2 * Copyright (c) 2011-2018 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_SYNC / 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 HAMMER2_INODE_DIRTYDATA
)) &&
179 (ip
->flags
& HAMMER2_INODE_ISDELETED
) == 0) {
180 hammer2_inode_sideq_t
*ipul
;
182 ipul
= kmalloc(sizeof(*ipul
), pmp
->minode
, M_WAITOK
| M_ZERO
);
185 hammer2_spin_ex(&pmp
->list_spin
);
186 if ((ip
->flags
& HAMMER2_INODE_ONSIDEQ
) == 0) {
188 atomic_set_int(&ip
->flags
, HAMMER2_INODE_ONSIDEQ
);
189 TAILQ_INSERT_TAIL(&pmp
->sideq
, ipul
, entry
);
191 hammer2_spin_unex(&pmp
->list_spin
);
193 hammer2_spin_unex(&pmp
->list_spin
);
194 kfree(ipul
, pmp
->minode
);
195 hammer2_inode_drop(ip
); /* vp ref */
197 /* retain ref from vp for ipul */
199 hammer2_inode_drop(ip
); /* vp ref */
203 * XXX handle background sync when ip dirty, kernel will no longer
204 * notify us regarding this inode because there is no longer a
205 * vnode attached to it.
212 * Currently this function synchronizes the front-end inode state to the
213 * backend chain topology, then flushes the inode's chain and sub-topology
214 * to backend media. This function does not flush the root topology down to
219 hammer2_vop_fsync(struct vop_fsync_args
*ap
)
230 hammer2_trans_init(ip
->pmp
, 0);
233 * Clean out buffer cache, wait for I/O's to complete.
235 vfsync(vp
, ap
->a_waitfor
, 1, NULL
, NULL
);
236 bio_track_wait(&vp
->v_track_write
, 0, 0);
239 * Flush any inode changes
241 hammer2_inode_lock(ip
, 0);
242 if (ip
->flags
& (HAMMER2_INODE_RESIZED
|HAMMER2_INODE_MODIFIED
))
243 error1
= hammer2_inode_chain_sync(ip
);
246 * Flush dirty chains related to the inode.
248 * NOTE! XXX We do not currently flush to the volume root, ultimately
249 * we will want to have a shortcut for the flushed inode stored
250 * in the volume root for recovery purposes.
252 error2
= hammer2_inode_chain_flush(ip
);
255 hammer2_inode_unlock(ip
);
256 hammer2_trans_done(ip
->pmp
);
263 hammer2_vop_access(struct vop_access_args
*ap
)
265 hammer2_inode_t
*ip
= VTOI(ap
->a_vp
);
270 hammer2_inode_lock(ip
, HAMMER2_RESOLVE_SHARED
);
271 uid
= hammer2_to_unix_xid(&ip
->meta
.uid
);
272 gid
= hammer2_to_unix_xid(&ip
->meta
.gid
);
273 error
= vop_helper_access(ap
, uid
, gid
, ip
->meta
.mode
, ip
->meta
.uflags
);
274 hammer2_inode_unlock(ip
);
281 hammer2_vop_getattr(struct vop_getattr_args
*ap
)
287 hammer2_chain_t
*chain
;
296 hammer2_inode_lock(ip
, HAMMER2_RESOLVE_SHARED
);
298 vap
->va_fsid
= pmp
->mp
->mnt_stat
.f_fsid
.val
[0];
299 vap
->va_fileid
= ip
->meta
.inum
;
300 vap
->va_mode
= ip
->meta
.mode
;
301 vap
->va_nlink
= ip
->meta
.nlinks
;
302 vap
->va_uid
= hammer2_to_unix_xid(&ip
->meta
.uid
);
303 vap
->va_gid
= hammer2_to_unix_xid(&ip
->meta
.gid
);
306 vap
->va_size
= ip
->meta
.size
; /* protected by shared lock */
307 vap
->va_blocksize
= HAMMER2_PBUFSIZE
;
308 vap
->va_flags
= ip
->meta
.uflags
;
309 hammer2_time_to_timespec(ip
->meta
.ctime
, &vap
->va_ctime
);
310 hammer2_time_to_timespec(ip
->meta
.mtime
, &vap
->va_mtime
);
311 hammer2_time_to_timespec(ip
->meta
.mtime
, &vap
->va_atime
);
314 if (ip
->meta
.type
== HAMMER2_OBJTYPE_DIRECTORY
) {
316 * Can't really calculate directory use sans the files under
317 * it, just assume one block for now.
319 vap
->va_bytes
+= HAMMER2_INODE_BYTES
;
321 for (i
= 0; i
< ip
->cluster
.nchains
; ++i
) {
322 if ((chain
= ip
->cluster
.array
[i
].chain
) != NULL
) {
324 chain
->bref
.embed
.stats
.data_count
) {
326 chain
->bref
.embed
.stats
.data_count
;
331 vap
->va_type
= hammer2_get_vtype(ip
->meta
.type
);
333 vap
->va_uid_uuid
= ip
->meta
.uid
;
334 vap
->va_gid_uuid
= ip
->meta
.gid
;
335 vap
->va_vaflags
= VA_UID_UUID_VALID
| VA_GID_UUID_VALID
|
338 hammer2_inode_unlock(ip
);
345 hammer2_vop_setattr(struct vop_setattr_args
*ap
)
356 hammer2_update_time(&ctime
);
362 if (hammer2_vfs_enospace(ip
, 0, ap
->a_cred
) > 1)
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
);
434 kflags
|= NOTE_WRITE
;
436 hammer2_extend_file(ip
, vap
->va_size
);
437 kflags
|= NOTE_WRITE
| NOTE_EXTEND
;
439 hammer2_inode_modify(ip
);
440 ip
->meta
.mtime
= ctime
;
441 vclrflags(vp
, VLASTWRITETS
);
449 /* atime not supported */
450 if (vap
->va_atime
.tv_sec
!= VNOVAL
) {
451 hammer2_inode_modify(ip
);
452 ip
->meta
.atime
= hammer2_timespec_to_time(&vap
->va_atime
);
453 kflags
|= NOTE_ATTRIB
;
456 if (vap
->va_mode
!= (mode_t
)VNOVAL
) {
457 mode_t cur_mode
= ip
->meta
.mode
;
458 uid_t cur_uid
= hammer2_to_unix_xid(&ip
->meta
.uid
);
459 gid_t cur_gid
= hammer2_to_unix_xid(&ip
->meta
.gid
);
461 error
= vop_helper_chmod(ap
->a_vp
, vap
->va_mode
, ap
->a_cred
,
462 cur_uid
, cur_gid
, &cur_mode
);
463 if (error
== 0 && ip
->meta
.mode
!= cur_mode
) {
464 hammer2_inode_modify(ip
);
465 ip
->meta
.mode
= cur_mode
;
466 ip
->meta
.ctime
= ctime
;
467 kflags
|= NOTE_ATTRIB
;
471 if (vap
->va_mtime
.tv_sec
!= VNOVAL
) {
472 hammer2_inode_modify(ip
);
473 ip
->meta
.mtime
= hammer2_timespec_to_time(&vap
->va_mtime
);
474 kflags
|= NOTE_ATTRIB
;
475 vclrflags(vp
, VLASTWRITETS
);
480 * If a truncation occurred we must call chain_sync() now in order
481 * to trim the related data chains, otherwise a later expansion can
484 * If an extend occured that changed the DIRECTDATA state, we must
485 * call inode_fsync now in order to prepare the inode's indirect
488 * WARNING! This means we are making an adjustment to the inode's
489 * chain outside of sync/fsync, and not just to inode->meta, which
490 * may result in some consistency issues if a crash were to occur
491 * at just the wrong time.
493 if (ip
->flags
& HAMMER2_INODE_RESIZED
)
494 hammer2_inode_chain_sync(ip
);
499 hammer2_inode_unlock(ip
);
500 hammer2_trans_done(ip
->pmp
);
501 hammer2_knote(ip
->vp
, kflags
);
508 hammer2_vop_readdir(struct vop_readdir_args
*ap
)
510 hammer2_xop_readdir_t
*xop
;
511 hammer2_blockref_t bref
;
526 saveoff
= uio
->uio_offset
;
531 * Setup cookies directory entry cookies if requested
533 if (ap
->a_ncookies
) {
534 ncookies
= uio
->uio_resid
/ 16 + 1;
537 cookies
= kmalloc(ncookies
* sizeof(off_t
), M_TEMP
, M_WAITOK
);
544 hammer2_inode_lock(ip
, HAMMER2_RESOLVE_SHARED
);
547 * Handle artificial entries. To ensure that only positive 64 bit
548 * quantities are returned to userland we always strip off bit 63.
