup cookies directory entry cookies if requested
510 if (ap
->a_ncookies
) {
511 ncookies
= uio
->uio_resid
/ 16 + 1;
514 cookies
= kmalloc(ncookies
* sizeof(off_t
), M_TEMP
, M_WAITOK
);
521 hammer2_inode_lock(ip
, HAMMER2_RESOLVE_SHARED
);
524 * Handle artificial entries. To ensure that only positive 64 bit
525 * quantities are returned to userland we always strip off bit 63.
526 * The hash code is designed such that codes 0x0000-0x7FFF are not
527 * used, allowing us to use these codes for articial entries.
529 * Entry 0 is used for '.' and entry 1 is used for '..'. Do not
530 * allow '..' to cross the mount point into (e.g.) the super-root.
533 inum
= ip
->meta
.inum
& HAMMER2_DIRHASH_USERMSK
;
534 r
= vop_write_dirent(&error
, uio
, inum
, DT_DIR
, 1, ".");
538 cookies
[cookie_index
] = saveoff
;
541 if (cookie_index
== ncookies
)
547 * Be careful with lockorder when accessing ".."
549 * (ip is the current dir. xip is the parent dir).
551 inum
= ip
->meta
.inum
& HAMMER2_DIRHASH_USERMSK
;
552 if (ip
!= ip
->pmp
->iroot
)
553 inum
= ip
->meta
.iparent
& HAMMER2_DIRHASH_USERMSK
;
554 r
= vop_write_dirent(&error
, uio
, inum
, DT_DIR
, 2, "..");
558 cookies
[cookie_index
] = saveoff
;
561 if (cookie_index
== ncookies
)
565 lkey
= saveoff
| HAMMER2_DIRHASH_VISIBLE
;
566 if (hammer2_debug
& 0x0020)
567 kprintf("readdir: lkey %016jx\n", lkey
);
572 * Use XOP for cluster scan.
574 * parent is the inode cluster, already locked for us. Don't
575 * double lock shared locks as this will screw up upgrades.
577 xop
= hammer2_xop_alloc(ip
, 0);
579 hammer2_xop_start(&xop
->head
, hammer2_xop_readdir
);
582 const hammer2_inode_data_t
*ripdata
;
586 error
= hammer2_xop_collect(&xop
->head
, 0);
587 error
= hammer2_error_to_errno(error
);
591 if (cookie_index
== ncookies
)
593 if (hammer2_debug
& 0x0020)
594 kprintf("cluster chain %p %p\n",
595 xop
->head
.cluster
.focus
,
596 (xop
->head
.cluster
.focus
?
597 xop
->head
.cluster
.focus
->data
: (void *)-1));
598 hammer2_cluster_bref(&xop
->head
.cluster
, &bref
);
600 if (bref
.type
== HAMMER2_BREF_TYPE_INODE
) {
602 &hammer2_cluster_rdata(&xop
->head
.cluster
)->ipdata
;
603 dtype
= hammer2_get_dtype(ripdata
->meta
.type
);
604 saveoff
= bref
.key
& HAMMER2_DIRHASH_USERMSK
;
605 r
= vop_write_dirent(&error
, uio
,
607 HAMMER2_DIRHASH_USERMSK
,
609 ripdata
->meta
.name_len
,
614 cookies
[cookie_index
] = saveoff
;
616 } else if (bref
.type
== HAMMER2_BREF_TYPE_DIRENT
) {
617 dtype
= hammer2_get_dtype(bref
.embed
.dirent
.type
);
618 saveoff
= bref
.key
& HAMMER2_DIRHASH_USERMSK
;
619 if (bref
.embed
.dirent
.namlen
<=
620 sizeof(bref
.check
.buf
)) {
621 dname
= bref
.check
.buf
;
624 hammer2_cluster_rdata(&xop
->head
.cluster
)->buf
;
626 r
= vop_write_dirent(&error
, uio
,
627 bref
.embed
.dirent
.inum
,
629 bref
.embed
.dirent
.namlen
,
634 cookies
[cookie_index
] = saveoff
;
637 /* XXX chain error */
638 kprintf("bad chain type readdir %d\n", bref
.type
);
641 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
642 if (error
== ENOENT
) {
645 saveoff
= (hammer2_key_t
)-1;
647 saveoff
= bref
.key
& HAMMER2_DIRHASH_USERMSK
;
650 hammer2_inode_unlock(ip
);
652 *ap
->a_eofflag
= eofflag
;
653 if (hammer2_debug
& 0x0020)
654 kprintf("readdir: done at %016jx\n", saveoff
);
655 uio
->uio_offset
= saveoff
& ~HAMMER2_DIRHASH_VISIBLE
;
656 if (error
&& cookie_index
== 0) {
658 kfree(cookies
, M_TEMP
);
660 *ap
->a_cookies
= NULL
;
664 *ap
->a_ncookies
= cookie_index
;
665 *ap
->a_cookies
= cookies
;
672 * hammer2_vop_readlink { vp, uio, cred }
676 hammer2_vop_readlink(struct vop_readlink_args
*ap
)
683 if (vp
->v_type
!= VLNK
)
687 error
= hammer2_read_file(ip
, ap
->a_uio
, 0);
693 hammer2_vop_read(struct vop_read_args
*ap
)
703 * Read operations supported on this vnode?
706 if (vp
->v_type
!= VREG
)
716 seqcount
= ap
->a_ioflag
>> 16;
717 bigread
= (uio
->uio_resid
> 100 * 1024 * 1024);
719 error
= hammer2_read_file(ip
, uio
, seqcount
);
725 hammer2_vop_write(struct vop_write_args
*ap
)
736 * Read operations supported on this vnode?
739 if (vp
->v_type
!= VREG
)
746 ioflag
= ap
->a_ioflag
;
751 switch (hammer2_vfs_enospace(ip
, uio
->uio_resid
, ap
->a_cred
)) {
755 ioflag
|= IO_DIRECT
; /* semi-synchronous */
761 seqcount
= ioflag
>> 16;
764 * Check resource limit
766 if (uio
->uio_resid
> 0 && (td
= uio
->uio_td
) != NULL
&& td
->td_proc
&&
767 uio
->uio_offset
+ uio
->uio_resid
>
768 td
->td_proc
->p_rlimit
[RLIMIT_FSIZE
].rlim_cur
) {
769 lwpsignal(td
->td_proc
, td
->td_lwp
, SIGXFSZ
);
774 * The transaction interlocks against flush initiations
775 * (note: but will run concurrently with the actual flush).
777 * To avoid deadlocking against the VM system, we must flag any
778 * transaction related to the buffer cache or other direct
779 * VM page manipulation.
781 if (uio
->uio_segflg
== UIO_NOCOPY
)
782 hammer2_trans_init(ip
->pmp
, HAMMER2_TRANS_BUFCACHE
);
784 hammer2_trans_init(ip
->pmp
, 0);
785 error
= hammer2_write_file(ip
, uio
, ioflag
, seqcount
);
786 hammer2_trans_done(ip
->pmp
);
792 * Perform read operations on a file or symlink given an UNLOCKED
795 * The passed ip is not locked.
