2 * Copyright (c) 2007-2008 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * $DragonFly: src/sys/vfs/hammer/hammer_vnops.c,v 1.85 2008/07/10 04:44:33 dillon Exp $
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
40 #include <sys/fcntl.h>
41 #include <sys/namecache.h>
42 #include <sys/vnode.h>
43 #include <sys/lockf.h>
44 #include <sys/event.h>
46 #include <sys/dirent.h>
47 #include <vm/vm_extern.h>
48 #include <vfs/fifofs/fifo.h>
54 /*static int hammer_vop_vnoperate(struct vop_generic_args *);*/
55 static int hammer_vop_fsync(struct vop_fsync_args *);
56 static int hammer_vop_read(struct vop_read_args *);
57 static int hammer_vop_write(struct vop_write_args *);
58 static int hammer_vop_access(struct vop_access_args *);
59 static int hammer_vop_advlock(struct vop_advlock_args *);
60 static int hammer_vop_close(struct vop_close_args *);
61 static int hammer_vop_ncreate(struct vop_ncreate_args *);
62 static int hammer_vop_getattr(struct vop_getattr_args *);
63 static int hammer_vop_nresolve(struct vop_nresolve_args *);
64 static int hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
65 static int hammer_vop_nlink(struct vop_nlink_args *);
66 static int hammer_vop_nmkdir(struct vop_nmkdir_args *);
67 static int hammer_vop_nmknod(struct vop_nmknod_args *);
68 static int hammer_vop_open(struct vop_open_args *);
69 static int hammer_vop_pathconf(struct vop_pathconf_args *);
70 static int hammer_vop_print(struct vop_print_args *);
71 static int hammer_vop_readdir(struct vop_readdir_args *);
72 static int hammer_vop_readlink(struct vop_readlink_args *);
73 static int hammer_vop_nremove(struct vop_nremove_args *);
74 static int hammer_vop_nrename(struct vop_nrename_args *);
75 static int hammer_vop_nrmdir(struct vop_nrmdir_args *);
76 static int hammer_vop_setattr(struct vop_setattr_args *);
77 static int hammer_vop_strategy(struct vop_strategy_args *);
78 static int hammer_vop_bmap(struct vop_bmap_args *ap);
79 static int hammer_vop_nsymlink(struct vop_nsymlink_args *);
80 static int hammer_vop_nwhiteout(struct vop_nwhiteout_args *);
81 static int hammer_vop_ioctl(struct vop_ioctl_args *);
82 static int hammer_vop_mountctl(struct vop_mountctl_args *);
84 static int hammer_vop_fifoclose (struct vop_close_args *);
85 static int hammer_vop_fiforead (struct vop_read_args *);
86 static int hammer_vop_fifowrite (struct vop_write_args *);
88 static int hammer_vop_specclose (struct vop_close_args *);
89 static int hammer_vop_specread (struct vop_read_args *);
90 static int hammer_vop_specwrite (struct vop_write_args *);
92 struct vop_ops hammer_vnode_vops = {
93 .vop_default = vop_defaultop,
94 .vop_fsync = hammer_vop_fsync,
95 .vop_getpages = vop_stdgetpages,
96 .vop_putpages = vop_stdputpages,
97 .vop_read = hammer_vop_read,
98 .vop_write = hammer_vop_write,
99 .vop_access = hammer_vop_access,
100 .vop_advlock = hammer_vop_advlock,
101 .vop_close = hammer_vop_close,
102 .vop_ncreate = hammer_vop_ncreate,
103 .vop_getattr = hammer_vop_getattr,
104 .vop_inactive = hammer_vop_inactive,
105 .vop_reclaim = hammer_vop_reclaim,
106 .vop_nresolve = hammer_vop_nresolve,
107 .vop_nlookupdotdot = hammer_vop_nlookupdotdot,
108 .vop_nlink = hammer_vop_nlink,
109 .vop_nmkdir = hammer_vop_nmkdir,
110 .vop_nmknod = hammer_vop_nmknod,
111 .vop_open = hammer_vop_open,
112 .vop_pathconf = hammer_vop_pathconf,
113 .vop_print = hammer_vop_print,
114 .vop_readdir = hammer_vop_readdir,
115 .vop_readlink = hammer_vop_readlink,
116 .vop_nremove = hammer_vop_nremove,
117 .vop_nrename = hammer_vop_nrename,
118 .vop_nrmdir = hammer_vop_nrmdir,
119 .vop_setattr = hammer_vop_setattr,
120 .vop_bmap = hammer_vop_bmap,
121 .vop_strategy = hammer_vop_strategy,
122 .vop_nsymlink = hammer_vop_nsymlink,
123 .vop_nwhiteout = hammer_vop_nwhiteout,
124 .vop_ioctl = hammer_vop_ioctl,
125 .vop_mountctl = hammer_vop_mountctl
128 struct vop_ops hammer_spec_vops = {
129 .vop_default = spec_vnoperate,
130 .vop_fsync = hammer_vop_fsync,
131 .vop_read = hammer_vop_specread,
132 .vop_write = hammer_vop_specwrite,
133 .vop_access = hammer_vop_access,
134 .vop_close = hammer_vop_specclose,
135 .vop_getattr = hammer_vop_getattr,
136 .vop_inactive = hammer_vop_inactive,
137 .vop_reclaim = hammer_vop_reclaim,
138 .vop_setattr = hammer_vop_setattr
141 struct vop_ops hammer_fifo_vops = {
142 .vop_default = fifo_vnoperate,
143 .vop_fsync = hammer_vop_fsync,
144 .vop_read = hammer_vop_fiforead,
145 .vop_write = hammer_vop_fifowrite,
146 .vop_access = hammer_vop_access,
147 .vop_close = hammer_vop_fifoclose,
148 .vop_getattr = hammer_vop_getattr,
149 .vop_inactive = hammer_vop_inactive,
150 .vop_reclaim = hammer_vop_reclaim,
151 .vop_setattr = hammer_vop_setattr
154 #ifdef DEBUG_TRUNCATE
155 struct hammer_inode *HammerTruncIp;
158 static int hammer_dounlink(hammer_transaction_t trans, struct nchandle *nch,
159 struct vnode *dvp, struct ucred *cred, int flags);
160 static int hammer_vop_strategy_read(struct vop_strategy_args *ap);
161 static int hammer_vop_strategy_write(struct vop_strategy_args *ap);
166 hammer_vop_vnoperate(struct vop_generic_args *)
168 return (VOCALL(&hammer_vnode_vops, ap));
173 * hammer_vop_fsync { vp, waitfor }
175 * fsync() an inode to disk and wait for it to be completely committed
176 * such that the information would not be undone if a crash occured after
181 hammer_vop_fsync(struct vop_fsync_args *ap)
183 hammer_inode_t ip = VTOI(ap->a_vp);
185 vfsync(ap->a_vp, ap->a_waitfor, 1, NULL, NULL);
186 hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
187 if (ap->a_waitfor == MNT_WAIT)
188 hammer_wait_inode(ip);
193 * hammer_vop_read { vp, uio, ioflag, cred }
197 hammer_vop_read(struct vop_read_args *ap)
199 struct hammer_transaction trans;
210 if (ap->a_vp->v_type != VREG)
217 * Allow the UIO's size to override the sequential heuristic.