549 * The hash code is designed such that codes 0x0000-0x7FFF are not
550 * used, allowing us to use these codes for articial entries.
552 * Entry 0 is used for '.' and entry 1 is used for '..'. Do not
553 * allow '..' to cross the mount point into (e.g.) the super-root.
556 inum
= ip
->meta
.inum
& HAMMER2_DIRHASH_USERMSK
;
557 r
= vop_write_dirent(&error
, uio
, inum
, DT_DIR
, 1, ".");
561 cookies
[cookie_index
] = saveoff
;
564 if (cookie_index
== ncookies
)
570 * Be careful with lockorder when accessing ".."
572 * (ip is the current dir. xip is the parent dir).
574 inum
= ip
->meta
.inum
& HAMMER2_DIRHASH_USERMSK
;
575 if (ip
!= ip
->pmp
->iroot
)
576 inum
= ip
->meta
.iparent
& HAMMER2_DIRHASH_USERMSK
;
577 r
= vop_write_dirent(&error
, uio
, inum
, DT_DIR
, 2, "..");
581 cookies
[cookie_index
] = saveoff
;
584 if (cookie_index
== ncookies
)
588 lkey
= saveoff
| HAMMER2_DIRHASH_VISIBLE
;
589 if (hammer2_debug
& 0x0020)
590 kprintf("readdir: lkey %016jx\n", lkey
);
595 * Use XOP for cluster scan.
597 * parent is the inode cluster, already locked for us. Don't
598 * double lock shared locks as this will screw up upgrades.
600 xop
= hammer2_xop_alloc(ip
, 0);
602 hammer2_xop_start(&xop
->head
, hammer2_xop_readdir
);
605 const hammer2_inode_data_t
*ripdata
;
609 error
= hammer2_xop_collect(&xop
->head
, 0);
610 error
= hammer2_error_to_errno(error
);
614 if (cookie_index
== ncookies
)
616 if (hammer2_debug
& 0x0020)
617 kprintf("cluster chain %p %p\n",
618 xop
->head
.cluster
.focus
,
619 (xop
->head
.cluster
.focus
?
620 xop
->head
.cluster
.focus
->data
: (void *)-1));
621 hammer2_cluster_bref(&xop
->head
.cluster
, &bref
);
623 if (bref
.type
== HAMMER2_BREF_TYPE_INODE
) {
625 &hammer2_cluster_rdata(&xop
->head
.cluster
)->ipdata
;
626 dtype
= hammer2_get_dtype(ripdata
->meta
.type
);
627 saveoff
= bref
.key
& HAMMER2_DIRHASH_USERMSK
;
628 r
= vop_write_dirent(&error
, uio
,
630 HAMMER2_DIRHASH_USERMSK
,
632 ripdata
->meta
.name_len
,
637 cookies
[cookie_index
] = saveoff
;
639 } else if (bref
.type
== HAMMER2_BREF_TYPE_DIRENT
) {
640 dtype
= hammer2_get_dtype(bref
.embed
.dirent
.type
);
641 saveoff
= bref
.key
& HAMMER2_DIRHASH_USERMSK
;
642 if (bref
.embed
.dirent
.namlen
<=
643 sizeof(bref
.check
.buf
)) {
644 dname
= bref
.check
.buf
;
647 hammer2_cluster_rdata(&xop
->head
.cluster
)->buf
;
649 r
= vop_write_dirent(&error
, uio
,
650 bref
.embed
.dirent
.inum
,
652 bref
.embed
.dirent
.namlen
,
657 cookies
[cookie_index
] = saveoff
;
660 /* XXX chain error */
661 kprintf("bad chain type readdir %d\n", bref
.type
);
664 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
665 if (error
== ENOENT
) {
668 saveoff
= (hammer2_key_t
)-1;
670 saveoff
= bref
.key
& HAMMER2_DIRHASH_USERMSK
;
673 hammer2_inode_unlock(ip
);
675 *ap
->a_eofflag
= eofflag
;
676 if (hammer2_debug
& 0x0020)
677 kprintf("readdir: done at %016jx\n", saveoff
);
678 uio
->uio_offset
= saveoff
& ~HAMMER2_DIRHASH_VISIBLE
;
679 if (error
&& cookie_index
== 0) {
681 kfree(cookies
, M_TEMP
);
683 *ap
->a_cookies
= NULL
;
687 *ap
->a_ncookies
= cookie_index
;
688 *ap
->a_cookies
= cookies
;
695 * hammer2_vop_readlink { vp, uio, cred }
699 hammer2_vop_readlink(struct vop_readlink_args
*ap
)
706 if (vp
->v_type
!= VLNK
)
710 error
= hammer2_read_file(ip
, ap
->a_uio
, 0);
716 hammer2_vop_read(struct vop_read_args
*ap
)
726 * Read operations supported on this vnode?
729 if (vp
->v_type
!= VREG
)
739 seqcount
= ap
->a_ioflag
>> 16;
740 bigread
= (uio
->uio_resid
> 100 * 1024 * 1024);
742 error
= hammer2_read_file(ip
, uio
, seqcount
);
748 hammer2_vop_write(struct vop_write_args
*ap
)
759 * Read operations supported on this vnode?
762 if (vp
->v_type
!= VREG
)
769 ioflag
= ap
->a_ioflag
;
774 switch (hammer2_vfs_enospace(ip
, uio
->uio_resid
, ap
->a_cred
)) {
778 ioflag
|= IO_DIRECT
; /* semi-synchronous */
784 seqcount
= ioflag
>> 16;
787 * Check resource limit
789 if (uio
->uio_resid
> 0 && (td
= uio
->uio_td
) != NULL
&& td
->td_proc
&&
790 uio
->uio_offset
+ uio
->uio_resid
>
791 td
->td_proc
->p_rlimit
[RLIMIT_FSIZE
].rlim_cur
) {
792 lwpsignal(td
->td_proc
, td
->td_lwp
, SIGXFSZ
);
797 * The transaction interlocks against flush initiations
798 * (note: but will run concurrently with the actual flush).
800 * To avoid deadlocking against the VM system, we must flag any
801 * transaction related to the buffer cache or other direct
802 * VM page manipulation.
804 if (uio
->uio_segflg
== UIO_NOCOPY
)
805 hammer2_trans_init(ip
->pmp
, HAMMER2_TRANS_BUFCACHE
);
807 hammer2_trans_init(ip
->pmp
, 0);
808 error
= hammer2_write_file(ip
, uio
, ioflag
, seqcount
);
809 hammer2_trans_done(ip
->pmp
);
815 * Perform read operations on a file or symlink given an UNLOCKED
818 * The passed ip is not locked.
822 hammer2_read_file(hammer2_inode_t
*ip
, struct uio
*uio
, int seqcount
)
833 * WARNING! Assumes that the kernel interlocks size changes at the
836 hammer2_mtx_sh(&ip
->lock
);
837 hammer2_mtx_sh(&ip
->truncate_lock
);
838 size
= ip
->meta
.size
;
839 hammer2_mtx_unlock(&ip
->lock
);
841 while (uio
->uio_resid
> 0 && uio
->uio_offset
< size
) {
848 lblksize
= hammer2_calc_logical(ip
, uio
->uio_offset
,
853 error
= cluster_readx(ip
->vp
, leof
, lbase
, lblksize
,
854 B_NOTMETA
| B_KVABIO
,
859 if (uio
->uio_segflg
== UIO_NOCOPY
) {
860 bp
= getblk(ip
->vp
, lbase
, lblksize
,
861 GETBLK_BHEAVY
| GETBLK_KVABIO
, 0);
862 if (bp
->b_flags
& B_CACHE
) {
865 if (bp
->b_xio
.xio_npages
!= 16)
866 kprintf("NPAGES BAD\n");
867 for (i
= 0; i
< bp
->b_xio
.xio_npages
; ++i
) {
869 m
= bp
->b_xio
.xio_pages
[i
];
870 if (m
== NULL
|| m
->valid
== 0) {
871 kprintf("bp %016jx %016jx pg %d inv",
874 kprintf("m->object %p/%p", m
->object
, ip
->vp
->v_object
);
880 kprintf("b_flags %08x, b_error %d\n", bp
->b_flags
, bp
->b_error
);
884 error
= bread_kvabio(ip
->vp
, lbase
, lblksize
, &bp
);
891 loff
= (int)(uio
->uio_offset
- lbase
);
893 if (n
> uio
->uio_resid
)
895 if (n
> size
- uio
->uio_offset
)
896 n
= (int)(size
- uio
->uio_offset
);
897 bp
->b_flags
|= B_AGE
;
898 uiomovebp(bp
, (char *)bp
->b_data
+ loff
, n
, uio
);
901 hammer2_mtx_unlock(&ip
->truncate_lock
);
907 * Write to the file represented by the inode via the logical buffer cache.