799 hammer2_read_file(hammer2_inode_t
*ip
, struct uio
*uio
, int seqcount
)
810 * WARNING! Assumes that the kernel interlocks size changes at the
813 hammer2_mtx_sh(&ip
->lock
);
814 hammer2_mtx_sh(&ip
->truncate_lock
);
815 size
= ip
->meta
.size
;
816 hammer2_mtx_unlock(&ip
->lock
);
818 while (uio
->uio_resid
> 0 && uio
->uio_offset
< size
) {
825 lblksize
= hammer2_calc_logical(ip
, uio
->uio_offset
,
830 error
= cluster_readx(ip
->vp
, leof
, lbase
, lblksize
,
831 B_NOTMETA
| B_KVABIO
,
836 if (uio
->uio_segflg
== UIO_NOCOPY
) {
837 bp
= getblk(ip
->vp
, lbase
, lblksize
,
838 GETBLK_BHEAVY
| GETBLK_KVABIO
, 0);
839 if (bp
->b_flags
& B_CACHE
) {
842 if (bp
->b_xio
.xio_npages
!= 16)
843 kprintf("NPAGES BAD\n");
844 for (i
= 0; i
< bp
->b_xio
.xio_npages
; ++i
) {
846 m
= bp
->b_xio
.xio_pages
[i
];
847 if (m
== NULL
|| m
->valid
== 0) {
848 kprintf("bp %016jx %016jx pg %d inv",
851 kprintf("m->object %p/%p", m
->object
, ip
->vp
->v_object
);
857 kprintf("b_flags %08x, b_error %d\n", bp
->b_flags
, bp
->b_error
);
861 error
= bread_kvabio(ip
->vp
, lbase
, lblksize
, &bp
);
868 loff
= (int)(uio
->uio_offset
- lbase
);
870 if (n
> uio
->uio_resid
)
872 if (n
> size
- uio
->uio_offset
)
873 n
= (int)(size
- uio
->uio_offset
);
874 bp
->b_flags
|= B_AGE
;
875 uiomovebp(bp
, (char *)bp
->b_data
+ loff
, n
, uio
);
878 hammer2_mtx_unlock(&ip
->truncate_lock
);
884 * Write to the file represented by the inode via the logical buffer cache.
885 * The inode may represent a regular file or a symlink.
887 * The inode must not be locked.
891 hammer2_write_file(hammer2_inode_t
*ip
, struct uio
*uio
,
892 int ioflag
, int seqcount
)
894 hammer2_key_t old_eof
;
895 hammer2_key_t new_eof
;
904 * WARNING! Assumes that the kernel interlocks size changes at the
907 hammer2_mtx_ex(&ip
->lock
);
908 hammer2_mtx_sh(&ip
->truncate_lock
);
909 if (ioflag
& IO_APPEND
)
910 uio
->uio_offset
= ip
->meta
.size
;
911 old_eof
= ip
->meta
.size
;
914 * Extend the file if necessary. If the write fails at some point
915 * we will truncate it back down to cover as much as we were able
918 * Doing this now makes it easier to calculate buffer sizes in
925 if (uio
->uio_offset
+ uio
->uio_resid
> old_eof
) {
926 new_eof
= uio
->uio_offset
+ uio
->uio_resid
;
928 hammer2_extend_file(ip
, new_eof
);
929 kflags
|= NOTE_EXTEND
;
933 hammer2_mtx_unlock(&ip
->lock
);
938 while (uio
->uio_resid
> 0) {
947 * Don't allow the buffer build to blow out the buffer
950 if ((ioflag
& IO_RECURSE
) == 0)
951 bwillwrite(HAMMER2_PBUFSIZE
);
954 * This nominally tells us how much we can cluster and
955 * what the logical buffer size needs to be. Currently
956 * we don't try to cluster the write and just handle one
959 lblksize
= hammer2_calc_logical(ip
, uio
->uio_offset
,
961 loff
= (int)(uio
->uio_offset
- lbase
);
963 KKASSERT(lblksize
<= 65536);
966 * Calculate bytes to copy this transfer and whether the
967 * copy completely covers the buffer or not.
971 if (n
> uio
->uio_resid
) {
973 if (loff
== lbase
&& uio
->uio_offset
+ n
== new_eof
)
981 if (lbase
>= new_eof
)
987 if (uio
->uio_segflg
== UIO_NOCOPY
) {
989 * Issuing a write with the same data backing the
990 * buffer. Instantiate the buffer to collect the
991 * backing vm pages, then read-in any missing bits.
993 * This case is used by vop_stdputpages().
995 bp
= getblk(ip
->vp
, lbase
, lblksize
,
996 GETBLK_BHEAVY
| GETBLK_KVABIO
, 0);
997 if ((bp
->b_flags
& B_CACHE
) == 0) {
999 error
= bread_kvabio(ip
->vp
, lbase
,
1002 } else if (trivial
) {
1004 * Even though we are entirely overwriting the buffer
1005 * we may still have to zero it out to avoid a
1006 * mmap/write visibility issue.
1008 bp
= getblk(ip
->vp
, lbase
, lblksize
,
1009 GETBLK_BHEAVY
| GETBLK_KVABIO
, 0);
1010 if ((bp
->b_flags
& B_CACHE
) == 0)
1014 * Partial overwrite, read in any missing bits then
1015 * replace the portion being written.
1017 * (The strategy code will detect zero-fill physical
1018 * blocks for this case).
1020 error
= bread_kvabio(ip
->vp
, lbase
, lblksize
, &bp
);
1031 * Ok, copy the data in
1034 error
= uiomovebp(bp
, bp
->b_data
+ loff
, n
, uio
);
1035 kflags
|= NOTE_WRITE
;
1043 * WARNING: Pageout daemon will issue UIO_NOCOPY writes
1044 * with IO_SYNC or IO_ASYNC set. These writes
1045 * must be handled as the pageout daemon expects.
1047 * NOTE! H2 relies on cluster_write() here because it
1048 * cannot preallocate disk blocks at the logical
1049 * level due to not knowing what the compression
1050 * size will be at this time.
1052 * We must use cluster_write() here and we depend
1053 * on the write-behind feature to flush buffers
1054 * appropriately. If we let the buffer daemons do
1055 * it the block allocations will be all over the
1058 if (ioflag
& IO_SYNC
) {
1060 } else if ((ioflag
& IO_DIRECT
) && endofblk
) {
1062 } else if (ioflag
& IO_ASYNC
) {
1064 } else if (ip
->vp
->v_mount
->mnt_flag
& MNT_NOCLUSTERW
) {
1068 bp
->b_flags
|= B_CLUSTEROK
;
1069 cluster_write(bp
, new_eof
, lblksize
, seqcount
);
1071 bp
->b_flags
|= B_CLUSTEROK
;
1078 * Cleanup. If we extended the file EOF but failed to write through
1079 * the entire write is a failure and we have to back-up.