219 blksize = hammer_blocksize(uio->uio_offset);
220 seqcount = (uio->uio_resid + (blksize - 1)) / blksize;
221 ioseqcount = ap->a_ioflag >> 16;
222 if (seqcount < ioseqcount)
223 seqcount = ioseqcount;
225 hammer_start_transaction(&trans, ip->hmp);
228 * Access the data typically in HAMMER_BUFSIZE blocks via the
229 * buffer cache, but HAMMER may use a variable block size based
232 while (uio->uio_resid > 0 && uio->uio_offset < ip->ino_data.size) {
236 blksize = hammer_blocksize(uio->uio_offset);
237 offset = (int)uio->uio_offset & (blksize - 1);
238 base_offset = uio->uio_offset - offset;
240 if (hammer_debug_cluster_enable) {
242 * Use file_limit to prevent cluster_read() from
243 * creating buffers of the wrong block size past
246 file_limit = ip->ino_data.size;
247 if (base_offset < HAMMER_XDEMARC &&
248 file_limit > HAMMER_XDEMARC) {
249 file_limit = HAMMER_XDEMARC;
251 error = cluster_read(ap->a_vp,
252 file_limit, base_offset,
256 error = bread(ap->a_vp, base_offset, blksize, &bp);
259 kprintf("error %d\n", error);
264 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
265 n = blksize - offset;
266 if (n > uio->uio_resid)
268 if (n > ip->ino_data.size - uio->uio_offset)
269 n = (int)(ip->ino_data.size - uio->uio_offset);
270 error = uiomove((char *)bp->b_data + offset, n, uio);
272 /* data has a lower priority then meta-data */
273 bp->b_flags |= B_AGE;
278 if ((ip->flags & HAMMER_INODE_RO) == 0 &&
279 (ip->hmp->mp->mnt_flag & MNT_NOATIME) == 0) {
280 ip->ino_data.atime = trans.time;
281 hammer_modify_inode(ip, HAMMER_INODE_ATIME);
283 hammer_done_transaction(&trans);
288 * hammer_vop_write { vp, uio, ioflag, cred }
292 hammer_vop_write(struct vop_write_args *ap)
294 struct hammer_transaction trans;
295 struct hammer_inode *ip;
307 if (ap->a_vp->v_type != VREG)
312 seqcount = ap->a_ioflag >> 16;
314 if (ip->flags & HAMMER_INODE_RO)
318 * Create a transaction to cover the operations we perform.
320 hammer_start_transaction(&trans, hmp);
326 if (ap->a_ioflag & IO_APPEND)
327 uio->uio_offset = ip->ino_data.size;
330 * Check for illegal write offsets. Valid range is 0...2^63-1.
332 * NOTE: the base_off assignment is required to work around what
333 * I consider to be a GCC-4 optimization bug.
335 if (uio->uio_offset < 0) {
336 hammer_done_transaction(&trans);
339 base_offset = uio->uio_offset + uio->uio_resid; /* work around gcc-4 */
340 if (uio->uio_resid > 0 && base_offset <= 0) {
341 hammer_done_transaction(&trans);
346 * Access the data typically in HAMMER_BUFSIZE blocks via the
347 * buffer cache, but HAMMER may use a variable block size based
350 while (uio->uio_resid > 0) {
355 if ((error = hammer_checkspace(hmp, HAMMER_CHECKSPACE_SLOP_WRITE)) != 0)
358 blksize = hammer_blocksize(uio->uio_offset);
361 * Do not allow HAMMER to blow out the buffer cache. Very
362 * large UIOs can lockout other processes due to bwillwrite()
365 * The hammer inode is not locked during these operations.
366 * The vnode is locked which can interfere with the pageout
367 * daemon for non-UIO_NOCOPY writes but should not interfere
368 * with the buffer cache. Even so, we cannot afford to
369 * allow the pageout daemon to build up too many dirty buffer
372 /*if (((int)uio->uio_offset & (blksize - 1)) == 0)*/
376 * Do not allow HAMMER to blow out system memory by
377 * accumulating too many records. Records are so well
378 * decoupled from the buffer cache that it is possible
379 * for userland to push data out to the media via
380 * direct-write, but build up the records queued to the
381 * backend faster then the backend can flush them out.
382 * HAMMER has hit its write limit but the frontend has
383 * no pushback to slow it down.
385 if (hmp->rsv_recs > hammer_limit_recs / 2) {
387 * Get the inode on the flush list
389 if (ip->rsv_recs >= 64)
390 hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
391 else if (ip->rsv_recs >= 16)
392 hammer_flush_inode(ip, 0);
395 * Keep the flusher going if the system keeps
398 delta = hmp->count_newrecords -
399 hmp->last_newrecords;
400 if (delta < 0 || delta > hammer_limit_recs / 2) {
401 hmp->last_newrecords = hmp->count_newrecords;
402 hammer_sync_hmp(hmp, MNT_NOWAIT);
406 * If we have gotten behind start slowing
409 delta = (hmp->rsv_recs - hammer_limit_recs) *
410 hz / hammer_limit_recs;
412 tsleep(&trans, 0, "hmrslo", delta);
416 * Calculate the blocksize at the current offset and figure
417 * out how much we can actually write.
419 blkmask = blksize - 1;
420 offset = (int)uio->uio_offset & blkmask;
421 base_offset = uio->uio_offset & ~(int64_t)blkmask;
422 n = blksize - offset;
423 if (n > uio->uio_resid)
425 if (uio->uio_offset + n > ip->ino_data.size) {
426 vnode_pager_setsize(ap->a_vp, uio->uio_offset + n);
430 if (uio->uio_segflg == UIO_NOCOPY) {
432 * Issuing a write with the same data backing the
433 * buffer. Instantiate the buffer to collect the
434 * backing vm pages, then read-in any missing bits.
436 * This case is used by vop_stdputpages().
438 bp = getblk(ap->a_vp, base_offset,
439 blksize, GETBLK_BHEAVY, 0);
440 if ((bp->b_flags & B_CACHE) == 0) {
442 error = bread(ap->a_vp, base_offset,
445 } else if (offset == 0 && uio->uio_resid >= blksize) {
447 * Even though we are entirely overwriting the buffer
448 * we may still have to zero it out to avoid a
449 * mmap/write visibility issue.
451 bp = getblk(ap->a_vp, base_offset, blksize, GETBLK_BHEAVY, 0);
452 if ((bp->b_flags & B_CACHE) == 0)
454 } else if (base_offset >= ip->ino_data.size) {
456 * If the base offset of the buffer is beyond the
457 * file EOF, we don't have to issue a read.
459 bp = getblk(ap->a_vp, base_offset,
460 blksize, GETBLK_BHEAVY, 0);
464 * Partial overwrite, read in any missing bits then
465 * replace the portion being written.
467 error = bread(ap->a_vp, base_offset, blksize, &bp);
472 error = uiomove((char *)bp->b_data + offset,
477 * If we screwed up we have to undo any VM size changes we
483 vtruncbuf(ap->a_vp, ip->ino_data.size,
484 hammer_blocksize(ip->ino_data.size));
488 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
489 if (ip->ino_data.size < uio->uio_offset) {
490 ip->ino_data.size = uio->uio_offset;
491 flags = HAMMER_INODE_DDIRTY;
492 vnode_pager_setsize(ap->a_vp, ip->ino_data.size);
496 ip->ino_data.mtime = trans.time;
497 flags |= HAMMER_INODE_MTIME | HAMMER_INODE_BUFS;
498 hammer_modify_inode(ip, flags);
501 * Final buffer disposition.
503 bp->b_flags |= B_AGE;
504 if (ap->a_ioflag & IO_SYNC) {
506 } else if (ap->a_ioflag & IO_DIRECT) {
512 hammer_done_transaction(&trans);
517 * hammer_vop_access { vp, mode, cred }
521 hammer_vop_access(struct vop_access_args *ap)
523 struct hammer_inode *ip = VTOI(ap->a_vp);
528 uid = hammer_to_unix_xid(&ip->ino_data.uid);
529 gid = hammer_to_unix_xid(&ip->ino_data.gid);
531 error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
532 ip->ino_data.uflags);
537 * hammer_vop_advlock { vp, id, op, fl, flags }
541 hammer_vop_advlock(struct vop_advlock_args *ap)
543 hammer_inode_t ip = VTOI(ap->a_vp);
545 return (lf_advlock(ap, &ip->advlock, ip->ino_data.size));
549 * hammer_vop_close { vp, fflag }
553 hammer_vop_close(struct vop_close_args *ap)
555 hammer_inode_t ip = VTOI(ap->a_vp);
557 if ((ip->flags | ip->sync_flags) & HAMMER_INODE_MODMASK)
558 hammer_inode_waitreclaims(ip->hmp);
559 return (vop_stdclose(ap));
563 * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
565 * The operating system has already ensured that the directory entry
566 * does not exist and done all appropriate namespace locking.