908 * The inode may represent a regular file or a symlink.
910 * The inode must not be locked.
914 hammer2_write_file(hammer2_inode_t
*ip
, struct uio
*uio
,
915 int ioflag
, int seqcount
)
917 hammer2_key_t old_eof
;
918 hammer2_key_t new_eof
;
927 * WARNING! Assumes that the kernel interlocks size changes at the
930 hammer2_mtx_ex(&ip
->lock
);
931 hammer2_mtx_sh(&ip
->truncate_lock
);
932 if (ioflag
& IO_APPEND
)
933 uio
->uio_offset
= ip
->meta
.size
;
934 old_eof
= ip
->meta
.size
;
937 * Extend the file if necessary. If the write fails at some point
938 * we will truncate it back down to cover as much as we were able
941 * Doing this now makes it easier to calculate buffer sizes in
948 if (uio
->uio_offset
+ uio
->uio_resid
> old_eof
) {
949 new_eof
= uio
->uio_offset
+ uio
->uio_resid
;
951 hammer2_extend_file(ip
, new_eof
);
952 kflags
|= NOTE_EXTEND
;
956 hammer2_mtx_unlock(&ip
->lock
);
961 while (uio
->uio_resid
> 0) {
970 * Don't allow the buffer build to blow out the buffer
973 if ((ioflag
& IO_RECURSE
) == 0)
974 bwillwrite(HAMMER2_PBUFSIZE
);
977 * This nominally tells us how much we can cluster and
978 * what the logical buffer size needs to be. Currently
979 * we don't try to cluster the write and just handle one
982 lblksize
= hammer2_calc_logical(ip
, uio
->uio_offset
,
984 loff
= (int)(uio
->uio_offset
- lbase
);
986 KKASSERT(lblksize
<= 65536);
989 * Calculate bytes to copy this transfer and whether the
990 * copy completely covers the buffer or not.
994 if (n
> uio
->uio_resid
) {
996 if (loff
== lbase
&& uio
->uio_offset
+ n
== new_eof
)
1004 if (lbase
>= new_eof
)
1010 if (uio
->uio_segflg
== UIO_NOCOPY
) {
1012 * Issuing a write with the same data backing the
1013 * buffer. Instantiate the buffer to collect the
1014 * backing vm pages, then read-in any missing bits.
1016 * This case is used by vop_stdputpages().
1018 bp
= getblk(ip
->vp
, lbase
, lblksize
,
1019 GETBLK_BHEAVY
| GETBLK_KVABIO
, 0);
1020 if ((bp
->b_flags
& B_CACHE
) == 0) {
1022 error
= bread_kvabio(ip
->vp
, lbase
,
1025 } else if (trivial
) {
1027 * Even though we are entirely overwriting the buffer
1028 * we may still have to zero it out to avoid a
1029 * mmap/write visibility issue.
1031 bp
= getblk(ip
->vp
, lbase
, lblksize
,
1032 GETBLK_BHEAVY
| GETBLK_KVABIO
, 0);
1033 if ((bp
->b_flags
& B_CACHE
) == 0)
1037 * Partial overwrite, read in any missing bits then
1038 * replace the portion being written.
1040 * (The strategy code will detect zero-fill physical
1041 * blocks for this case).
1043 error
= bread_kvabio(ip
->vp
, lbase
, lblksize
, &bp
);
1054 * Ok, copy the data in
1057 error
= uiomovebp(bp
, bp
->b_data
+ loff
, n
, uio
);
1058 kflags
|= NOTE_WRITE
;
1066 * WARNING: Pageout daemon will issue UIO_NOCOPY writes
1067 * with IO_SYNC or IO_ASYNC set. These writes
1068 * must be handled as the pageout daemon expects.
1070 * NOTE! H2 relies on cluster_write() here because it
1071 * cannot preallocate disk blocks at the logical
1072 * level due to not knowing what the compression
1073 * size will be at this time.
1075 * We must use cluster_write() here and we depend
1076 * on the write-behind feature to flush buffers
1077 * appropriately. If we let the buffer daemons do
1078 * it the block allocations will be all over the
1081 if (ioflag
& IO_SYNC
) {
1083 } else if ((ioflag
& IO_DIRECT
) && endofblk
) {
1085 } else if (ioflag
& IO_ASYNC
) {
1087 } else if (ip
->vp
->v_mount
->mnt_flag
& MNT_NOCLUSTERW
) {
1091 bp
->b_flags
|= B_CLUSTEROK
;
1092 cluster_write(bp
, new_eof
, lblksize
, seqcount
);
1094 bp
->b_flags
|= B_CLUSTEROK
;
1101 * Cleanup. If we extended the file EOF but failed to write through
1102 * the entire write is a failure and we have to back-up.
1104 if (error
&& new_eof
!= old_eof
) {
1105 hammer2_mtx_unlock(&ip
->truncate_lock
);
1106 hammer2_mtx_ex(&ip
->lock
);
1107 hammer2_mtx_ex(&ip
->truncate_lock
);
1108 hammer2_truncate_file(ip
, old_eof
);
1109 if (ip
->flags
& HAMMER2_INODE_MODIFIED
)
1110 hammer2_inode_chain_sync(ip
);
1111 hammer2_mtx_unlock(&ip
->lock
);
1112 } else if (modified
) {
1113 struct vnode
*vp
= ip
->vp
;
1115 hammer2_mtx_ex(&ip
->lock
);
1116 hammer2_inode_modify(ip
);
1117 if (uio
->uio_segflg
== UIO_NOCOPY
) {
1118 if (vp
->v_flag
& VLASTWRITETS
) {
1120 (unsigned long)vp
->v_lastwrite_ts
.tv_sec
*
1122 vp
->v_lastwrite_ts
.tv_nsec
/ 1000;
1125 hammer2_update_time(&ip
->meta
.mtime
);
1126 vclrflags(vp
, VLASTWRITETS
);
1131 * REMOVED - handled by hammer2_extend_file(). Do not issue
1132 * a chain_sync() outside of a sync/fsync except for DIRECTDATA
1135 * Under normal conditions we only issue a chain_sync if
1136 * the inode's DIRECTDATA state changed.
1138 if (ip
->flags
& HAMMER2_INODE_RESIZED
)
1139 hammer2_inode_chain_sync(ip
);
1141 hammer2_mtx_unlock(&ip
->lock
);
1142 hammer2_knote(ip
->vp
, kflags
);
1144 hammer2_trans_assert_strategy(ip
->pmp
);
1145 hammer2_mtx_unlock(&ip
->truncate_lock
);
1151 * Truncate the size of a file. The inode must not be locked.
1153 * We must unconditionally set HAMMER2_INODE_RESIZED to properly
1154 * ensure that any on-media data beyond the new file EOF has been destroyed.
1156 * WARNING: nvtruncbuf() can only be safely called without the inode lock
1157 * held due to the way our write thread works. If the truncation
1158 * occurs in the middle of a buffer, nvtruncbuf() is responsible
1159 * for dirtying that buffer and zeroing out trailing bytes.
1161 * WARNING! Assumes that the kernel interlocks size changes at the
1164 * WARNING! Caller assumes responsibility for removing dead blocks
1165 * if INODE_RESIZED is set.