1081 if (error
&& new_eof
!= old_eof
) {
1082 hammer2_mtx_unlock(&ip
->truncate_lock
);
1083 hammer2_mtx_ex(&ip
->lock
);
1084 hammer2_mtx_ex(&ip
->truncate_lock
);
1085 hammer2_truncate_file(ip
, old_eof
);
1086 if (ip
->flags
& HAMMER2_INODE_MODIFIED
)
1087 hammer2_inode_chain_sync(ip
);
1088 hammer2_mtx_unlock(&ip
->lock
);
1089 } else if (modified
) {
1090 hammer2_mtx_ex(&ip
->lock
);
1091 hammer2_inode_modify(ip
);
1092 hammer2_update_time(&ip
->meta
.mtime
);
1093 if (ip
->flags
& HAMMER2_INODE_MODIFIED
)
1094 hammer2_inode_chain_sync(ip
);
1095 hammer2_mtx_unlock(&ip
->lock
);
1096 hammer2_knote(ip
->vp
, kflags
);
1098 hammer2_trans_assert_strategy(ip
->pmp
);
1099 hammer2_mtx_unlock(&ip
->truncate_lock
);
1105 * Truncate the size of a file. The inode must not be locked.
1107 * We must unconditionally set HAMMER2_INODE_RESIZED to properly
1108 * ensure that any on-media data beyond the new file EOF has been destroyed.
1110 * WARNING: nvtruncbuf() can only be safely called without the inode lock
1111 * held due to the way our write thread works. If the truncation
1112 * occurs in the middle of a buffer, nvtruncbuf() is responsible
1113 * for dirtying that buffer and zeroing out trailing bytes.
1115 * WARNING! Assumes that the kernel interlocks size changes at the
1118 * WARNING! Caller assumes responsibility for removing dead blocks
1119 * if INODE_RESIZED is set.
1123 hammer2_truncate_file(hammer2_inode_t
*ip
, hammer2_key_t nsize
)
1125 hammer2_key_t lbase
;
1128 hammer2_mtx_unlock(&ip
->lock
);
1130 nblksize
= hammer2_calc_logical(ip
, nsize
, &lbase
, NULL
);
1131 nvtruncbuf(ip
->vp
, nsize
,
1132 nblksize
, (int)nsize
& (nblksize
- 1),
1135 hammer2_mtx_ex(&ip
->lock
);
1136 KKASSERT((ip
->flags
& HAMMER2_INODE_RESIZED
) == 0);
1137 ip
->osize
= ip
->meta
.size
;
1138 ip
->meta
.size
= nsize
;
1139 atomic_set_int(&ip
->flags
, HAMMER2_INODE_RESIZED
);
1140 hammer2_inode_modify(ip
);
1144 * Extend the size of a file. The inode must not be locked.
1146 * Even though the file size is changing, we do not have to set the
1147 * INODE_RESIZED bit unless the file size crosses the EMBEDDED_BYTES
1148 * boundary. When this occurs a hammer2_inode_chain_sync() is required
1149 * to prepare the inode cluster's indirect block table, otherwise
1150 * async execution of the strategy code will implode on us.
1152 * WARNING! Assumes that the kernel interlocks size changes at the
1155 * WARNING! Caller assumes responsibility for transitioning out
1156 * of the inode DIRECTDATA mode if INODE_RESIZED is set.
1160 hammer2_extend_file(hammer2_inode_t
*ip
, hammer2_key_t nsize
)
1162 hammer2_key_t lbase
;
1163 hammer2_key_t osize
;
1167 KKASSERT((ip
->flags
& HAMMER2_INODE_RESIZED
) == 0);
1168 hammer2_inode_modify(ip
);
1169 osize
= ip
->meta
.size
;
1171 ip
->meta
.size
= nsize
;
1173 if (osize
<= HAMMER2_EMBEDDED_BYTES
&& nsize
> HAMMER2_EMBEDDED_BYTES
) {
1174 atomic_set_int(&ip
->flags
, HAMMER2_INODE_RESIZED
);
1175 hammer2_inode_chain_sync(ip
);
1178 hammer2_mtx_unlock(&ip
->lock
);
1180 oblksize
= hammer2_calc_logical(ip
, osize
, &lbase
, NULL
);
1181 nblksize
= hammer2_calc_logical(ip
, nsize
, &lbase
, NULL
);
1187 hammer2_mtx_ex(&ip
->lock
);
1192 hammer2_vop_nresolve(struct vop_nresolve_args
*ap
)
1194 hammer2_xop_nresolve_t
*xop
;
1195 hammer2_inode_t
*ip
;
1196 hammer2_inode_t
*dip
;
1197 struct namecache
*ncp
;
1201 dip
= VTOI(ap
->a_dvp
);
1202 xop
= hammer2_xop_alloc(dip
, 0);
1204 ncp
= ap
->a_nch
->ncp
;
1205 hammer2_xop_setname(&xop
->head
, ncp
->nc_name
, ncp
->nc_nlen
);
1208 * Note: In DragonFly the kernel handles '.' and '..'.
1210 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1211 hammer2_xop_start(&xop
->head
, hammer2_xop_nresolve
);
1213 error
= hammer2_xop_collect(&xop
->head
, 0);
1214 error
= hammer2_error_to_errno(error
);
1218 ip
= hammer2_inode_get(dip
->pmp
, dip
, &xop
->head
.cluster
, -1);
1220 hammer2_inode_unlock(dip
);
1223 * Acquire the related vnode
1225 * NOTE: For error processing, only ENOENT resolves the namecache
1226 * entry to NULL, otherwise we just return the error and
1227 * leave the namecache unresolved.
1229 * NOTE: multiple hammer2_inode structures can be aliased to the
1230 * same chain element, for example for hardlinks. This
1231 * use case does not 'reattach' inode associations that
1232 * might already exist, but always allocates a new one.
1234 * WARNING: inode structure is locked exclusively via inode_get
1235 * but chain was locked shared. inode_unlock()
1236 * will handle it properly.
1239 vp
= hammer2_igetv(ip
, &error
); /* error set to UNIX error */
1242 cache_setvp(ap
->a_nch
, vp
);
1243 } else if (error
== ENOENT
) {
1244 cache_setvp(ap
->a_nch
, NULL
);
1246 hammer2_inode_unlock(ip
);
1249 * The vp should not be released until after we've disposed
1250 * of our locks, because it might cause vop_inactive() to
1257 cache_setvp(ap
->a_nch
, NULL
);
1259 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1260 KASSERT(error
|| ap
->a_nch
->ncp
->nc_vp
!= NULL
,
1261 ("resolve error %d/%p ap %p\n",
1262 error
, ap
->a_nch
->ncp
->nc_vp
, ap
));
1269 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args
*ap
)
1271 hammer2_inode_t
*dip
;
1275 dip
= VTOI(ap
->a_dvp
);
1276 inum
= dip
->meta
.iparent
;
1280 error
= hammer2_vfs_vget(ap
->a_dvp
->v_mount
, NULL
,
1290 hammer2_vop_nmkdir(struct vop_nmkdir_args
*ap
)
1292 hammer2_inode_t
*dip
;
1293 hammer2_inode_t
*nip
;
1294 struct namecache
*ncp
;
1295 const uint8_t *name
;
1300 dip
= VTOI(ap
->a_dvp
);
1301 if (dip
->pmp
->ronly
)
1303 if (hammer2_vfs_enospace(dip
, 0, ap
->a_cred
) > 1)
1306 ncp
= ap
->a_nch
->ncp
;
1307 name
= ncp
->nc_name
;
1308 name_len
= ncp
->nc_nlen
;
1310 hammer2_pfs_memory_wait(dip
->pmp
);
1311 hammer2_trans_init(dip
->pmp
, 0);
1313 inum
= hammer2_trans_newinum(dip
->pmp
);
1316 * Create the actual inode as a hidden file in the iroot, then
1317 * create the directory entry. The creation of the actual inode
1318 * sets its nlinks to 1 which is the value we desire.