570 hammer_vop_ncreate(struct vop_ncreate_args *ap)
572 struct hammer_transaction trans;
573 struct hammer_inode *dip;
574 struct hammer_inode *nip;
575 struct nchandle *nch;
579 dip = VTOI(ap->a_dvp);
581 if (dip->flags & HAMMER_INODE_RO)
583 if ((error = hammer_checkspace(dip->hmp, HAMMER_CHECKSPACE_SLOP_CREATE)) != 0)
587 * Create a transaction to cover the operations we perform.
589 hammer_start_transaction(&trans, dip->hmp);
592 * Create a new filesystem object of the requested type. The
593 * returned inode will be referenced and shared-locked to prevent
594 * it from being moved to the flusher.
597 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred,
600 hkprintf("hammer_create_inode error %d\n", error);
601 hammer_done_transaction(&trans);
607 * Add the new filesystem object to the directory. This will also
608 * bump the inode's link count.
610 error = hammer_ip_add_directory(&trans, dip,
611 nch->ncp->nc_name, nch->ncp->nc_nlen,
614 hkprintf("hammer_ip_add_directory error %d\n", error);
620 hammer_rel_inode(nip, 0);
621 hammer_done_transaction(&trans);
624 error = hammer_get_vnode(nip, ap->a_vpp);
625 hammer_done_transaction(&trans);
626 hammer_rel_inode(nip, 0);
628 cache_setunresolved(ap->a_nch);
629 cache_setvp(ap->a_nch, *ap->a_vpp);
636 * hammer_vop_getattr { vp, vap }
638 * Retrieve an inode's attribute information. When accessing inodes
639 * historically we fake the atime field to ensure consistent results.
640 * The atime field is stored in the B-Tree element and allowed to be
641 * updated without cycling the element.
645 hammer_vop_getattr(struct vop_getattr_args *ap)
647 struct hammer_inode *ip = VTOI(ap->a_vp);
648 struct vattr *vap = ap->a_vap;
651 * We want the fsid to be different when accessing a filesystem
652 * with different as-of's so programs like diff don't think
653 * the files are the same.
655 * We also want the fsid to be the same when comparing snapshots,
656 * or when comparing mirrors (which might be backed by different
657 * physical devices). HAMMER fsids are based on the PFS's
660 * XXX there is a chance of collision here. The va_fsid reported
661 * by stat is different from the more involved fsid used in the
664 vap->va_fsid = ip->pfsm->fsid_udev ^ (u_int32_t)ip->obj_asof ^
665 (u_int32_t)(ip->obj_asof >> 32);
667 vap->va_fileid = ip->ino_leaf.base.obj_id;
668 vap->va_mode = ip->ino_data.mode;
669 vap->va_nlink = ip->ino_data.nlinks;
670 vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
671 vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
674 vap->va_size = ip->ino_data.size;
677 * We must provide a consistent atime and mtime for snapshots
678 * so people can do a 'tar cf - ... | md5' on them and get
679 * consistent results.
681 if (ip->flags & HAMMER_INODE_RO) {
682 hammer_time_to_timespec(ip->ino_data.ctime, &vap->va_atime);
683 hammer_time_to_timespec(ip->ino_data.ctime, &vap->va_mtime);
685 hammer_time_to_timespec(ip->ino_data.atime, &vap->va_atime);
686 hammer_time_to_timespec(ip->ino_data.mtime, &vap->va_mtime);
688 hammer_time_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
689 vap->va_flags = ip->ino_data.uflags;
690 vap->va_gen = 1; /* hammer inums are unique for all time */
691 vap->va_blocksize = HAMMER_BUFSIZE;
692 if (ip->ino_data.size >= HAMMER_XDEMARC) {
693 vap->va_bytes = (ip->ino_data.size + HAMMER_XBUFMASK64) &
695 } else if (ip->ino_data.size > HAMMER_BUFSIZE / 2) {
696 vap->va_bytes = (ip->ino_data.size + HAMMER_BUFMASK64) &
699 vap->va_bytes = (ip->ino_data.size + 15) & ~15;
701 vap->va_type = hammer_get_vnode_type(ip->ino_data.obj_type);
702 vap->va_filerev = 0; /* XXX */
703 /* mtime uniquely identifies any adjustments made to the file XXX */
704 vap->va_fsmid = ip->ino_data.mtime;
705 vap->va_uid_uuid = ip->ino_data.uid;
706 vap->va_gid_uuid = ip->ino_data.gid;
707 vap->va_fsid_uuid = ip->hmp->fsid;
708 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
711 switch (ip->ino_data.obj_type) {
712 case HAMMER_OBJTYPE_CDEV:
713 case HAMMER_OBJTYPE_BDEV:
714 vap->va_rmajor = ip->ino_data.rmajor;
715 vap->va_rminor = ip->ino_data.rminor;
724 * hammer_vop_nresolve { nch, dvp, cred }
726 * Locate the requested directory entry.
730 hammer_vop_nresolve(struct vop_nresolve_args *ap)
732 struct hammer_transaction trans;
733 struct namecache *ncp;
737 struct hammer_cursor cursor;
746 u_int32_t localization;
749 * Misc initialization, plus handle as-of name extensions. Look for
750 * the '@@' extension. Note that as-of files and directories cannot
753 dip = VTOI(ap->a_dvp);
754 ncp = ap->a_nch->ncp;
755 asof = dip->obj_asof;
757 flags = dip->flags & HAMMER_INODE_RO;
760 hammer_simple_transaction(&trans, dip->hmp);
762 for (i = 0; i < nlen; ++i) {
763 if (ncp->nc_name[i] == '@' && ncp->nc_name[i+1] == '@') {
764 asof = hammer_str_to_tid(ncp->nc_name + i + 2,
765 &ispfs, &localization);
766 if (asof != HAMMER_MAX_TID)
767 flags |= HAMMER_INODE_RO;
774 * If this is a PFS softlink we dive into the PFS
776 if (ispfs && nlen == 0) {
777 ip = hammer_get_inode(&trans, dip, HAMMER_OBJID_ROOT,
781 error = hammer_get_vnode(ip, &vp);
782 hammer_rel_inode(ip, 0);
788 cache_setvp(ap->a_nch, vp);
795 * If there is no path component the time extension is relative to
799 ip = hammer_get_inode(&trans, dip, dip->obj_id,
800 asof, dip->obj_localization,
803 error = hammer_get_vnode(ip, &vp);
804 hammer_rel_inode(ip, 0);
810 cache_setvp(ap->a_nch, vp);
817 * Calculate the namekey and setup the key range for the scan. This
818 * works kinda like a chained hash table where the lower 32 bits
819 * of the namekey synthesize the chain.
821 * The key range is inclusive of both key_beg and key_end.
823 namekey = hammer_directory_namekey(ncp->nc_name, nlen);
825 error = hammer_init_cursor(&trans, &cursor, &dip->cache[1], dip);
826 cursor.key_beg.localization = dip->obj_localization +
827 HAMMER_LOCALIZE_MISC;
828 cursor.key_beg.obj_id = dip->obj_id;
829 cursor.key_beg.key = namekey;
830 cursor.key_beg.create_tid = 0;
831 cursor.key_beg.delete_tid = 0;
832 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
833 cursor.key_beg.obj_type = 0;
835 cursor.key_end = cursor.key_beg;
836 cursor.key_end.key |= 0xFFFFFFFFULL;
838 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
841 * Scan all matching records (the chain), locate the one matching
842 * the requested path component.
844 * The hammer_ip_*() functions merge in-memory records with on-disk
845 * records for the purposes of the search.
848 localization = HAMMER_DEF_LOCALIZATION;
851 error = hammer_ip_first(&cursor);
853 error = hammer_ip_resolve_data(&cursor);
856 if (nlen == cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF &&
857 bcmp(ncp->nc_name, cursor.data->entry.name, nlen) == 0) {
858 obj_id = cursor.data->entry.obj_id;
859 localization = cursor.data->entry.localization;
862 error = hammer_ip_next(&cursor);
865 hammer_done_cursor(&cursor);
867 ip = hammer_get_inode(&trans, dip, obj_id,
871 error = hammer_get_vnode(ip, &vp);
872 hammer_rel_inode(ip, 0);
878 cache_setvp(ap->a_nch, vp);
881 } else if (error == ENOENT) {
882 cache_setvp(ap->a_nch, NULL);
885 hammer_done_transaction(&trans);
890 * hammer_vop_nlookupdotdot { dvp, vpp, cred }
892 * Locate the parent directory of a directory vnode.