1169 hammer2_truncate_file(hammer2_inode_t
*ip
, hammer2_key_t nsize
)
1171 hammer2_key_t lbase
;
1174 hammer2_mtx_unlock(&ip
->lock
);
1176 nblksize
= hammer2_calc_logical(ip
, nsize
, &lbase
, NULL
);
1177 nvtruncbuf(ip
->vp
, nsize
,
1178 nblksize
, (int)nsize
& (nblksize
- 1),
1181 hammer2_mtx_ex(&ip
->lock
);
1182 KKASSERT((ip
->flags
& HAMMER2_INODE_RESIZED
) == 0);
1183 ip
->osize
= ip
->meta
.size
;
1184 ip
->meta
.size
= nsize
;
1185 atomic_set_int(&ip
->flags
, HAMMER2_INODE_RESIZED
);
1186 hammer2_inode_modify(ip
);
1190 * Extend the size of a file. The inode must not be locked.
1192 * Even though the file size is changing, we do not have to set the
1193 * INODE_RESIZED bit unless the file size crosses the EMBEDDED_BYTES
1194 * boundary. When this occurs a hammer2_inode_chain_sync() is required
1195 * to prepare the inode cluster's indirect block table, otherwise
1196 * async execution of the strategy code will implode on us.
1198 * WARNING! Assumes that the kernel interlocks size changes at the
1201 * WARNING! Caller assumes responsibility for transitioning out
1202 * of the inode DIRECTDATA mode if INODE_RESIZED is set.
1206 hammer2_extend_file(hammer2_inode_t
*ip
, hammer2_key_t nsize
)
1208 hammer2_key_t lbase
;
1209 hammer2_key_t osize
;
1213 KKASSERT((ip
->flags
& HAMMER2_INODE_RESIZED
) == 0);
1214 hammer2_inode_modify(ip
);
1215 osize
= ip
->meta
.size
;
1217 ip
->meta
.size
= nsize
;
1220 * We must issue a chain_sync() when the DIRECTDATA state changes
1221 * to prevent confusion between the flush code and the in-memory
1222 * state. This is not perfect because we are doing it outside of
1223 * a sync/fsync operation, so it might not be fully synchronized
1224 * with the meta-data topology flush.
1226 if (osize
<= HAMMER2_EMBEDDED_BYTES
&& nsize
> HAMMER2_EMBEDDED_BYTES
) {
1227 atomic_set_int(&ip
->flags
, HAMMER2_INODE_RESIZED
);
1228 hammer2_inode_chain_sync(ip
);
1231 hammer2_mtx_unlock(&ip
->lock
);
1233 oblksize
= hammer2_calc_logical(ip
, osize
, &lbase
, NULL
);
1234 nblksize
= hammer2_calc_logical(ip
, nsize
, &lbase
, NULL
);
1240 hammer2_mtx_ex(&ip
->lock
);
1245 hammer2_vop_nresolve(struct vop_nresolve_args
*ap
)
1247 hammer2_xop_nresolve_t
*xop
;
1248 hammer2_inode_t
*ip
;
1249 hammer2_inode_t
*dip
;
1250 struct namecache
*ncp
;
1254 dip
= VTOI(ap
->a_dvp
);
1255 xop
= hammer2_xop_alloc(dip
, 0);
1257 ncp
= ap
->a_nch
->ncp
;
1258 hammer2_xop_setname(&xop
->head
, ncp
->nc_name
, ncp
->nc_nlen
);
1261 * Note: In DragonFly the kernel handles '.' and '..'.
1263 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1264 hammer2_xop_start(&xop
->head
, hammer2_xop_nresolve
);
1266 error
= hammer2_xop_collect(&xop
->head
, 0);
1267 error
= hammer2_error_to_errno(error
);
1271 ip
= hammer2_inode_get(dip
->pmp
, dip
, &xop
->head
.cluster
, -1);
1273 hammer2_inode_unlock(dip
);
1276 * Acquire the related vnode
1278 * NOTE: For error processing, only ENOENT resolves the namecache
1279 * entry to NULL, otherwise we just return the error and
1280 * leave the namecache unresolved.
1282 * NOTE: multiple hammer2_inode structures can be aliased to the
1283 * same chain element, for example for hardlinks. This
1284 * use case does not 'reattach' inode associations that
1285 * might already exist, but always allocates a new one.
1287 * WARNING: inode structure is locked exclusively via inode_get
1288 * but chain was locked shared. inode_unlock()
1289 * will handle it properly.
1292 vp
= hammer2_igetv(ip
, &error
); /* error set to UNIX error */
1295 cache_setvp(ap
->a_nch
, vp
);
1296 } else if (error
== ENOENT
) {
1297 cache_setvp(ap
->a_nch
, NULL
);
1299 hammer2_inode_unlock(ip
);
1302 * The vp should not be released until after we've disposed
1303 * of our locks, because it might cause vop_inactive() to
1310 cache_setvp(ap
->a_nch
, NULL
);
1312 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1313 KASSERT(error
|| ap
->a_nch
->ncp
->nc_vp
!= NULL
,
1314 ("resolve error %d/%p ap %p\n",
1315 error
, ap
->a_nch
->ncp
->nc_vp
, ap
));
1322 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args
*ap
)
1324 hammer2_inode_t
*dip
;
1328 dip
= VTOI(ap
->a_dvp
);
1329 inum
= dip
->meta
.iparent
;
1333 error
= hammer2_vfs_vget(ap
->a_dvp
->v_mount
, NULL
,
1343 hammer2_vop_nmkdir(struct vop_nmkdir_args
*ap
)
1345 hammer2_inode_t
*dip
;
1346 hammer2_inode_t
*nip
;
1347 struct namecache
*ncp
;
1348 const uint8_t *name
;
1353 dip
= VTOI(ap
->a_dvp
);
1354 if (dip
->pmp
->ronly
)
1356 if (hammer2_vfs_enospace(dip
, 0, ap
->a_cred
) > 1)
1359 ncp
= ap
->a_nch
->ncp
;
1360 name
= ncp
->nc_name
;
1361 name_len
= ncp
->nc_nlen
;
1363 hammer2_pfs_memory_wait(dip
->pmp
);
1364 hammer2_trans_init(dip
->pmp
, 0);
1366 inum
= hammer2_trans_newinum(dip
->pmp
);
1369 * Create the actual inode as a hidden file in the iroot, then
1370 * create the directory entry. The creation of the actual inode
1371 * sets its nlinks to 1 which is the value we desire.
1373 nip
= hammer2_inode_create(dip
->pmp
->iroot
, dip
, ap
->a_vap
, ap
->a_cred
,
1378 error
= hammer2_error_to_errno(error
);
1380 error
= hammer2_dirent_create(dip
, name
, name_len
,
1381 nip
->meta
.inum
, nip
->meta
.type
);
1382 /* returns UNIX error code */
1386 hammer2_inode_unlink_finisher(nip
, 0);
1387 hammer2_inode_unlock(nip
);
1392 *ap
->a_vpp
= hammer2_igetv(nip
, &error
);
1393 hammer2_inode_unlock(nip
);
1397 * Update dip's mtime
1399 * We can use a shared inode lock and allow the meta.mtime update
1400 * SMP race. hammer2_inode_modify() is MPSAFE w/a shared lock.
1405 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1406 hammer2_update_time(&mtime
);
1407 hammer2_inode_modify(dip
);
1408 dip
->meta
.mtime
= mtime
;
1409 hammer2_inode_unlock(dip
);
1412 hammer2_trans_done(dip
->pmp
);
1415 cache_setunresolved(ap
->a_nch
);
1416 cache_setvp(ap
->a_nch
, *ap
->a_vpp
);
1417 hammer2_knote(ap
->a_dvp
, NOTE_WRITE
| NOTE_LINK
);
1424 hammer2_vop_open(struct vop_open_args
*ap
)
1426 return vop_stdopen(ap
);
1430 * hammer2_vop_advlock { vp, id, op, fl, flags }
1434 hammer2_vop_advlock(struct vop_advlock_args
*ap
)
1436 hammer2_inode_t
*ip
= VTOI(ap
->a_vp
);
1439 size
= ip
->meta
.size
;
1440 return (lf_advlock(ap
, &ip
->advlock
, size
));
1445 hammer2_vop_close(struct vop_close_args
*ap
)
1447 return vop_stdclose(ap
);
1451 * hammer2_vop_nlink { nch, dvp, vp, cred }
1453 * Create a hardlink from (vp) to {dvp, nch}.