1320 nip
= hammer2_inode_create(dip
->pmp
->iroot
, dip
, ap
->a_vap
, ap
->a_cred
,
1325 error
= hammer2_error_to_errno(error
);
1327 error
= hammer2_dirent_create(dip
, name
, name_len
,
1328 nip
->meta
.inum
, nip
->meta
.type
);
1329 /* returns UNIX error code */
1333 hammer2_inode_unlink_finisher(nip
, 0);
1334 hammer2_inode_unlock(nip
);
1339 *ap
->a_vpp
= hammer2_igetv(nip
, &error
);
1340 hammer2_inode_unlock(nip
);
1344 * Update dip's mtime
1346 * We can use a shared inode lock and allow the meta.mtime update
1347 * SMP race. hammer2_inode_modify() is MPSAFE w/a shared lock.
1352 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1353 hammer2_update_time(&mtime
);
1354 hammer2_inode_modify(dip
);
1355 dip
->meta
.mtime
= mtime
;
1356 hammer2_inode_unlock(dip
);
1359 hammer2_trans_done(dip
->pmp
);
1362 cache_setunresolved(ap
->a_nch
);
1363 cache_setvp(ap
->a_nch
, *ap
->a_vpp
);
1364 hammer2_knote(ap
->a_dvp
, NOTE_WRITE
| NOTE_LINK
);
1371 hammer2_vop_open(struct vop_open_args
*ap
)
1373 return vop_stdopen(ap
);
1377 * hammer2_vop_advlock { vp, id, op, fl, flags }
1381 hammer2_vop_advlock(struct vop_advlock_args
*ap
)
1383 hammer2_inode_t
*ip
= VTOI(ap
->a_vp
);
1386 size
= ip
->meta
.size
;
1387 return (lf_advlock(ap
, &ip
->advlock
, size
));
1392 hammer2_vop_close(struct vop_close_args
*ap
)
1394 return vop_stdclose(ap
);
1398 * hammer2_vop_nlink { nch, dvp, vp, cred }
1400 * Create a hardlink from (vp) to {dvp, nch}.
1404 hammer2_vop_nlink(struct vop_nlink_args
*ap
)
1406 hammer2_inode_t
*tdip
; /* target directory to create link in */
1407 hammer2_inode_t
*ip
; /* inode we are hardlinking to */
1408 struct namecache
*ncp
;
1409 const uint8_t *name
;
1413 if (ap
->a_dvp
->v_mount
!= ap
->a_vp
->v_mount
)
1416 tdip
= VTOI(ap
->a_dvp
);
1417 if (tdip
->pmp
->ronly
)
1419 if (hammer2_vfs_enospace(tdip
, 0, ap
->a_cred
) > 1)
1422 ncp
= ap
->a_nch
->ncp
;
1423 name
= ncp
->nc_name
;
1424 name_len
= ncp
->nc_nlen
;
1427 * ip represents the file being hardlinked. The file could be a
1428 * normal file or a hardlink target if it has already been hardlinked.
1429 * (with the new semantics, it will almost always be a hardlink
1432 * Bump nlinks and potentially also create or move the hardlink
1433 * target in the parent directory common to (ip) and (tdip). The
1434 * consolidation code can modify ip->cluster. The returned cluster
1437 ip
= VTOI(ap
->a_vp
);
1438 KASSERT(ip
->pmp
, ("ip->pmp is NULL %p %p", ip
, ip
->pmp
));
1439 hammer2_pfs_memory_wait(ip
->pmp
);
1440 hammer2_trans_init(ip
->pmp
, 0);
1443 * Target should be an indexed inode or there's no way we will ever
1444 * be able to find it!
1446 KKASSERT((ip
->meta
.name_key
& HAMMER2_DIRHASH_VISIBLE
) == 0);
1451 * Can return NULL and error == EXDEV if the common parent
1452 * crosses a directory with the xlink flag set.
1454 hammer2_inode_lock(tdip
, 0);
1455 hammer2_inode_lock(ip
, 0);
1458 * Create the directory entry and bump nlinks.
1461 error
= hammer2_dirent_create(tdip
, name
, name_len
,
1462 ip
->meta
.inum
, ip
->meta
.type
);
1463 hammer2_inode_modify(ip
);
1468 * Update dip's mtime
1472 hammer2_update_time(&mtime
);
1473 hammer2_inode_modify(tdip
);
1474 tdip
->meta
.mtime
= mtime
;
1476 cache_setunresolved(ap
->a_nch
);
1477 cache_setvp(ap
->a_nch
, ap
->a_vp
);
1479 hammer2_inode_unlock(ip
);
1480 hammer2_inode_unlock(tdip
);
1482 hammer2_trans_done(ip
->pmp
);
1483 hammer2_knote(ap
->a_vp
, NOTE_LINK
);
1484 hammer2_knote(ap
->a_dvp
, NOTE_WRITE
);
1490 * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1492 * The operating system has already ensured that the directory entry
1493 * does not exist and done all appropriate namespace locking.
1497 hammer2_vop_ncreate(struct vop_ncreate_args
*ap
)
1499 hammer2_inode_t
*dip
;
1500 hammer2_inode_t
*nip
;
1501 struct namecache
*ncp
;
1502 const uint8_t *name
;
1507 dip
= VTOI(ap
->a_dvp
);
1508 if (dip
->pmp
->ronly
)
1510 if (hammer2_vfs_enospace(dip
, 0, ap
->a_cred
) > 1)
1513 ncp
= ap
->a_nch
->ncp
;
1514 name
= ncp
->nc_name
;
1515 name_len
= ncp
->nc_nlen
;
1516 hammer2_pfs_memory_wait(dip
->pmp
);
1517 hammer2_trans_init(dip
->pmp
, 0);
1519 inum
= hammer2_trans_newinum(dip
->pmp
);
1522 * Create the actual inode as a hidden file in the iroot, then
1523 * create the directory entry. The creation of the actual inode
1524 * sets its nlinks to 1 which is the value we desire.