894 * dvp is referenced but not locked. *vpp must be returned referenced and
895 * locked. A parent_obj_id of 0 does not necessarily indicate that we are
896 * at the root, instead it could indicate that the directory we were in was
899 * NOTE: as-of sequences are not linked into the directory structure. If
900 * we are at the root with a different asof then the mount point, reload
901 * the same directory with the mount point's asof. I'm not sure what this
902 * will do to NFS. We encode ASOF stamps in NFS file handles so it might not
903 * get confused, but it hasn't been tested.
907 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
909 struct hammer_transaction trans;
910 struct hammer_inode *dip;
911 struct hammer_inode *ip;
912 int64_t parent_obj_id;
913 u_int32_t parent_obj_localization;
917 dip = VTOI(ap->a_dvp);
918 asof = dip->obj_asof;
921 * Whos are parent? This could be the root of a pseudo-filesystem
922 * whos parent is in another localization domain.
924 parent_obj_id = dip->ino_data.parent_obj_id;
925 if (dip->obj_id == HAMMER_OBJID_ROOT)
926 parent_obj_localization = dip->ino_data.ext.obj.parent_obj_localization;
928 parent_obj_localization = dip->obj_localization;
930 if (parent_obj_id == 0) {
931 if (dip->obj_id == HAMMER_OBJID_ROOT &&
932 asof != dip->hmp->asof) {
933 parent_obj_id = dip->obj_id;
934 asof = dip->hmp->asof;
935 *ap->a_fakename = kmalloc(19, M_TEMP, M_WAITOK);
936 ksnprintf(*ap->a_fakename, 19, "0x%016llx",
944 hammer_simple_transaction(&trans, dip->hmp);
946 ip = hammer_get_inode(&trans, dip, parent_obj_id,
947 asof, parent_obj_localization,
950 error = hammer_get_vnode(ip, ap->a_vpp);
951 hammer_rel_inode(ip, 0);
955 hammer_done_transaction(&trans);
960 * hammer_vop_nlink { nch, dvp, vp, cred }
964 hammer_vop_nlink(struct vop_nlink_args *ap)
966 struct hammer_transaction trans;
967 struct hammer_inode *dip;
968 struct hammer_inode *ip;
969 struct nchandle *nch;
973 dip = VTOI(ap->a_dvp);
976 if (dip->flags & HAMMER_INODE_RO)
978 if (ip->flags & HAMMER_INODE_RO)
980 if ((error = hammer_checkspace(dip->hmp, HAMMER_CHECKSPACE_SLOP_CREATE)) != 0)
984 * Create a transaction to cover the operations we perform.
986 hammer_start_transaction(&trans, dip->hmp);
989 * Add the filesystem object to the directory. Note that neither
990 * dip nor ip are referenced or locked, but their vnodes are
991 * referenced. This function will bump the inode's link count.
993 error = hammer_ip_add_directory(&trans, dip,
994 nch->ncp->nc_name, nch->ncp->nc_nlen,
1001 cache_setunresolved(nch);
1002 cache_setvp(nch, ap->a_vp);
1004 hammer_done_transaction(&trans);
1009 * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
1011 * The operating system has already ensured that the directory entry
1012 * does not exist and done all appropriate namespace locking.
1016 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
1018 struct hammer_transaction trans;
1019 struct hammer_inode *dip;
1020 struct hammer_inode *nip;
1021 struct nchandle *nch;
1025 dip = VTOI(ap->a_dvp);
1027 if (dip->flags & HAMMER_INODE_RO)
1029 if ((error = hammer_checkspace(dip->hmp, HAMMER_CHECKSPACE_SLOP_CREATE)) != 0)
1033 * Create a transaction to cover the operations we perform.
1035 hammer_start_transaction(&trans, dip->hmp);
1038 * Create a new filesystem object of the requested type. The
1039 * returned inode will be referenced but not locked.
1041 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred,
1044 hkprintf("hammer_mkdir error %d\n", error);
1045 hammer_done_transaction(&trans);
1050 * Add the new filesystem object to the directory. This will also
1051 * bump the inode's link count.
1053 error = hammer_ip_add_directory(&trans, dip,
1054 nch->ncp->nc_name, nch->ncp->nc_nlen,
1057 hkprintf("hammer_mkdir (add) error %d\n", error);
1063 hammer_rel_inode(nip, 0);
1066 error = hammer_get_vnode(nip, ap->a_vpp);
1067 hammer_rel_inode(nip, 0);
1069 cache_setunresolved(ap->a_nch);
1070 cache_setvp(ap->a_nch, *ap->a_vpp);
1073 hammer_done_transaction(&trans);
1078 * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
1080 * The operating system has already ensured that the directory entry
1081 * does not exist and done all appropriate namespace locking.
1085 hammer_vop_nmknod(struct vop_nmknod_args *ap)
1087 struct hammer_transaction trans;
1088 struct hammer_inode *dip;
1089 struct hammer_inode *nip;
1090 struct nchandle *nch;
1094 dip = VTOI(ap->a_dvp);
1096 if (dip->flags & HAMMER_INODE_RO)
1098 if ((error = hammer_checkspace(dip->hmp, HAMMER_CHECKSPACE_SLOP_CREATE)) != 0)
1102 * Create a transaction to cover the operations we perform.
1104 hammer_start_transaction(&trans, dip->hmp);
1107 * Create a new filesystem object of the requested type. The
1108 * returned inode will be referenced but not locked.
1110 * If mknod specifies a directory a pseudo-fs is created.
1112 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred,
1115 hammer_done_transaction(&trans);
1121 * Add the new filesystem object to the directory. This will also
1122 * bump the inode's link count.
1124 error = hammer_ip_add_directory(&trans, dip,
1125 nch->ncp->nc_name, nch->ncp->nc_nlen,
1132 hammer_rel_inode(nip, 0);
1135 error = hammer_get_vnode(nip, ap->a_vpp);
1136 hammer_rel_inode(nip, 0);
1138 cache_setunresolved(ap->a_nch);
1139 cache_setvp(ap->a_nch, *ap->a_vpp);
1142 hammer_done_transaction(&trans);
1147 * hammer_vop_open { vp, mode, cred, fp }
1151 hammer_vop_open(struct vop_open_args *ap)
1155 ip = VTOI(ap->a_vp);
1157 if ((ap->a_mode & FWRITE) && (ip->flags & HAMMER_INODE_RO))
1159 return(vop_stdopen(ap));
1163 * hammer_vop_pathconf { vp, name, retval }
1167 hammer_vop_pathconf(struct vop_pathconf_args *ap)
1173 * hammer_vop_print { vp }
1177 hammer_vop_print(struct vop_print_args *ap)
1183 * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
1187 hammer_vop_readdir(struct vop_readdir_args *ap)
1189 struct hammer_transaction trans;
1190 struct hammer_cursor cursor;
1191 struct hammer_inode *ip;
1193 hammer_base_elm_t base;
1202 ip = VTOI(ap->a_vp);
1204 saveoff = uio->uio_offset;
1206 if (ap->a_ncookies) {
1207 ncookies = uio->uio_resid / 16 + 1;
1208 if (ncookies > 1024)
1210 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1218 hammer_simple_transaction(&trans, ip->hmp);
1221 * Handle artificial entries
1225 r = vop_write_dirent(&error, uio, ip->obj_id, DT_DIR, 1, ".");
1229 cookies[cookie_index] = saveoff;
1232 if (cookie_index == ncookies)
1236 if (ip->ino_data.parent_obj_id) {
1237 r = vop_write_dirent(&error, uio,
1238 ip->ino_data.parent_obj_id,
1241 r = vop_write_dirent(&error, uio,
1242 ip->obj_id, DT_DIR, 2, "..");
1247 cookies[cookie_index] = saveoff;
1250 if (cookie_index == ncookies)
1255 * Key range (begin and end inclusive) to scan. Directory keys
1256 * directly translate to a 64 bit 'seek' position.