1457 hammer2_vop_nlink(struct vop_nlink_args
*ap
)
1459 hammer2_inode_t
*tdip
; /* target directory to create link in */
1460 hammer2_inode_t
*ip
; /* inode we are hardlinking to */
1461 struct namecache
*ncp
;
1462 const uint8_t *name
;
1466 if (ap
->a_dvp
->v_mount
!= ap
->a_vp
->v_mount
)
1469 tdip
= VTOI(ap
->a_dvp
);
1470 if (tdip
->pmp
->ronly
)
1472 if (hammer2_vfs_enospace(tdip
, 0, ap
->a_cred
) > 1)
1475 ncp
= ap
->a_nch
->ncp
;
1476 name
= ncp
->nc_name
;
1477 name_len
= ncp
->nc_nlen
;
1480 * ip represents the file being hardlinked. The file could be a
1481 * normal file or a hardlink target if it has already been hardlinked.
1482 * (with the new semantics, it will almost always be a hardlink
1485 * Bump nlinks and potentially also create or move the hardlink
1486 * target in the parent directory common to (ip) and (tdip). The
1487 * consolidation code can modify ip->cluster. The returned cluster
1490 ip
= VTOI(ap
->a_vp
);
1491 KASSERT(ip
->pmp
, ("ip->pmp is NULL %p %p", ip
, ip
->pmp
));
1492 hammer2_pfs_memory_wait(ip
->pmp
);
1493 hammer2_trans_init(ip
->pmp
, 0);
1496 * Target should be an indexed inode or there's no way we will ever
1497 * be able to find it!
1499 KKASSERT((ip
->meta
.name_key
& HAMMER2_DIRHASH_VISIBLE
) == 0);
1504 * Can return NULL and error == EXDEV if the common parent
1505 * crosses a directory with the xlink flag set.
1507 hammer2_inode_lock(tdip
, 0);
1508 hammer2_inode_lock(ip
, 0);
1511 * Create the directory entry and bump nlinks.
1514 error
= hammer2_dirent_create(tdip
, name
, name_len
,
1515 ip
->meta
.inum
, ip
->meta
.type
);
1516 hammer2_inode_modify(ip
);
1521 * Update dip's mtime
1525 hammer2_update_time(&mtime
);
1526 hammer2_inode_modify(tdip
);
1527 tdip
->meta
.mtime
= mtime
;
1529 cache_setunresolved(ap
->a_nch
);
1530 cache_setvp(ap
->a_nch
, ap
->a_vp
);
1532 hammer2_inode_unlock(ip
);
1533 hammer2_inode_unlock(tdip
);
1535 hammer2_trans_done(ip
->pmp
);
1536 hammer2_knote(ap
->a_vp
, NOTE_LINK
);
1537 hammer2_knote(ap
->a_dvp
, NOTE_WRITE
);
1543 * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1545 * The operating system has already ensured that the directory entry
1546 * does not exist and done all appropriate namespace locking.
1550 hammer2_vop_ncreate(struct vop_ncreate_args
*ap
)
1552 hammer2_inode_t
*dip
;
1553 hammer2_inode_t
*nip
;
1554 struct namecache
*ncp
;
1555 const uint8_t *name
;
1560 dip
= VTOI(ap
->a_dvp
);
1561 if (dip
->pmp
->ronly
)
1563 if (hammer2_vfs_enospace(dip
, 0, ap
->a_cred
) > 1)
1566 ncp
= ap
->a_nch
->ncp
;
1567 name
= ncp
->nc_name
;
1568 name_len
= ncp
->nc_nlen
;
1569 hammer2_pfs_memory_wait(dip
->pmp
);
1570 hammer2_trans_init(dip
->pmp
, 0);
1572 inum
= hammer2_trans_newinum(dip
->pmp
);
1575 * Create the actual inode as a hidden file in the iroot, then
1576 * create the directory entry. The creation of the actual inode
1577 * sets its nlinks to 1 which is the value we desire.
1579 nip
= hammer2_inode_create(dip
->pmp
->iroot
, dip
, ap
->a_vap
, ap
->a_cred
,
1585 error
= hammer2_error_to_errno(error
);
1587 error
= hammer2_dirent_create(dip
, name
, name_len
,
1588 nip
->meta
.inum
, nip
->meta
.type
);
1592 hammer2_inode_unlink_finisher(nip
, 0);
1593 hammer2_inode_unlock(nip
);
1598 *ap
->a_vpp
= hammer2_igetv(nip
, &error
);
1599 hammer2_inode_unlock(nip
);
1603 * Update dip's mtime
1608 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1609 hammer2_update_time(&mtime
);
1610 hammer2_inode_modify(dip
);
1611 dip
->meta
.mtime
= mtime
;
1612 hammer2_inode_unlock(dip
);
1615 hammer2_trans_done(dip
->pmp
);
1618 cache_setunresolved(ap
->a_nch
);
1619 cache_setvp(ap
->a_nch
, *ap
->a_vpp
);
1620 hammer2_knote(ap
->a_dvp
, NOTE_WRITE
);
1626 * Make a device node (typically a fifo)
1630 hammer2_vop_nmknod(struct vop_nmknod_args
*ap
)
1632 hammer2_inode_t
*dip
;
1633 hammer2_inode_t
*nip
;
1634 struct namecache
*ncp
;
1635 const uint8_t *name
;
1640 dip
= VTOI(ap
->a_dvp
);
1641 if (dip
->pmp
->ronly
)
1643 if (hammer2_vfs_enospace(dip
, 0, ap
->a_cred
) > 1)
1646 ncp
= ap
->a_nch
->ncp
;
1647 name
= ncp
->nc_name
;
1648 name_len
= ncp
->nc_nlen
;
1649 hammer2_pfs_memory_wait(dip
->pmp
);
1650 hammer2_trans_init(dip
->pmp
, 0);
1653 * Create the device inode and then create the directory entry.
1655 inum
= hammer2_trans_newinum(dip
->pmp
);
1656 nip
= hammer2_inode_create(dip
->pmp
->iroot
, dip
, ap
->a_vap
, ap
->a_cred
,
1661 error
= hammer2_dirent_create(dip
, name
, name_len
,
1662 nip
->meta
.inum
, nip
->meta
.type
);
1666 hammer2_inode_unlink_finisher(nip
, 0);
1667 hammer2_inode_unlock(nip
);
1672 *ap
->a_vpp
= hammer2_igetv(nip
, &error
);
1673 hammer2_inode_unlock(nip
);
1677 * Update dip's mtime
1682 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1683 hammer2_update_time(&mtime
);
1684 hammer2_inode_modify(dip
);
1685 dip
->meta
.mtime
= mtime
;
1686 hammer2_inode_unlock(dip
);
1689 hammer2_trans_done(dip
->pmp
);
1692 cache_setunresolved(ap
->a_nch
);
1693 cache_setvp(ap
->a_nch
, *ap
->a_vpp
);
1694 hammer2_knote(ap
->a_dvp
, NOTE_WRITE
);
1700 * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1704 hammer2_vop_nsymlink(struct vop_nsymlink_args
*ap
)
1706 hammer2_inode_t
*dip
;
1707 hammer2_inode_t
*nip
;
1708 struct namecache
*ncp
;
1709 const uint8_t *name
;
1714 dip
= VTOI(ap
->a_dvp
);
1715 if (dip
->pmp
->ronly
)
1717 if (hammer2_vfs_enospace(dip
, 0, ap
->a_cred
) > 1)
1720 ncp
= ap
->a_nch
->ncp
;
1721 name
= ncp
->nc_name
;
1722 name_len
= ncp
->nc_nlen
;
1723 hammer2_pfs_memory_wait(dip
->pmp
);
1724 hammer2_trans_init(dip
->pmp
, 0);
1726 ap
->a_vap
->va_type
= VLNK
; /* enforce type */
1729 * Create the softlink as an inode and then create the directory
1732 inum
= hammer2_trans_newinum(dip
->pmp
);
1734 nip
= hammer2_inode_create(dip
->pmp
->iroot
, dip
, ap
->a_vap
, ap
->a_cred
,
1739 error
= hammer2_dirent_create(dip
, name
, name_len
,
1740 nip
->meta
.inum
, nip
->meta
.type
);
1744 hammer2_inode_unlink_finisher(nip
, 0);
1745 hammer2_inode_unlock(nip
);
1749 hammer2_trans_done(dip
->pmp
);
1752 *ap
->a_vpp
= hammer2_igetv(nip
, &error
);
1755 * Build the softlink (~like file data) and finalize the namecache.