1526 nip
= hammer2_inode_create(dip
->pmp
->iroot
, dip
, ap
->a_vap
, ap
->a_cred
,
1532 error
= hammer2_error_to_errno(error
);
1534 error
= hammer2_dirent_create(dip
, name
, name_len
,
1535 nip
->meta
.inum
, nip
->meta
.type
);
1539 hammer2_inode_unlink_finisher(nip
, 0);
1540 hammer2_inode_unlock(nip
);
1545 *ap
->a_vpp
= hammer2_igetv(nip
, &error
);
1546 hammer2_inode_unlock(nip
);
1550 * Update dip's mtime
1555 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1556 hammer2_update_time(&mtime
);
1557 hammer2_inode_modify(dip
);
1558 dip
->meta
.mtime
= mtime
;
1559 hammer2_inode_unlock(dip
);
1562 hammer2_trans_done(dip
->pmp
);
1565 cache_setunresolved(ap
->a_nch
);
1566 cache_setvp(ap
->a_nch
, *ap
->a_vpp
);
1567 hammer2_knote(ap
->a_dvp
, NOTE_WRITE
);
1573 * Make a device node (typically a fifo)
1577 hammer2_vop_nmknod(struct vop_nmknod_args
*ap
)
1579 hammer2_inode_t
*dip
;
1580 hammer2_inode_t
*nip
;
1581 struct namecache
*ncp
;
1582 const uint8_t *name
;
1587 dip
= VTOI(ap
->a_dvp
);
1588 if (dip
->pmp
->ronly
)
1590 if (hammer2_vfs_enospace(dip
, 0, ap
->a_cred
) > 1)
1593 ncp
= ap
->a_nch
->ncp
;
1594 name
= ncp
->nc_name
;
1595 name_len
= ncp
->nc_nlen
;
1596 hammer2_pfs_memory_wait(dip
->pmp
);
1597 hammer2_trans_init(dip
->pmp
, 0);
1600 * Create the device inode and then create the directory entry.
1602 inum
= hammer2_trans_newinum(dip
->pmp
);
1603 nip
= hammer2_inode_create(dip
->pmp
->iroot
, dip
, ap
->a_vap
, ap
->a_cred
,
1608 error
= hammer2_dirent_create(dip
, name
, name_len
,
1609 nip
->meta
.inum
, nip
->meta
.type
);
1613 hammer2_inode_unlink_finisher(nip
, 0);
1614 hammer2_inode_unlock(nip
);
1619 *ap
->a_vpp
= hammer2_igetv(nip
, &error
);
1620 hammer2_inode_unlock(nip
);
1624 * Update dip's mtime
1629 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1630 hammer2_update_time(&mtime
);
1631 hammer2_inode_modify(dip
);
1632 dip
->meta
.mtime
= mtime
;
1633 hammer2_inode_unlock(dip
);
1636 hammer2_trans_done(dip
->pmp
);
1639 cache_setunresolved(ap
->a_nch
);
1640 cache_setvp(ap
->a_nch
, *ap
->a_vpp
);
1641 hammer2_knote(ap
->a_dvp
, NOTE_WRITE
);
1647 * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1651 hammer2_vop_nsymlink(struct vop_nsymlink_args
*ap
)
1653 hammer2_inode_t
*dip
;
1654 hammer2_inode_t
*nip
;
1655 struct namecache
*ncp
;
1656 const uint8_t *name
;
1661 dip
= VTOI(ap
->a_dvp
);
1662 if (dip
->pmp
->ronly
)
1664 if (hammer2_vfs_enospace(dip
, 0, ap
->a_cred
) > 1)
1667 ncp
= ap
->a_nch
->ncp
;
1668 name
= ncp
->nc_name
;
1669 name_len
= ncp
->nc_nlen
;
1670 hammer2_pfs_memory_wait(dip
->pmp
);
1671 hammer2_trans_init(dip
->pmp
, 0);
1673 ap
->a_vap
->va_type
= VLNK
; /* enforce type */
1676 * Create the softlink as an inode and then create the directory
1679 inum
= hammer2_trans_newinum(dip
->pmp
);
1681 nip
= hammer2_inode_create(dip
->pmp
->iroot
, dip
, ap
->a_vap
, ap
->a_cred
,
1686 error
= hammer2_dirent_create(dip
, name
, name_len
,
1687 nip
->meta
.inum
, nip
->meta
.type
);
1691 hammer2_inode_unlink_finisher(nip
, 0);
1692 hammer2_inode_unlock(nip
);
1696 hammer2_trans_done(dip
->pmp
);
1699 *ap
->a_vpp
= hammer2_igetv(nip
, &error
);
1702 * Build the softlink (~like file data) and finalize the namecache.
1709 bytes
= strlen(ap
->a_target
);
1711 hammer2_inode_unlock(nip
);
1712 bzero(&auio
, sizeof(auio
));
1713 bzero(&aiov
, sizeof(aiov
));
1714 auio
.uio_iov
= &aiov
;
1715 auio
.uio_segflg
= UIO_SYSSPACE
;
1716 auio
.uio_rw
= UIO_WRITE
;
1717 auio
.uio_resid
= bytes
;
1718 auio
.uio_iovcnt
= 1;
1719 auio
.uio_td
= curthread
;
1720 aiov
.iov_base
= ap
->a_target
;
1721 aiov
.iov_len
= bytes
;
1722 error
= hammer2_write_file(nip
, &auio
, IO_APPEND
, 0);
1723 /* XXX handle error */
1726 hammer2_inode_unlock(nip
);
1730 * Update dip's mtime
1735 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1736 hammer2_update_time(&mtime
);
1737 hammer2_inode_modify(dip
);
1738 dip
->meta
.mtime
= mtime
;
1739 hammer2_inode_unlock(dip
);
1742 hammer2_trans_done(dip
->pmp
);
1745 * Finalize namecache
1748 cache_setunresolved(ap
->a_nch
);
1749 cache_setvp(ap
->a_nch
, *ap
->a_vpp
);
1750 hammer2_knote(ap
->a_dvp
, NOTE_WRITE
);
1756 * hammer2_vop_nremove { nch, dvp, cred }
1760 hammer2_vop_nremove(struct vop_nremove_args
*ap
)
1762 hammer2_xop_unlink_t
*xop
;
1763 hammer2_inode_t
*dip
;
1764 hammer2_inode_t
*ip
;
1765 struct namecache
*ncp
;
1769 dip
= VTOI(ap
->a_dvp
);
1770 if (dip
->pmp
->ronly
)
1773 /* allow removals, except user to also bulkfree */
1774 if (hammer2_vfs_enospace(dip
, 0, ap
->a_cred
) > 1)
1778 ncp
= ap
->a_nch
->ncp
;
1780 hammer2_pfs_memory_wait(dip
->pmp
);
1781 hammer2_trans_init(dip
->pmp
, 0);
1782 hammer2_inode_lock(dip
, 0);
1785 * The unlink XOP unlinks the path from the directory and
1786 * locates and returns the cluster associated with the real inode.
1787 * We have to handle nlinks here on the frontend.
1789 xop
= hammer2_xop_alloc(dip
, HAMMER2_XOP_MODIFYING
);
1790 hammer2_xop_setname(&xop
->head
, ncp
->nc_name
, ncp
->nc_nlen
);
1793 * The namecache entry is locked so nobody can use this namespace.