1258 hammer_init_cursor(&trans, &cursor, &ip->cache[1], ip);
1259 cursor.key_beg.localization = ip->obj_localization +
1260 HAMMER_LOCALIZE_MISC;
1261 cursor.key_beg.obj_id = ip->obj_id;
1262 cursor.key_beg.create_tid = 0;
1263 cursor.key_beg.delete_tid = 0;
1264 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1265 cursor.key_beg.obj_type = 0;
1266 cursor.key_beg.key = saveoff;
1268 cursor.key_end = cursor.key_beg;
1269 cursor.key_end.key = HAMMER_MAX_KEY;
1270 cursor.asof = ip->obj_asof;
1271 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1273 error = hammer_ip_first(&cursor);
1275 while (error == 0) {
1276 error = hammer_ip_resolve_data(&cursor);
1279 base = &cursor.leaf->base;
1280 saveoff = base->key;
1281 KKASSERT(cursor.leaf->data_len > HAMMER_ENTRY_NAME_OFF);
1283 if (base->obj_id != ip->obj_id)
1284 panic("readdir: bad record at %p", cursor.node);
1287 * Convert pseudo-filesystems into softlinks
1289 dtype = hammer_get_dtype(cursor.leaf->base.obj_type);
1290 r = vop_write_dirent(
1291 &error, uio, cursor.data->entry.obj_id,
1293 cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF ,
1294 (void *)cursor.data->entry.name);
1299 cookies[cookie_index] = base->key;
1301 if (cookie_index == ncookies)
1303 error = hammer_ip_next(&cursor);
1305 hammer_done_cursor(&cursor);
1308 hammer_done_transaction(&trans);
1311 *ap->a_eofflag = (error == ENOENT);
1312 uio->uio_offset = saveoff;
1313 if (error && cookie_index == 0) {
1314 if (error == ENOENT)
1317 kfree(cookies, M_TEMP);
1318 *ap->a_ncookies = 0;
1319 *ap->a_cookies = NULL;
1322 if (error == ENOENT)
1325 *ap->a_ncookies = cookie_index;
1326 *ap->a_cookies = cookies;
1333 * hammer_vop_readlink { vp, uio, cred }
1337 hammer_vop_readlink(struct vop_readlink_args *ap)
1339 struct hammer_transaction trans;
1340 struct hammer_cursor cursor;
1341 struct hammer_inode *ip;
1343 u_int32_t localization;
1344 hammer_pseudofs_inmem_t pfsm;
1347 ip = VTOI(ap->a_vp);
1350 * Shortcut if the symlink data was stuffed into ino_data.
1352 * Also expand special "@@PFS%05d" softlinks.
1354 if (ip->ino_data.size <= HAMMER_INODE_BASESYMLEN) {
1358 ptr = ip->ino_data.ext.symlink;
1359 bytes = (int)ip->ino_data.size;
1360 if (bytes == 10 && strncmp(ptr, "@@PFS", 5) == 0) {
1361 hammer_simple_transaction(&trans, ip->hmp);
1362 bcopy(ptr + 5, buf, 5);
1364 localization = strtoul(buf, NULL, 10) << 16;
1365 pfsm = hammer_load_pseudofs(&trans, localization,
1368 if (pfsm->pfsd.mirror_flags &
1369 HAMMER_PFSD_SLAVE) {
1370 ksnprintf(buf, sizeof(buf),
1372 pfsm->pfsd.sync_end_tid,
1373 localization >> 16);
1375 ksnprintf(buf, sizeof(buf),
1378 localization >> 16);
1381 bytes = strlen(buf);
1384 hammer_rel_pseudofs(trans.hmp, pfsm);
1385 hammer_done_transaction(&trans);
1387 error = uiomove(ptr, bytes, ap->a_uio);
1394 hammer_simple_transaction(&trans, ip->hmp);
1395 hammer_init_cursor(&trans, &cursor, &ip->cache[1], ip);
1398 * Key range (begin and end inclusive) to scan. Directory keys
1399 * directly translate to a 64 bit 'seek' position.
1401 cursor.key_beg.localization = ip->obj_localization +
1402 HAMMER_LOCALIZE_MISC;
1403 cursor.key_beg.obj_id = ip->obj_id;
1404 cursor.key_beg.create_tid = 0;
1405 cursor.key_beg.delete_tid = 0;
1406 cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
1407 cursor.key_beg.obj_type = 0;
1408 cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
1409 cursor.asof = ip->obj_asof;
1410 cursor.flags |= HAMMER_CURSOR_ASOF;
1412 error = hammer_ip_lookup(&cursor);
1414 error = hammer_ip_resolve_data(&cursor);
1416 KKASSERT(cursor.leaf->data_len >=
1417 HAMMER_SYMLINK_NAME_OFF);
1418 error = uiomove(cursor.data->symlink.name,
1419 cursor.leaf->data_len -
1420 HAMMER_SYMLINK_NAME_OFF,
1424 hammer_done_cursor(&cursor);
1425 hammer_done_transaction(&trans);
1430 * hammer_vop_nremove { nch, dvp, cred }
1434 hammer_vop_nremove(struct vop_nremove_args *ap)
1436 struct hammer_transaction trans;
1437 struct hammer_inode *dip;
1440 dip = VTOI(ap->a_dvp);
1442 if (hammer_nohistory(dip) == 0 &&
1443 (error = hammer_checkspace(dip->hmp, HAMMER_CHECKSPACE_SLOP_REMOVE)) != 0) {
1447 hammer_start_transaction(&trans, dip->hmp);
1448 error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1449 hammer_done_transaction(&trans);
1455 * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1459 hammer_vop_nrename(struct vop_nrename_args *ap)
1461 struct hammer_transaction trans;
1462 struct namecache *fncp;
1463 struct namecache *tncp;
1464 struct hammer_inode *fdip;
1465 struct hammer_inode *tdip;
1466 struct hammer_inode *ip;
1467 struct hammer_cursor cursor;
1471 fdip = VTOI(ap->a_fdvp);
1472 tdip = VTOI(ap->a_tdvp);
1473 fncp = ap->a_fnch->ncp;
1474 tncp = ap->a_tnch->ncp;
1475 ip = VTOI(fncp->nc_vp);
1476 KKASSERT(ip != NULL);
1478 if (fdip->flags & HAMMER_INODE_RO)
1480 if (tdip->flags & HAMMER_INODE_RO)
1482 if (ip->flags & HAMMER_INODE_RO)
1484 if ((error = hammer_checkspace(fdip->hmp, HAMMER_CHECKSPACE_SLOP_CREATE)) != 0)
1487 hammer_start_transaction(&trans, fdip->hmp);
1490 * Remove tncp from the target directory and then link ip as
1491 * tncp. XXX pass trans to dounlink
1493 * Force the inode sync-time to match the transaction so it is
1494 * in-sync with the creation of the target directory entry.
1496 error = hammer_dounlink(&trans, ap->a_tnch, ap->a_tdvp, ap->a_cred, 0);
1497 if (error == 0 || error == ENOENT) {
1498 error = hammer_ip_add_directory(&trans, tdip,
1499 tncp->nc_name, tncp->nc_nlen,
1502 ip->ino_data.parent_obj_id = tdip->obj_id;
1503 hammer_modify_inode(ip, HAMMER_INODE_DDIRTY);
1507 goto failed; /* XXX */
1510 * Locate the record in the originating directory and remove it.
1512 * Calculate the namekey and setup the key range for the scan. This
1513 * works kinda like a chained hash table where the lower 32 bits
1514 * of the namekey synthesize the chain.
1516 * The key range is inclusive of both key_beg and key_end.
1518 namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
1520 hammer_init_cursor(&trans, &cursor, &fdip->cache[1], fdip);
1521 cursor.key_beg.localization = fdip->obj_localization +
1522 HAMMER_LOCALIZE_MISC;
1523 cursor.key_beg.obj_id = fdip->obj_id;
1524 cursor.key_beg.key = namekey;
1525 cursor.key_beg.create_tid = 0;
1526 cursor.key_beg.delete_tid = 0;
1527 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1528 cursor.key_beg.obj_type = 0;
1530 cursor.key_end = cursor.key_beg;
1531 cursor.key_end.key |= 0xFFFFFFFFULL;
1532 cursor.asof = fdip->obj_asof;
1533 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1536 * Scan all matching records (the chain), locate the one matching
1537 * the requested path component.