1762 bytes
= strlen(ap
->a_target
);
1764 hammer2_inode_unlock(nip
);
1765 bzero(&auio
, sizeof(auio
));
1766 bzero(&aiov
, sizeof(aiov
));
1767 auio
.uio_iov
= &aiov
;
1768 auio
.uio_segflg
= UIO_SYSSPACE
;
1769 auio
.uio_rw
= UIO_WRITE
;
1770 auio
.uio_resid
= bytes
;
1771 auio
.uio_iovcnt
= 1;
1772 auio
.uio_td
= curthread
;
1773 aiov
.iov_base
= ap
->a_target
;
1774 aiov
.iov_len
= bytes
;
1775 error
= hammer2_write_file(nip
, &auio
, IO_APPEND
, 0);
1776 /* XXX handle error */
1779 hammer2_inode_unlock(nip
);
1783 * Update dip's mtime
1788 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1789 hammer2_update_time(&mtime
);
1790 hammer2_inode_modify(dip
);
1791 dip
->meta
.mtime
= mtime
;
1792 hammer2_inode_unlock(dip
);
1795 hammer2_trans_done(dip
->pmp
);
1798 * Finalize namecache
1801 cache_setunresolved(ap
->a_nch
);
1802 cache_setvp(ap
->a_nch
, *ap
->a_vpp
);
1803 hammer2_knote(ap
->a_dvp
, NOTE_WRITE
);
1809 * hammer2_vop_nremove { nch, dvp, cred }
1813 hammer2_vop_nremove(struct vop_nremove_args
*ap
)
1815 hammer2_xop_unlink_t
*xop
;
1816 hammer2_inode_t
*dip
;
1817 hammer2_inode_t
*ip
;
1818 struct namecache
*ncp
;
1822 dip
= VTOI(ap
->a_dvp
);
1823 if (dip
->pmp
->ronly
)
1826 /* allow removals, except user to also bulkfree */
1827 if (hammer2_vfs_enospace(dip
, 0, ap
->a_cred
) > 1)
1831 ncp
= ap
->a_nch
->ncp
;
1833 hammer2_pfs_memory_wait(dip
->pmp
);
1834 hammer2_trans_init(dip
->pmp
, 0);
1835 hammer2_inode_lock(dip
, 0);
1838 * The unlink XOP unlinks the path from the directory and
1839 * locates and returns the cluster associated with the real inode.
1840 * We have to handle nlinks here on the frontend.
1842 xop
= hammer2_xop_alloc(dip
, HAMMER2_XOP_MODIFYING
);
1843 hammer2_xop_setname(&xop
->head
, ncp
->nc_name
, ncp
->nc_nlen
);
1846 * The namecache entry is locked so nobody can use this namespace.
1847 * Calculate isopen to determine if this namespace has an open vp
1848 * associated with it and resolve the vp only if it does.
1850 * We try to avoid resolving the vnode if nobody has it open, but
1851 * note that the test is via this namespace only.
1853 isopen
= cache_isopen(ap
->a_nch
);
1855 xop
->dopermanent
= 0;
1856 hammer2_xop_start(&xop
->head
, hammer2_xop_unlink
);
1859 * Collect the real inode and adjust nlinks, destroy the real
1860 * inode if nlinks transitions to 0 and it was the real inode
1861 * (else it has already been removed).
1863 error
= hammer2_xop_collect(&xop
->head
, 0);
1864 error
= hammer2_error_to_errno(error
);
1865 hammer2_inode_unlock(dip
);
1868 ip
= hammer2_inode_get(dip
->pmp
, dip
, &xop
->head
.cluster
, -1);
1869 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1871 hammer2_inode_unlink_finisher(ip
, isopen
);
1872 hammer2_inode_unlock(ip
);
1875 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1879 * Update dip's mtime
1884 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1885 hammer2_update_time(&mtime
);
1886 hammer2_inode_modify(dip
);
1887 dip
->meta
.mtime
= mtime
;
1888 hammer2_inode_unlock(dip
);
1891 hammer2_inode_run_sideq(dip
->pmp
, 0);
1892 hammer2_trans_done(dip
->pmp
);
1894 cache_unlink(ap
->a_nch
);
1895 hammer2_knote(ap
->a_dvp
, NOTE_WRITE
);
1901 * hammer2_vop_nrmdir { nch, dvp, cred }
1905 hammer2_vop_nrmdir(struct vop_nrmdir_args
*ap
)
1907 hammer2_xop_unlink_t
*xop
;
1908 hammer2_inode_t
*dip
;
1909 hammer2_inode_t
*ip
;
1910 struct namecache
*ncp
;
1914 dip
= VTOI(ap
->a_dvp
);
1915 if (dip
->pmp
->ronly
)
1918 /* allow removals, except user to also bulkfree */
1919 if (hammer2_vfs_enospace(dip
, 0, ap
->a_cred
) > 1)
1923 hammer2_pfs_memory_wait(dip
->pmp
);
1924 hammer2_trans_init(dip
->pmp
, 0);
1925 hammer2_inode_lock(dip
, 0);
1927 xop
= hammer2_xop_alloc(dip
, HAMMER2_XOP_MODIFYING
);
1929 ncp
= ap
->a_nch
->ncp
;
1930 hammer2_xop_setname(&xop
->head
, ncp
->nc_name
, ncp
->nc_nlen
);
1931 isopen
= cache_isopen(ap
->a_nch
);
1933 xop
->dopermanent
= 0;
1934 hammer2_xop_start(&xop
->head
, hammer2_xop_unlink
);
1937 * Collect the real inode and adjust nlinks, destroy the real
1938 * inode if nlinks transitions to 0 and it was the real inode
1939 * (else it has already been removed).
1941 error
= hammer2_xop_collect(&xop
->head
, 0);
1942 error
= hammer2_error_to_errno(error
);
1943 hammer2_inode_unlock(dip
);
1946 ip
= hammer2_inode_get(dip
->pmp
, dip
, &xop
->head
.cluster
, -1);
1947 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1949 hammer2_inode_unlink_finisher(ip
, isopen
);
1950 hammer2_inode_unlock(ip
);
1953 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1957 * Update dip's mtime
1962 hammer2_inode_lock(dip
, HAMMER2_RESOLVE_SHARED
);
1963 hammer2_update_time(&mtime
);
1964 hammer2_inode_modify(dip
);
1965 dip
->meta
.mtime
= mtime
;
1966 hammer2_inode_unlock(dip
);
1969 hammer2_inode_run_sideq(dip
->pmp
, 0);
1970 hammer2_trans_done(dip
->pmp
);
1972 cache_unlink(ap
->a_nch
);
1973 hammer2_knote(ap
->a_dvp
, NOTE_WRITE
| NOTE_LINK
);
1979 * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1983 hammer2_vop_nrename(struct vop_nrename_args
*ap
)
1985 struct namecache
*fncp
;
1986 struct namecache
*tncp
;
1987 hammer2_inode_t
*fdip
; /* source directory */
1988 hammer2_inode_t
*tdip
; /* target directory */
1989 hammer2_inode_t
*ip
; /* file being renamed */
1990 hammer2_inode_t
*tip
; /* replaced target during rename or NULL */
1991 const uint8_t *fname
;
1993 const uint8_t *tname
;
2000 if (ap
->a_fdvp
->v_mount
!= ap
->a_tdvp
->v_mount
)
2002 if (ap
->a_fdvp
->v_mount
!= ap
->a_fnch
->ncp
->nc_vp
->v_mount
)
2005 fdip
= VTOI(ap
->a_fdvp
); /* source directory */
2006 tdip
= VTOI(ap
->a_tdvp
); /* target directory */
2008 if (fdip
->pmp
->ronly
)
2010 if (hammer2_vfs_enospace(fdip
, 0, ap
->a_cred
) > 1)
2013 fncp
= ap
->a_fnch
->ncp
; /* entry name in source */
2014 fname
= fncp
->nc_name
;
2015 fname_len
= fncp
->nc_nlen
;
2017 tncp
= ap
->a_tnch
->ncp
; /* entry name in target */
2018 tname
= tncp
->nc_name
;
2019 tname_len
= tncp
->nc_nlen
;
2021 hammer2_pfs_memory_wait(tdip
->pmp
);
2022 hammer2_trans_init(tdip
->pmp
, 0);
2027 ip
= VTOI(fncp
->nc_vp
);
2028 hammer2_inode_ref(ip
); /* extra ref */
2031 * Lookup the target name to determine if a directory entry
2032 * is being overwritten. We only hold related inode locks
2033 * temporarily, the operating system is expected to protect
2034 * against rename races.