1794 * Calculate isopen to determine if this namespace has an open vp
1795 * associated with it and resolve the vp only if it does.
1797 * We try to avoid resolving the vnode if nobody has it open, but
1798 * note that the test is via this namespace only.
1800 isopen
= cache_isopen(ap
->a_nch
);
1802 xop
->dopermanent
= 0;
1803 hammer2_xop_start(&xop
->head
, hammer2_xop_unlink
);
1806 * Collect the real inode and adjust nlinks, destroy the real
1807 * inode if nlinks transitions to 0 and it was the real inode
1808 * (else it has already been removed).
1810 error
= hammer2_xop_collect(&xop
->head
, 0);
1811 error
= hammer2_error_to_errno(error
);
1812 hammer2_inode_unlock(dip
);
1815 ip
= hammer2_inode_get(dip
->pmp
, dip
, &xop
->head
.cluster
, -1);
1816 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1818 hammer2_inode_unlink_finisher(ip
, isopen
);
1819 hammer2_inode_unlock(ip
);
1822 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1826 * Update dip's mtime
1831 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1832 hammer2_update_time(&mtime
);
1833 hammer2_inode_modify(dip
);
1834 dip
->meta
.mtime
= mtime
;
1835 hammer2_inode_unlock(dip
);
1838 hammer2_inode_run_sideq(dip
->pmp
, 0);
1839 hammer2_trans_done(dip
->pmp
);
1841 cache_unlink(ap
->a_nch
);
1842 hammer2_knote(ap
->a_dvp
, NOTE_WRITE
);
1848 * hammer2_vop_nrmdir { nch, dvp, cred }
1852 hammer2_vop_nrmdir(struct vop_nrmdir_args
*ap
)
1854 hammer2_xop_unlink_t
*xop
;
1855 hammer2_inode_t
*dip
;
1856 hammer2_inode_t
*ip
;
1857 struct namecache
*ncp
;
1861 dip
= VTOI(ap
->a_dvp
);
1862 if (dip
->pmp
->ronly
)
1865 /* allow removals, except user to also bulkfree */
1866 if (hammer2_vfs_enospace(dip
, 0, ap
->a_cred
) > 1)
1870 hammer2_pfs_memory_wait(dip
->pmp
);
1871 hammer2_trans_init(dip
->pmp
, 0);
1872 hammer2_inode_lock(dip
, 0);
1874 xop
= hammer2_xop_alloc(dip
, HAMMER2_XOP_MODIFYING
);
1876 ncp
= ap
->a_nch
->ncp
;
1877 hammer2_xop_setname(&xop
->head
, ncp
->nc_name
, ncp
->nc_nlen
);
1878 isopen
= cache_isopen(ap
->a_nch
);
1880 xop
->dopermanent
= 0;
1881 hammer2_xop_start(&xop
->head
, hammer2_xop_unlink
);
1884 * Collect the real inode and adjust nlinks, destroy the real
1885 * inode if nlinks transitions to 0 and it was the real inode
1886 * (else it has already been removed).
1888 error
= hammer2_xop_collect(&xop
->head
, 0);
1889 error
= hammer2_error_to_errno(error
);
1890 hammer2_inode_unlock(dip
);
1893 ip
= hammer2_inode_get(dip
->pmp
, dip
, &xop
->head
.cluster
, -1);
1894 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1896 hammer2_inode_unlink_finisher(ip
, isopen
);
1897 hammer2_inode_unlock(ip
);
1900 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1904 * Update dip's mtime
1909 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1910 hammer2_update_time(&mtime
);
1911 hammer2_inode_modify(dip
);
1912 dip
->meta
.mtime
= mtime
;
1913 hammer2_inode_unlock(dip
);
1916 hammer2_inode_run_sideq(dip
->pmp
, 0);
1917 hammer2_trans_done(dip
->pmp
);
1919 cache_unlink(ap
->a_nch
);
1920 hammer2_knote(ap
->a_dvp
, NOTE_WRITE
| NOTE_LINK
);
1926 * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1930 hammer2_vop_nrename(struct vop_nrename_args
*ap
)
1932 struct namecache
*fncp
;
1933 struct namecache
*tncp
;
1934 hammer2_inode_t
*fdip
; /* source directory */
1935 hammer2_inode_t
*tdip
; /* target directory */
1936 hammer2_inode_t
*ip
; /* file being renamed */
1937 hammer2_inode_t
*tip
; /* replaced target during rename or NULL */
1938 const uint8_t *fname
;
1940 const uint8_t *tname
;
1947 if (ap
->a_fdvp
->v_mount
!= ap
->a_tdvp
->v_mount
)
1949 if (ap
->a_fdvp
->v_mount
!= ap
->a_fnch
->ncp
->nc_vp
->v_mount
)
1952 fdip
= VTOI(ap
->a_fdvp
); /* source directory */
1953 tdip
= VTOI(ap
->a_tdvp
); /* target directory */
1955 if (fdip
->pmp
->ronly
)
1957 if (hammer2_vfs_enospace(fdip
, 0, ap
->a_cred
) > 1)
1960 fncp
= ap
->a_fnch
->ncp
; /* entry name in source */
1961 fname
= fncp
->nc_name
;
1962 fname_len
= fncp
->nc_nlen
;
1964 tncp
= ap
->a_tnch
->ncp
; /* entry name in target */
1965 tname
= tncp
->nc_name
;
1966 tname_len
= tncp
->nc_nlen
;
1968 hammer2_pfs_memory_wait(tdip
->pmp
);
1969 hammer2_trans_init(tdip
->pmp
, 0);
1974 ip
= VTOI(fncp
->nc_vp
);
1975 hammer2_inode_ref(ip
); /* extra ref */
1978 * Lookup the target name to determine if a directory entry
1979 * is being overwritten. We only hold related inode locks
1980 * temporarily, the operating system is expected to protect
1981 * against rename races.
1983 tip
= tncp
->nc_vp
? VTOI(tncp
->nc_vp
) : NULL
;
1985 hammer2_inode_ref(tip
); /* extra ref */
1988 * Can return NULL and error == EXDEV if the common parent
1989 * crosses a directory with the xlink flag set.
1991 * For now try to avoid deadlocks with a simple pointer address
1992 * test. (tip) can be NULL.
1996 hammer2_inode_lock(fdip
, 0);
1997 hammer2_inode_lock(tdip
, 0);
1999 hammer2_inode_lock(tdip
, 0);
2000 hammer2_inode_lock(fdip
, 0);
2004 hammer2_inode_lock(ip
, 0);
2005 hammer2_inode_lock(tip
, 0);
2007 hammer2_inode_lock(tip
, 0);
2008 hammer2_inode_lock(ip
, 0);
2011 hammer2_inode_lock(ip
, 0);
2016 * Delete the target namespace.
2018 * REMOVED - NOW FOLDED INTO XOP_NRENAME OPERATION
2021 hammer2_xop_unlink_t
*xop2
;
2022 hammer2_inode_t
*tip
;
2026 * The unlink XOP unlinks the path from the directory and
2027 * locates and returns the cluster associated with the real
2028 * inode. We have to handle nlinks here on the frontend.