1539 * The hammer_ip_*() functions merge in-memory records with on-disk
1540 * records for the purposes of the search.
1542 error = hammer_ip_first(&cursor);
1543 while (error == 0) {
1544 if (hammer_ip_resolve_data(&cursor) != 0)
1546 nlen = cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF;
1548 if (fncp->nc_nlen == nlen &&
1549 bcmp(fncp->nc_name, cursor.data->entry.name, nlen) == 0) {
1552 error = hammer_ip_next(&cursor);
1556 * If all is ok we have to get the inode so we can adjust nlinks.
1558 * WARNING: hammer_ip_del_directory() may have to terminate the
1559 * cursor to avoid a recursion. It's ok to call hammer_done_cursor()
1563 error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
1566 * XXX A deadlock here will break rename's atomicy for the purposes
1567 * of crash recovery.
1569 if (error == EDEADLK) {
1570 hammer_done_cursor(&cursor);
1575 * Cleanup and tell the kernel that the rename succeeded.
1577 hammer_done_cursor(&cursor);
1579 cache_rename(ap->a_fnch, ap->a_tnch);
1582 hammer_done_transaction(&trans);
1587 * hammer_vop_nrmdir { nch, dvp, cred }
1591 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1593 struct hammer_transaction trans;
1594 struct hammer_inode *dip;
1597 dip = VTOI(ap->a_dvp);
1599 if (hammer_nohistory(dip) == 0 &&
1600 (error = hammer_checkspace(dip->hmp, HAMMER_CHECKSPACE_SLOP_REMOVE)) != 0) {
1604 hammer_start_transaction(&trans, dip->hmp);
1605 error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1606 hammer_done_transaction(&trans);
1612 * hammer_vop_setattr { vp, vap, cred }
1616 hammer_vop_setattr(struct vop_setattr_args *ap)
1618 struct hammer_transaction trans;
1620 struct hammer_inode *ip;
1625 int64_t aligned_size;
1629 ip = ap->a_vp->v_data;
1632 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1634 if (ip->flags & HAMMER_INODE_RO)
1636 if (hammer_nohistory(ip) == 0 &&
1637 (error = hammer_checkspace(ip->hmp, HAMMER_CHECKSPACE_SLOP_REMOVE)) != 0) {
1641 hammer_start_transaction(&trans, ip->hmp);
1644 if (vap->va_flags != VNOVAL) {
1645 flags = ip->ino_data.uflags;
1646 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1647 hammer_to_unix_xid(&ip->ino_data.uid),
1650 if (ip->ino_data.uflags != flags) {
1651 ip->ino_data.uflags = flags;
1652 modflags |= HAMMER_INODE_DDIRTY;
1654 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1661 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1665 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
1666 mode_t cur_mode = ip->ino_data.mode;
1667 uid_t cur_uid = hammer_to_unix_xid(&ip->ino_data.uid);
1668 gid_t cur_gid = hammer_to_unix_xid(&ip->ino_data.gid);
1672 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
1674 &cur_uid, &cur_gid, &cur_mode);
1676 hammer_guid_to_uuid(&uuid_uid, cur_uid);
1677 hammer_guid_to_uuid(&uuid_gid, cur_gid);
1678 if (bcmp(&uuid_uid, &ip->ino_data.uid,
1679 sizeof(uuid_uid)) ||
1680 bcmp(&uuid_gid, &ip->ino_data.gid,
1681 sizeof(uuid_gid)) ||
1682 ip->ino_data.mode != cur_mode
1684 ip->ino_data.uid = uuid_uid;
1685 ip->ino_data.gid = uuid_gid;
1686 ip->ino_data.mode = cur_mode;
1688 modflags |= HAMMER_INODE_DDIRTY;
1691 while (vap->va_size != VNOVAL && ip->ino_data.size != vap->va_size) {
1692 switch(ap->a_vp->v_type) {
1694 if (vap->va_size == ip->ino_data.size)
1697 * XXX break atomicy, we can deadlock the backend
1698 * if we do not release the lock. Probably not a
1701 blksize = hammer_blocksize(vap->va_size);
1702 if (vap->va_size < ip->ino_data.size) {
1703 vtruncbuf(ap->a_vp, vap->va_size, blksize);
1706 vnode_pager_setsize(ap->a_vp, vap->va_size);
1709 ip->ino_data.size = vap->va_size;
1710 modflags |= HAMMER_INODE_DDIRTY;
1713 * on-media truncation is cached in the inode until
1714 * the inode is synchronized.
1717 hammer_ip_frontend_trunc(ip, vap->va_size);
1718 #ifdef DEBUG_TRUNCATE
1719 if (HammerTruncIp == NULL)
1722 if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1723 ip->flags |= HAMMER_INODE_TRUNCATED;
1724 ip->trunc_off = vap->va_size;
1725 #ifdef DEBUG_TRUNCATE
1726 if (ip == HammerTruncIp)
1727 kprintf("truncate1 %016llx\n", ip->trunc_off);
1729 } else if (ip->trunc_off > vap->va_size) {
1730 ip->trunc_off = vap->va_size;
1731 #ifdef DEBUG_TRUNCATE
1732 if (ip == HammerTruncIp)
1733 kprintf("truncate2 %016llx\n", ip->trunc_off);
1736 #ifdef DEBUG_TRUNCATE
1737 if (ip == HammerTruncIp)
1738 kprintf("truncate3 %016llx (ignored)\n", vap->va_size);
1744 * If truncating we have to clean out a portion of
1745 * the last block on-disk. We do this in the
1746 * front-end buffer cache.
1748 aligned_size = (vap->va_size + (blksize - 1)) &
1749 ~(int64_t)(blksize - 1);
1750 if (truncating && vap->va_size < aligned_size) {
1754 aligned_size -= blksize;
1756 offset = (int)vap->va_size & (blksize - 1);
1757 error = bread(ap->a_vp, aligned_size,
1759 hammer_ip_frontend_trunc(ip, aligned_size);
1761 bzero(bp->b_data + offset,
1765 kprintf("ERROR %d\n", error);
1771 if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1772 ip->flags |= HAMMER_INODE_TRUNCATED;
1773 ip->trunc_off = vap->va_size;
1774 } else if (ip->trunc_off > vap->va_size) {
1775 ip->trunc_off = vap->va_size;
1777 hammer_ip_frontend_trunc(ip, vap->va_size);
1778 ip->ino_data.size = vap->va_size;
1779 modflags |= HAMMER_INODE_DDIRTY;
1787 if (vap->va_atime.tv_sec != VNOVAL) {
1788 ip->ino_data.atime =
1789 hammer_timespec_to_time(&vap->va_atime);
1790 modflags |= HAMMER_INODE_ATIME;
1792 if (vap->va_mtime.tv_sec != VNOVAL) {
1793 ip->ino_data.mtime =
1794 hammer_timespec_to_time(&vap->va_mtime);
1795 modflags |= HAMMER_INODE_MTIME;
1797 if (vap->va_mode != (mode_t)VNOVAL) {
1798 mode_t cur_mode = ip->ino_data.mode;
1799 uid_t cur_uid = hammer_to_unix_xid(&ip->ino_data.uid);
1800 gid_t cur_gid = hammer_to_unix_xid(&ip->ino_data.gid);
1802 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
1803 cur_uid, cur_gid, &cur_mode);
1804 if (error == 0 && ip->ino_data.mode != cur_mode) {
1805 ip->ino_data.mode = cur_mode;
1806 modflags |= HAMMER_INODE_DDIRTY;
1811 hammer_modify_inode(ip, modflags);
1812 hammer_done_transaction(&trans);
1817 * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1821 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1823 struct hammer_transaction trans;
1824 struct hammer_inode *dip;
1825 struct hammer_inode *nip;
1826 struct nchandle *nch;
1827 hammer_record_t record;
1831 ap->a_vap->va_type = VLNK;
1834 dip = VTOI(ap->a_dvp);
1836 if (dip->flags & HAMMER_INODE_RO)
1838 if ((error = hammer_checkspace(dip->hmp, HAMMER_CHECKSPACE_SLOP_CREATE)) != 0)
1842 * Create a transaction to cover the operations we perform.