2036 tip
= tncp
->nc_vp
? VTOI(tncp
->nc_vp
) : NULL
;
2038 hammer2_inode_ref(tip
); /* extra ref */
2041 * Can return NULL and error == EXDEV if the common parent
2042 * crosses a directory with the xlink flag set.
2044 * For now try to avoid deadlocks with a simple pointer address
2045 * test. (tip) can be NULL.
2049 hammer2_inode_lock(fdip
, 0);
2050 hammer2_inode_lock(tdip
, 0);
2052 hammer2_inode_lock(tdip
, 0);
2053 hammer2_inode_lock(fdip
, 0);
2057 hammer2_inode_lock(ip
, 0);
2058 hammer2_inode_lock(tip
, 0);
2060 hammer2_inode_lock(tip
, 0);
2061 hammer2_inode_lock(ip
, 0);
2064 hammer2_inode_lock(ip
, 0);
2069 * Delete the target namespace.
2071 * REMOVED - NOW FOLDED INTO XOP_NRENAME OPERATION
2074 hammer2_xop_unlink_t
*xop2
;
2075 hammer2_inode_t
*tip
;
2079 * The unlink XOP unlinks the path from the directory and
2080 * locates and returns the cluster associated with the real
2081 * inode. We have to handle nlinks here on the frontend.
2083 xop2
= hammer2_xop_alloc(tdip
, HAMMER2_XOP_MODIFYING
);
2084 hammer2_xop_setname(&xop2
->head
, tname
, tname_len
);
2085 isopen
= cache_isopen(ap
->a_tnch
);
2087 xop2
->dopermanent
= 0;
2088 hammer2_xop_start(&xop2
->head
, hammer2_xop_unlink
);
2091 * Collect the real inode and adjust nlinks, destroy the real
2092 * inode if nlinks transitions to 0 and it was the real inode
2093 * (else it has already been removed).
2095 tnch_error
= hammer2_xop_collect(&xop2
->head
, 0);
2096 tnch_error
= hammer2_error_to_errno(tnch_error
);
2097 /* hammer2_inode_unlock(tdip); */
2099 if (tnch_error
== 0) {
2100 tip
= hammer2_inode_get(tdip
->pmp
, NULL
,
2101 &xop2
->head
.cluster
, -1);
2102 hammer2_xop_retire(&xop2
->head
, HAMMER2_XOPMASK_VOP
);
2104 hammer2_inode_unlink_finisher(tip
, isopen
);
2105 hammer2_inode_unlock(tip
);
2108 hammer2_xop_retire(&xop2
->head
, HAMMER2_XOPMASK_VOP
);
2110 /* hammer2_inode_lock(tdip, 0); */
2112 if (tnch_error
&& tnch_error
!= ENOENT
) {
2121 * Resolve the collision space for (tdip, tname, tname_len)
2123 * tdip must be held exclusively locked to prevent races since
2124 * multiple filenames can end up in the same collision space.
2127 hammer2_xop_scanlhc_t
*sxop
;
2128 hammer2_tid_t lhcbase
;
2130 tlhc
= hammer2_dirhash(tname
, tname_len
);
2132 sxop
= hammer2_xop_alloc(tdip
, HAMMER2_XOP_MODIFYING
);
2134 hammer2_xop_start(&sxop
->head
, hammer2_xop_scanlhc
);
2135 while ((error
= hammer2_xop_collect(&sxop
->head
, 0)) == 0) {
2136 if (tlhc
!= sxop
->head
.cluster
.focus
->bref
.key
)
2140 error
= hammer2_error_to_errno(error
);
2141 hammer2_xop_retire(&sxop
->head
, HAMMER2_XOPMASK_VOP
);
2144 if (error
!= ENOENT
)
2149 if ((lhcbase
^ tlhc
) & ~HAMMER2_DIRHASH_LOMASK
) {
2156 * Ready to go, issue the rename to the backend. Note that meta-data
2157 * updates to the related inodes occur separately from the rename
2160 * NOTE: While it is not necessary to update ip->meta.name*, doing
2161 * so aids catastrophic recovery and debugging.
2164 hammer2_xop_nrename_t
*xop4
;
2166 xop4
= hammer2_xop_alloc(fdip
, HAMMER2_XOP_MODIFYING
);
2168 xop4
->ip_key
= ip
->meta
.name_key
;
2169 hammer2_xop_setip2(&xop4
->head
, ip
);
2170 hammer2_xop_setip3(&xop4
->head
, tdip
);
2171 hammer2_xop_setname(&xop4
->head
, fname
, fname_len
);
2172 hammer2_xop_setname2(&xop4
->head
, tname
, tname_len
);
2173 hammer2_xop_start(&xop4
->head
, hammer2_xop_nrename
);
2175 error
= hammer2_xop_collect(&xop4
->head
, 0);
2176 error
= hammer2_error_to_errno(error
);
2177 hammer2_xop_retire(&xop4
->head
, HAMMER2_XOPMASK_VOP
);
2179 if (error
== ENOENT
)
2183 * Update inode meta-data.
2185 * WARNING! The in-memory inode (ip) structure does not
2186 * maintain a copy of the inode's filename buffer.
2189 (ip
->meta
.name_key
& HAMMER2_DIRHASH_VISIBLE
)) {
2190 hammer2_inode_modify(ip
);
2191 ip
->meta
.name_len
= tname_len
;
2192 ip
->meta
.name_key
= tlhc
;
2195 hammer2_inode_modify(ip
);
2196 ip
->meta
.iparent
= tdip
->meta
.inum
;
2204 * If no error, the backend has replaced the target directory entry.
2205 * We must adjust nlinks on the original replace target if it exists.
2207 if (error
== 0 && tip
) {
2210 isopen
= cache_isopen(ap
->a_tnch
);
2211 hammer2_inode_unlink_finisher(tip
, isopen
);
2215 * Update directory mtimes to represent the something changed.
2217 if (update_fdip
|| update_tdip
) {
2220 hammer2_update_time(&mtime
);
2222 hammer2_inode_modify(fdip
);
2223 fdip
->meta
.mtime
= mtime
;
2226 hammer2_inode_modify(tdip
);
2227 tdip
->meta
.mtime
= mtime
;
2231 hammer2_inode_unlock(tip
);
2232 hammer2_inode_drop(tip
);
2234 hammer2_inode_unlock(ip
);
2235 hammer2_inode_unlock(tdip
);
2236 hammer2_inode_unlock(fdip
);
2237 hammer2_inode_drop(ip
);
2238 hammer2_inode_run_sideq(fdip
->pmp
, 0);
2240 hammer2_trans_done(tdip
->pmp
);
2243 * Issue the namecache update after unlocking all the internal
2244 * hammer2 structures, otherwise we might deadlock.
2246 * WARNING! The target namespace must be updated atomically,
2247 * and we depend on cache_rename() to handle that for
2248 * us. Do not do a separate cache_unlink() because
2249 * that leaves a small window of opportunity for other
2250 * threads to allocate the target namespace before we
2251 * manage to complete our rename.
2253 * WARNING! cache_rename() (and cache_unlink()) will properly
2254 * set VREF_FINALIZE on any attached vnode. Do not
2255 * call cache_setunresolved() manually before-hand as
2256 * this will prevent the flag from being set later via
2257 * cache_rename(). If VREF_FINALIZE is not properly set
2258 * and the inode is no longer in the topology, related
2259 * chains can remain dirty indefinitely.