2030 xop2
= hammer2_xop_alloc(tdip
, HAMMER2_XOP_MODIFYING
);
2031 hammer2_xop_setname(&xop2
->head
, tname
, tname_len
);
2032 isopen
= cache_isopen(ap
->a_tnch
);
2034 xop2
->dopermanent
= 0;
2035 hammer2_xop_start(&xop2
->head
, hammer2_xop_unlink
);
2038 * Collect the real inode and adjust nlinks, destroy the real
2039 * inode if nlinks transitions to 0 and it was the real inode
2040 * (else it has already been removed).
2042 tnch_error
= hammer2_xop_collect(&xop2
->head
, 0);
2043 tnch_error
= hammer2_error_to_errno(tnch_error
);
2044 /* hammer2_inode_unlock(tdip); */
2046 if (tnch_error
== 0) {
2047 tip
= hammer2_inode_get(tdip
->pmp
, NULL
,
2048 &xop2
->head
.cluster
, -1);
2049 hammer2_xop_retire(&xop2
->head
, HAMMER2_XOPMASK_VOP
);
2051 hammer2_inode_unlink_finisher(tip
, isopen
);
2052 hammer2_inode_unlock(tip
);
2055 hammer2_xop_retire(&xop2
->head
, HAMMER2_XOPMASK_VOP
);
2057 /* hammer2_inode_lock(tdip, 0); */
2059 if (tnch_error
&& tnch_error
!= ENOENT
) {
2068 * Resolve the collision space for (tdip, tname, tname_len)
2070 * tdip must be held exclusively locked to prevent races since
2071 * multiple filenames can end up in the same collision space.
2074 hammer2_xop_scanlhc_t
*sxop
;
2075 hammer2_tid_t lhcbase
;
2077 tlhc
= hammer2_dirhash(tname
, tname_len
);
2079 sxop
= hammer2_xop_alloc(tdip
, HAMMER2_XOP_MODIFYING
);
2081 hammer2_xop_start(&sxop
->head
, hammer2_xop_scanlhc
);
2082 while ((error
= hammer2_xop_collect(&sxop
->head
, 0)) == 0) {
2083 if (tlhc
!= sxop
->head
.cluster
.focus
->bref
.key
)
2087 error
= hammer2_error_to_errno(error
);
2088 hammer2_xop_retire(&sxop
->head
, HAMMER2_XOPMASK_VOP
);
2091 if (error
!= ENOENT
)
2096 if ((lhcbase
^ tlhc
) & ~HAMMER2_DIRHASH_LOMASK
) {
2103 * Ready to go, issue the rename to the backend. Note that meta-data
2104 * updates to the related inodes occur separately from the rename
2107 * NOTE: While it is not necessary to update ip->meta.name*, doing
2108 * so aids catastrophic recovery and debugging.
2111 hammer2_xop_nrename_t
*xop4
;
2113 xop4
= hammer2_xop_alloc(fdip
, HAMMER2_XOP_MODIFYING
);
2115 xop4
->ip_key
= ip
->meta
.name_key
;
2116 hammer2_xop_setip2(&xop4
->head
, ip
);
2117 hammer2_xop_setip3(&xop4
->head
, tdip
);
2118 hammer2_xop_setname(&xop4
->head
, fname
, fname_len
);
2119 hammer2_xop_setname2(&xop4
->head
, tname
, tname_len
);
2120 hammer2_xop_start(&xop4
->head
, hammer2_xop_nrename
);
2122 error
= hammer2_xop_collect(&xop4
->head
, 0);
2123 error
= hammer2_error_to_errno(error
);
2124 hammer2_xop_retire(&xop4
->head
, HAMMER2_XOPMASK_VOP
);
2126 if (error
== ENOENT
)
2130 * Update inode meta-data.
2132 * WARNING! The in-memory inode (ip) structure does not
2133 * maintain a copy of the inode's filename buffer.
2136 (ip
->meta
.name_key
& HAMMER2_DIRHASH_VISIBLE
)) {
2137 hammer2_inode_modify(ip
);
2138 ip
->meta
.name_len
= tname_len
;
2139 ip
->meta
.name_key
= tlhc
;
2142 hammer2_inode_modify(ip
);
2143 ip
->meta
.iparent
= tdip
->meta
.inum
;
2151 * If no error, the backend has replaced the target directory entry.
2152 * We must adjust nlinks on the original replace target if it exists.
2154 if (error
== 0 && tip
) {
2157 isopen
= cache_isopen(ap
->a_tnch
);
2158 hammer2_inode_unlink_finisher(tip
, isopen
);
2162 * Update directory mtimes to represent the something changed.
2164 if (update_fdip
|| update_tdip
) {
2167 hammer2_update_time(&mtime
);
2169 hammer2_inode_modify(fdip
);
2170 fdip
->meta
.mtime
= mtime
;
2173 hammer2_inode_modify(tdip
);
2174 tdip
->meta
.mtime
= mtime
;
2178 hammer2_inode_unlock(tip
);
2179 hammer2_inode_drop(tip
);
2181 hammer2_inode_unlock(ip
);
2182 hammer2_inode_unlock(tdip
);
2183 hammer2_inode_unlock(fdip
);
2184 hammer2_inode_drop(ip
);
2185 hammer2_inode_run_sideq(fdip
->pmp
, 0);
2187 hammer2_trans_done(tdip
->pmp
);
2190 * Issue the namecache update after unlocking all the internal
2191 * hammer2 structures, otherwise we might deadlock.