1844 hammer_start_transaction(&trans, dip->hmp);
1847 * Create a new filesystem object of the requested type. The
1848 * returned inode will be referenced but not locked.
1851 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred,
1854 hammer_done_transaction(&trans);
1860 * Add a record representing the symlink. symlink stores the link
1861 * as pure data, not a string, and is no \0 terminated.
1864 bytes = strlen(ap->a_target);
1866 if (bytes <= HAMMER_INODE_BASESYMLEN) {
1867 bcopy(ap->a_target, nip->ino_data.ext.symlink, bytes);
1869 record = hammer_alloc_mem_record(nip, bytes);
1870 record->type = HAMMER_MEM_RECORD_GENERAL;
1872 record->leaf.base.localization = nip->obj_localization +
1873 HAMMER_LOCALIZE_MISC;
1874 record->leaf.base.key = HAMMER_FIXKEY_SYMLINK;
1875 record->leaf.base.rec_type = HAMMER_RECTYPE_FIX;
1876 record->leaf.data_len = bytes;
1877 KKASSERT(HAMMER_SYMLINK_NAME_OFF == 0);
1878 bcopy(ap->a_target, record->data->symlink.name, bytes);
1879 error = hammer_ip_add_record(&trans, record);
1883 * Set the file size to the length of the link.
1886 nip->ino_data.size = bytes;
1887 hammer_modify_inode(nip, HAMMER_INODE_DDIRTY);
1891 error = hammer_ip_add_directory(&trans, dip, nch->ncp->nc_name,
1892 nch->ncp->nc_nlen, nip);
1898 hammer_rel_inode(nip, 0);
1901 error = hammer_get_vnode(nip, ap->a_vpp);
1902 hammer_rel_inode(nip, 0);
1904 cache_setunresolved(ap->a_nch);
1905 cache_setvp(ap->a_nch, *ap->a_vpp);
1908 hammer_done_transaction(&trans);
1913 * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1917 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1919 struct hammer_transaction trans;
1920 struct hammer_inode *dip;
1923 dip = VTOI(ap->a_dvp);
1925 if (hammer_nohistory(dip) == 0 &&
1926 (error = hammer_checkspace(dip->hmp, HAMMER_CHECKSPACE_SLOP_CREATE)) != 0) {
1930 hammer_start_transaction(&trans, dip->hmp);
1931 error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp,
1932 ap->a_cred, ap->a_flags);
1933 hammer_done_transaction(&trans);
1939 * hammer_vop_ioctl { vp, command, data, fflag, cred }
1943 hammer_vop_ioctl(struct vop_ioctl_args *ap)
1945 struct hammer_inode *ip = ap->a_vp->v_data;
1947 return(hammer_ioctl(ip, ap->a_command, ap->a_data,
1948 ap->a_fflag, ap->a_cred));
1953 hammer_vop_mountctl(struct vop_mountctl_args *ap)
1958 mp = ap->a_head.a_ops->head.vv_mount;
1961 case MOUNTCTL_SET_EXPORT:
1962 if (ap->a_ctllen != sizeof(struct export_args))
1964 error = hammer_vfs_export(mp, ap->a_op,
1965 (const struct export_args *)ap->a_ctl);
1968 error = journal_mountctl(ap);
1975 * hammer_vop_strategy { vp, bio }
1977 * Strategy call, used for regular file read & write only. Note that the
1978 * bp may represent a cluster.
1980 * To simplify operation and allow better optimizations in the future,
1981 * this code does not make any assumptions with regards to buffer alignment
1986 hammer_vop_strategy(struct vop_strategy_args *ap)
1991 bp = ap->a_bio->bio_buf;
1995 error = hammer_vop_strategy_read(ap);
1998 error = hammer_vop_strategy_write(ap);
2001 bp->b_error = error = EINVAL;
2002 bp->b_flags |= B_ERROR;
2010 * Read from a regular file. Iterate the related records and fill in the
2011 * BIO/BUF. Gaps are zero-filled.
2013 * The support code in hammer_object.c should be used to deal with mixed
2014 * in-memory and on-disk records.
2016 * NOTE: Can be called from the cluster code with an oversized buf.
2022 hammer_vop_strategy_read(struct vop_strategy_args *ap)
2024 struct hammer_transaction trans;
2025 struct hammer_inode *ip;
2026 struct hammer_cursor cursor;
2027 hammer_base_elm_t base;
2028 hammer_off_t disk_offset;
2042 ip = ap->a_vp->v_data;
2045 * The zone-2 disk offset may have been set by the cluster code via
2046 * a BMAP operation, or else should be NOOFFSET.
2048 * Checking the high bits for a match against zone-2 should suffice.
2050 nbio = push_bio(bio);
2051 if ((nbio->bio_offset & HAMMER_OFF_ZONE_MASK) ==
2052 HAMMER_ZONE_RAW_BUFFER) {
2053 error = hammer_io_direct_read(ip->hmp, nbio);
2058 * Well, that sucked. Do it the hard way. If all the stars are
2059 * aligned we may still be able to issue a direct-read.
2061 hammer_simple_transaction(&trans, ip->hmp);
2062 hammer_init_cursor(&trans, &cursor, &ip->cache[1], ip);
2065 * Key range (begin and end inclusive) to scan. Note that the key's
2066 * stored in the actual records represent BASE+LEN, not BASE. The
2067 * first record containing bio_offset will have a key > bio_offset.
2069 cursor.key_beg.localization = ip->obj_localization +
2070 HAMMER_LOCALIZE_MISC;
2071 cursor.key_beg.obj_id = ip->obj_id;
2072 cursor.key_beg.create_tid = 0;
2073 cursor.key_beg.delete_tid = 0;
2074 cursor.key_beg.obj_type = 0;
2075 cursor.key_beg.key = bio->bio_offset + 1;
2076 cursor.asof = ip->obj_asof;
2077 cursor.flags |= HAMMER_CURSOR_ASOF;
2079 cursor.key_end = cursor.key_beg;
2080 KKASSERT(ip->ino_data.obj_type == HAMMER_OBJTYPE_REGFILE);
2082 if (ip->ino_data.obj_type == HAMMER_OBJTYPE_DBFILE) {
2083 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
2084 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
2085 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
2089 ran_end = bio->bio_offset + bp->b_bufsize;
2090 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
2091 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
2092 tmp64 = ran_end + MAXPHYS + 1; /* work-around GCC-4 bug */
2093 if (tmp64 < ran_end)
2094 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
2096 cursor.key_end.key = ran_end + MAXPHYS + 1;
2098 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
2100 error = hammer_ip_first(&cursor);
2103 while (error == 0) {
2105 * Get the base file offset of the record. The key for
2106 * data records is (base + bytes) rather then (base).
2108 base = &cursor.leaf->base;
2109 rec_offset = base->key - cursor.leaf->data_len;
2112 * Calculate the gap, if any, and zero-fill it.
2114 * n is the offset of the start of the record verses our
2115 * current seek offset in the bio.
2117 n = (int)(rec_offset - (bio->bio_offset + boff));
2119 if (n > bp->b_bufsize - boff)
2120 n = bp->b_bufsize - boff;
2121 bzero((char *)bp->b_data + boff, n);
2127 * Calculate the data offset in the record and the number
2128 * of bytes we can copy.
2130 * There are two degenerate cases. First, boff may already
2131 * be at bp->b_bufsize. Secondly, the data offset within
2132 * the record may exceed the record's size.
2136 n = cursor.leaf->data_len - roff;
2138 kprintf("strategy_read: bad n=%d roff=%d\n", n, roff);
2140 } else if (n > bp->b_bufsize - boff) {
2141 n = bp->b_bufsize - boff;
2145 * Deal with cached truncations. This cool bit of code
2146 * allows truncate()/ftruncate() to avoid having to sync
2149 * If the frontend is truncated then all backend records are
2150 * subject to the frontend's truncation.