2261 if (error
== 0 && tip
) {
2262 /*cache_unlink(ap->a_tnch); see above */
2263 /*cache_setunresolved(ap->a_tnch); see above */
2266 cache_rename(ap
->a_fnch
, ap
->a_tnch
);
2267 hammer2_knote(ap
->a_fdvp
, NOTE_WRITE
);
2268 hammer2_knote(ap
->a_tdvp
, NOTE_WRITE
);
2269 hammer2_knote(fncp
->nc_vp
, NOTE_RENAME
);
2276 * hammer2_vop_ioctl { vp, command, data, fflag, cred }
2280 hammer2_vop_ioctl(struct vop_ioctl_args
*ap
)
2282 hammer2_inode_t
*ip
;
2285 ip
= VTOI(ap
->a_vp
);
2287 error
= hammer2_ioctl(ip
, ap
->a_command
, (void *)ap
->a_data
,
2288 ap
->a_fflag
, ap
->a_cred
);
2294 hammer2_vop_mountctl(struct vop_mountctl_args
*ap
)
2301 case (MOUNTCTL_SET_EXPORT
):
2302 mp
= ap
->a_head
.a_ops
->head
.vv_mount
;
2305 if (ap
->a_ctllen
!= sizeof(struct export_args
))
2308 rc
= vfs_export(mp
, &pmp
->export
,
2309 (const struct export_args
*)ap
->a_ctl
);
2312 rc
= vop_stdmountctl(ap
);
2321 static void filt_hammer2detach(struct knote
*kn
);
2322 static int filt_hammer2read(struct knote
*kn
, long hint
);
2323 static int filt_hammer2write(struct knote
*kn
, long hint
);
2324 static int filt_hammer2vnode(struct knote
*kn
, long hint
);
2326 static struct filterops hammer2read_filtops
=
2327 { FILTEROP_ISFD
| FILTEROP_MPSAFE
,
2328 NULL
, filt_hammer2detach
, filt_hammer2read
};
2329 static struct filterops hammer2write_filtops
=
2330 { FILTEROP_ISFD
| FILTEROP_MPSAFE
,
2331 NULL
, filt_hammer2detach
, filt_hammer2write
};
2332 static struct filterops hammer2vnode_filtops
=
2333 { FILTEROP_ISFD
| FILTEROP_MPSAFE
,
2334 NULL
, filt_hammer2detach
, filt_hammer2vnode
};
2338 hammer2_vop_kqfilter(struct vop_kqfilter_args
*ap
)
2340 struct vnode
*vp
= ap
->a_vp
;
2341 struct knote
*kn
= ap
->a_kn
;
2343 switch (kn
->kn_filter
) {
2345 kn
->kn_fop
= &hammer2read_filtops
;
2348 kn
->kn_fop
= &hammer2write_filtops
;
2351 kn
->kn_fop
= &hammer2vnode_filtops
;
2354 return (EOPNOTSUPP
);
2357 kn
->kn_hook
= (caddr_t
)vp
;
2359 knote_insert(&vp
->v_pollinfo
.vpi_kqinfo
.ki_note
, kn
);
2365 filt_hammer2detach(struct knote
*kn
)
2367 struct vnode
*vp
= (void *)kn
->kn_hook
;
2369 knote_remove(&vp
->v_pollinfo
.vpi_kqinfo
.ki_note
, kn
);
2373 filt_hammer2read(struct knote
*kn
, long hint
)
2375 struct vnode
*vp
= (void *)kn
->kn_hook
;
2376 hammer2_inode_t
*ip
= VTOI(vp
);
2379 if (hint
== NOTE_REVOKE
) {
2380 kn
->kn_flags
|= (EV_EOF
| EV_NODATA
| EV_ONESHOT
);
2383 off
= ip
->meta
.size
- kn
->kn_fp
->f_offset
;
2384 kn
->kn_data
= (off
< INTPTR_MAX
) ? off
: INTPTR_MAX
;
2385 if (kn
->kn_sfflags
& NOTE_OLDAPI
)
2387 return (kn
->kn_data
!= 0);
2392 filt_hammer2write(struct knote
*kn
, long hint
)
2394 if (hint
== NOTE_REVOKE
)
2395 kn
->kn_flags
|= (EV_EOF
| EV_NODATA
| EV_ONESHOT
);
2401 filt_hammer2vnode(struct knote
*kn
, long hint
)
2403 if (kn
->kn_sfflags
& hint
)
2404 kn
->kn_fflags
|= hint
;
2405 if (hint
== NOTE_REVOKE
) {
2406 kn
->kn_flags
|= (EV_EOF
| EV_NODATA
);
2409 return (kn
->kn_fflags
!= 0);
2417 hammer2_vop_markatime(struct vop_markatime_args
*ap
)
2419 hammer2_inode_t
*ip
;
2432 hammer2_vop_fifokqfilter(struct vop_kqfilter_args
*ap
)
2436 error
= VOCALL(&fifo_vnode_vops
, &ap
->a_head
);
2438 error
= hammer2_vop_kqfilter(ap
);
2445 struct vop_ops hammer2_vnode_vops
= {
2446 .vop_default
= vop_defaultop
,
2447 .vop_fsync
= hammer2_vop_fsync
,
2448 .vop_getpages
= vop_stdgetpages
,
2449 .vop_putpages
= vop_stdputpages
,
2450 .vop_access
= hammer2_vop_access
,
2451 .vop_advlock
= hammer2_vop_advlock
,
2452 .vop_close
= hammer2_vop_close
,
2453 .vop_nlink
= hammer2_vop_nlink
,
2454 .vop_ncreate
= hammer2_vop_ncreate
,
2455 .vop_nsymlink
= hammer2_vop_nsymlink
,
2456 .vop_nremove
= hammer2_vop_nremove
,
2457 .vop_nrmdir
= hammer2_vop_nrmdir
,
2458 .vop_nrename
= hammer2_vop_nrename
,
2459 .vop_getattr
= hammer2_vop_getattr
,
2460 .vop_setattr
= hammer2_vop_setattr
,
2461 .vop_readdir
= hammer2_vop_readdir
,
2462 .vop_readlink
= hammer2_vop_readlink
,
2463 .vop_read
= hammer2_vop_read
,
2464 .vop_write
= hammer2_vop_write
,
2465 .vop_open
= hammer2_vop_open
,
2466 .vop_inactive
= hammer2_vop_inactive
,
2467 .vop_reclaim
= hammer2_vop_reclaim
,
2468 .vop_nresolve
= hammer2_vop_nresolve
,
2469 .vop_nlookupdotdot
= hammer2_vop_nlookupdotdot
,
2470 .vop_nmkdir
= hammer2_vop_nmkdir
,
2471 .vop_nmknod
= hammer2_vop_nmknod
,
2472 .vop_ioctl
= hammer2_vop_ioctl
,
2473 .vop_mountctl
= hammer2_vop_mountctl
,
2474 .vop_bmap
= hammer2_vop_bmap
,
2475 .vop_strategy
= hammer2_vop_strategy
,
2476 .vop_kqfilter
= hammer2_vop_kqfilter
2479 struct vop_ops hammer2_spec_vops
= {
2480 .vop_default
= vop_defaultop
,
2481 .vop_fsync
= hammer2_vop_fsync
,
2482 .vop_read
= vop_stdnoread
,
2483 .vop_write
= vop_stdnowrite
,
2484 .vop_access
= hammer2_vop_access
,
2485 .vop_close
= hammer2_vop_close
,
2486 .vop_markatime
= hammer2_vop_markatime
,
2487 .vop_getattr
= hammer2_vop_getattr
,
2488 .vop_inactive
= hammer2_vop_inactive
,
2489 .vop_reclaim
= hammer2_vop_reclaim
,
2490 .vop_setattr
= hammer2_vop_setattr
2493 struct vop_ops hammer2_fifo_vops
= {
2494 .vop_default
= fifo_vnoperate
,
2495 .vop_fsync
= hammer2_vop_fsync
,
2497 .vop_read
= hammer2_vop_fiforead
,
2498 .vop_write
= hammer2_vop_fifowrite
,
2500 .vop_access
= hammer2_vop_access
,
2502 .vop_close
= hammer2_vop_fifoclose
,
2504 .vop_markatime
= hammer2_vop_markatime
,
2505 .vop_getattr
= hammer2_vop_getattr
,
2506 .vop_inactive
= hammer2_vop_inactive
,
2507 .vop_reclaim
= hammer2_vop_reclaim
,
2508 .vop_setattr
= hammer2_vop_setattr
,
2509 .vop_kqfilter
= hammer2_vop_fifokqfilter