2193 if (error
== 0 && tip
) {
2194 cache_unlink(ap
->a_tnch
);
2195 cache_setunresolved(ap
->a_tnch
);
2198 cache_rename(ap
->a_fnch
, ap
->a_tnch
);
2199 hammer2_knote(ap
->a_fdvp
, NOTE_WRITE
);
2200 hammer2_knote(ap
->a_tdvp
, NOTE_WRITE
);
2201 hammer2_knote(fncp
->nc_vp
, NOTE_RENAME
);
2208 * hammer2_vop_ioctl { vp, command, data, fflag, cred }
2212 hammer2_vop_ioctl(struct vop_ioctl_args
*ap
)
2214 hammer2_inode_t
*ip
;
2217 ip
= VTOI(ap
->a_vp
);
2219 error
= hammer2_ioctl(ip
, ap
->a_command
, (void *)ap
->a_data
,
2220 ap
->a_fflag
, ap
->a_cred
);
2226 hammer2_vop_mountctl(struct vop_mountctl_args
*ap
)
2233 case (MOUNTCTL_SET_EXPORT
):
2234 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
2237 if (ap
->a_ctllen
!= sizeof(struct export_args
))
2240 rc
= vfs_export(mp
, &pmp
->export
,
2241 (const struct export_args
*)ap
->a_ctl
);
2244 rc
= vop_stdmountctl(ap
);
2253 static void filt_hammer2detach(struct knote
*kn
);
2254 static int filt_hammer2read(struct knote
*kn
, long hint
);
2255 static int filt_hammer2write(struct knote
*kn
, long hint
);
2256 static int filt_hammer2vnode(struct knote
*kn
, long hint
);
2258 static struct filterops hammer2read_filtops
=
2259 { FILTEROP_ISFD
| FILTEROP_MPSAFE
,
2260 NULL
, filt_hammer2detach
, filt_hammer2read
};
2261 static struct filterops hammer2write_filtops
=
2262 { FILTEROP_ISFD
| FILTEROP_MPSAFE
,
2263 NULL
, filt_hammer2detach
, filt_hammer2write
};
2264 static struct filterops hammer2vnode_filtops
=
2265 { FILTEROP_ISFD
| FILTEROP_MPSAFE
,
2266 NULL
, filt_hammer2detach
, filt_hammer2vnode
};
2270 hammer2_vop_kqfilter(struct vop_kqfilter_args
*ap
)
2272 struct vnode
*vp
= ap
->a_vp
;
2273 struct knote
*kn
= ap
->a_kn
;
2275 switch (kn
->kn_filter
) {
2277 kn
->kn_fop
= &hammer2read_filtops
;
2280 kn
->kn_fop
= &hammer2write_filtops
;
2283 kn
->kn_fop
= &hammer2vnode_filtops
;
2286 return (EOPNOTSUPP
);
2289 kn
->kn_hook
= (caddr_t
)vp
;
2291 knote_insert(&vp
->v_pollinfo
.vpi_kqinfo
.ki_note
, kn
);
2297 filt_hammer2detach(struct knote
*kn
)
2299 struct vnode
*vp
= (void *)kn
->kn_hook
;
2301 knote_remove(&vp
->v_pollinfo
.vpi_kqinfo
.ki_note
, kn
);
2305 filt_hammer2read(struct knote
*kn
, long hint
)
2307 struct vnode
*vp
= (void *)kn
->kn_hook
;
2308 hammer2_inode_t
*ip
= VTOI(vp
);
2311 if (hint
== NOTE_REVOKE
) {
2312 kn
->kn_flags
|= (EV_EOF
| EV_NODATA
| EV_ONESHOT
);
2315 off
= ip
->meta
.size
- kn
->kn_fp
->f_offset
;
2316 kn
->kn_data
= (off
< INTPTR_MAX
) ? off
: INTPTR_MAX
;
2317 if (kn
->kn_sfflags
& NOTE_OLDAPI
)
2319 return (kn
->kn_data
!= 0);
2324 filt_hammer2write(struct knote
*kn
, long hint
)
2326 if (hint
== NOTE_REVOKE
)
2327 kn
->kn_flags
|= (EV_EOF
| EV_NODATA
| EV_ONESHOT
);
2333 filt_hammer2vnode(struct knote
*kn
, long hint
)
2335 if (kn
->kn_sfflags
& hint
)
2336 kn
->kn_fflags
|= hint
;
2337 if (hint
== NOTE_REVOKE
) {
2338 kn
->kn_flags
|= (EV_EOF
| EV_NODATA
);
2341 return (kn
->kn_fflags
!= 0);
2349 hammer2_vop_markatime(struct vop_markatime_args
*ap
)
2351 hammer2_inode_t
*ip
;
2364 hammer2_vop_fifokqfilter(struct vop_kqfilter_args
*ap
)
2368 error
= VOCALL(&fifo_vnode_vops
, &ap
->a_head
);
2370 error
= hammer2_vop_kqfilter(ap
);
2377 struct vop_ops hammer2_vnode_vops
= {
2378 .vop_default
= vop_defaultop
,
2379 .vop_fsync
= hammer2_vop_fsync
,
2380 .vop_getpages
= vop_stdgetpages
,
2381 .vop_putpages
= vop_stdputpages
,
2382 .vop_access
= hammer2_vop_access
,
2383 .vop_advlock
= hammer2_vop_advlock
,
2384 .vop_close
= hammer2_vop_close
,
2385 .vop_nlink
= hammer2_vop_nlink
,
2386 .vop_ncreate
= hammer2_vop_ncreate
,
2387 .vop_nsymlink
= hammer2_vop_nsymlink
,
2388 .vop_nremove
= hammer2_vop_nremove
,
2389 .vop_nrmdir
= hammer2_vop_nrmdir
,
2390 .vop_nrename
= hammer2_vop_nrename
,
2391 .vop_getattr
= hammer2_vop_getattr
,
2392 .vop_setattr
= hammer2_vop_setattr
,
2393 .vop_readdir
= hammer2_vop_readdir
,
2394 .vop_readlink
= hammer2_vop_readlink
,
2395 .vop_getpages
= vop_stdgetpages
,
2396 .vop_putpages
= vop_stdputpages
,
2397 .vop_read
= hammer2_vop_read
,
2398 .vop_write
= hammer2_vop_write
,
2399 .vop_open
= hammer2_vop_open
,
2400 .vop_inactive
= hammer2_vop_inactive
,
2401 .vop_reclaim
= hammer2_vop_reclaim
,
2402 .vop_nresolve
= hammer2_vop_nresolve
,
2403 .vop_nlookupdotdot
= hammer2_vop_nlookupdotdot
,
2404 .vop_nmkdir
= hammer2_vop_nmkdir
,
2405 .vop_nmknod
= hammer2_vop_nmknod
,
2406 .vop_ioctl
= hammer2_vop_ioctl
,
2407 .vop_mountctl
= hammer2_vop_mountctl
,
2408 .vop_bmap
= hammer2_vop_bmap
,
2409 .vop_strategy
= hammer2_vop_strategy
,
2410 .vop_kqfilter
= hammer2_vop_kqfilter
2413 struct vop_ops hammer2_spec_vops
= {
2414 .vop_default
= vop_defaultop
,
2415 .vop_fsync
= hammer2_vop_fsync
,
2416 .vop_read
= vop_stdnoread
,
2417 .vop_write
= vop_stdnowrite
,
2418 .vop_access
= hammer2_vop_access
,
2419 .vop_close
= hammer2_vop_close
,
2420 .vop_markatime
= hammer2_vop_markatime
,
2421 .vop_getattr
= hammer2_vop_getattr
,
2422 .vop_inactive
= hammer2_vop_inactive
,
2423 .vop_reclaim
= hammer2_vop_reclaim
,
2424 .vop_setattr
= hammer2_vop_setattr
2427 struct vop_ops hammer2_fifo_vops
= {
2428 .vop_default
= fifo_vnoperate
,
2429 .vop_fsync
= hammer2_vop_fsync
,
2431 .vop_read
= hammer2_vop_fiforead
,
2432 .vop_write
= hammer2_vop_fifowrite
,
2434 .vop_access
= hammer2_vop_access
,
2436 .vop_close
= hammer2_vop_fifoclose
,
2438 .vop_markatime
= hammer2_vop_markatime
,
2439 .vop_getattr
= hammer2_vop_getattr
,
2440 .vop_inactive
= hammer2_vop_inactive
,
2441 .vop_reclaim
= hammer2_vop_reclaim
,
2442 .vop_setattr
= hammer2_vop_setattr
,
2443 .vop_kqfilter
= hammer2_vop_fifokqfilter