2152 * If the backend is truncated then backend records on-disk
2153 * (but not in-memory) are subject to the backend's
2154 * truncation. In-memory records owned by the backend
2155 * represent data written after the truncation point on the
2156 * backend and must not be truncated.
2158 * Truncate operations deal with frontend buffer cache
2159 * buffers and frontend-owned in-memory records synchronously.
2161 if (ip->flags & HAMMER_INODE_TRUNCATED) {
2162 if (hammer_cursor_ondisk(&cursor) ||
2163 cursor.iprec->flush_state == HAMMER_FST_FLUSH) {
2164 if (ip->trunc_off <= rec_offset)
2166 else if (ip->trunc_off < rec_offset + n)
2167 n = (int)(ip->trunc_off - rec_offset);
2170 if (ip->sync_flags & HAMMER_INODE_TRUNCATED) {
2171 if (hammer_cursor_ondisk(&cursor)) {
2172 if (ip->sync_trunc_off <= rec_offset)
2174 else if (ip->sync_trunc_off < rec_offset + n)
2175 n = (int)(ip->sync_trunc_off - rec_offset);
2180 * Try to issue a direct read into our bio if possible,
2181 * otherwise resolve the element data into a hammer_buffer
2184 * The buffer on-disk should be zerod past any real
2185 * truncation point, but may not be for any synthesized
2186 * truncation point from above.
2188 if (boff == 0 && n == bp->b_bufsize &&
2189 ((cursor.leaf->data_offset + roff) & HAMMER_BUFMASK) == 0) {
2190 disk_offset = hammer_blockmap_lookup(
2192 cursor.leaf->data_offset + roff,
2196 nbio->bio_offset = disk_offset;
2197 error = hammer_io_direct_read(trans.hmp, nbio);
2200 error = hammer_ip_resolve_data(&cursor);
2202 bcopy((char *)cursor.data + roff,
2203 (char *)bp->b_data + boff, n);
2210 * Iterate until we have filled the request.
2213 if (boff == bp->b_bufsize)
2215 error = hammer_ip_next(&cursor);
2219 * There may have been a gap after the last record
2221 if (error == ENOENT)
2223 if (error == 0 && boff != bp->b_bufsize) {
2224 KKASSERT(boff < bp->b_bufsize);
2225 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
2226 /* boff = bp->b_bufsize; */
2229 bp->b_error = error;
2231 bp->b_flags |= B_ERROR;
2236 hammer_cache_node(&ip->cache[1], cursor.node);
2237 hammer_done_cursor(&cursor);
2238 hammer_done_transaction(&trans);
2243 * BMAP operation - used to support cluster_read() only.
2245 * (struct vnode *vp, off_t loffset, off_t *doffsetp, int *runp, int *runb)
2247 * This routine may return EOPNOTSUPP if the opration is not supported for
2248 * the specified offset. The contents of the pointer arguments do not
2249 * need to be initialized in that case.
2251 * If a disk address is available and properly aligned return 0 with
2252 * *doffsetp set to the zone-2 address, and *runp / *runb set appropriately
2253 * to the run-length relative to that offset. Callers may assume that
2254 * *doffsetp is valid if 0 is returned, even if *runp is not sufficiently
2255 * large, so return EOPNOTSUPP if it is not sufficiently large.
2259 hammer_vop_bmap(struct vop_bmap_args *ap)
2261 struct hammer_transaction trans;
2262 struct hammer_inode *ip;
2263 struct hammer_cursor cursor;
2264 hammer_base_elm_t base;
2268 int64_t base_offset;
2269 int64_t base_disk_offset;
2270 int64_t last_offset;
2271 hammer_off_t last_disk_offset;
2272 hammer_off_t disk_offset;
2277 ip = ap->a_vp->v_data;
2280 * We can only BMAP regular files. We can't BMAP database files,
2283 if (ip->ino_data.obj_type != HAMMER_OBJTYPE_REGFILE)
2287 * bmap is typically called with runp/runb both NULL when used
2288 * for writing. We do not support BMAP for writing atm.
2290 if (ap->a_cmd != BUF_CMD_READ)
2294 * Scan the B-Tree to acquire blockmap addresses, then translate
2297 hammer_simple_transaction(&trans, ip->hmp);
2299 kprintf("bmap_beg %016llx ip->cache %p\n", ap->a_loffset, ip->cache[1]);
2301 hammer_init_cursor(&trans, &cursor, &ip->cache[1], ip);
2304 * Key range (begin and end inclusive) to scan. Note that the key's
2305 * stored in the actual records represent BASE+LEN, not BASE. The
2306 * first record containing bio_offset will have a key > bio_offset.
2308 cursor.key_beg.localization = ip->obj_localization +
2309 HAMMER_LOCALIZE_MISC;
2310 cursor.key_beg.obj_id = ip->obj_id;
2311 cursor.key_beg.create_tid = 0;
2312 cursor.key_beg.delete_tid = 0;
2313 cursor.key_beg.obj_type = 0;
2315 cursor.key_beg.key = ap->a_loffset - MAXPHYS + 1;
2317 cursor.key_beg.key = ap->a_loffset + 1;
2318 if (cursor.key_beg.key < 0)
2319 cursor.key_beg.key = 0;
2320 cursor.asof = ip->obj_asof;
2321 cursor.flags |= HAMMER_CURSOR_ASOF;
2323 cursor.key_end = cursor.key_beg;
2324 KKASSERT(ip->ino_data.obj_type == HAMMER_OBJTYPE_REGFILE);
2326 ran_end = ap->a_loffset + MAXPHYS;
2327 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
2328 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
2329 tmp64 = ran_end + MAXPHYS + 1; /* work-around GCC-4 bug */
2330 if (tmp64 < ran_end)
2331 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
2333 cursor.key_end.key = ran_end + MAXPHYS + 1;
2335 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
2337 error = hammer_ip_first(&cursor);
2338 base_offset = last_offset = 0;
2339 base_disk_offset = last_disk_offset = 0;
2341 while (error == 0) {
2343 * Get the base file offset of the record. The key for
2344 * data records is (base + bytes) rather then (base).
2346 * NOTE: rec_offset + rec_len may exceed the end-of-file.
2347 * The extra bytes should be zero on-disk and the BMAP op
2348 * should still be ok.
2350 base = &cursor.leaf->base;
2351 rec_offset = base->key - cursor.leaf->data_len;
2352 rec_len = cursor.leaf->data_len;
2355 * Incorporate any cached truncation.
2357 * NOTE: Modifications to rec_len based on synthesized
2358 * truncation points remove the guarantee that any extended
2359 * data on disk is zero (since the truncations may not have
2360 * taken place on-media yet).
2362 if (ip->flags & HAMMER_INODE_TRUNCATED) {
2363 if (hammer_cursor_ondisk(&cursor) ||
2364 cursor.iprec->flush_state == HAMMER_FST_FLUSH) {
2365 if (ip->trunc_off <= rec_offset)
2367 else if (ip->trunc_off < rec_offset + rec_len)
2368 rec_len = (int)(ip->trunc_off - rec_offset);
2371 if (ip->sync_flags & HAMMER_INODE_TRUNCATED) {
2372 if (hammer_cursor_ondisk(&cursor)) {
2373 if (ip->sync_trunc_off <= rec_offset)
2375 else if (ip->sync_trunc_off < rec_offset + rec_len)
2376 rec_len = (int)(ip->sync_trunc_off - rec_offset);
2381 * Accumulate information. If we have hit a discontiguous
2382 * block reset base_offset unless we are already beyond the
2383 * requested offset. If we are, that's it, we stop.
2385 disk_offset = hammer_blockmap_lookup(trans.hmp,
2386 cursor.leaf->data_offset,
2390 if (rec_offset != last_offset ||
2391 disk_offset != last_disk_offset) {
2392 if (rec_offset > ap->a_loffset)
2394 base_offset = rec_offset;
2395 base_disk_offset = disk_offset;
2397 last_offset = rec_offset + rec_len;
2398 last_disk_offset = disk_offset + rec_len;
2400 error = hammer_ip_next(&cursor);
2404 kprintf("BMAP %016llx: %016llx - %016llx\n",
2405 ap->a_loffset, base_offset, last_offset);
2406 kprintf("BMAP %16s: %016llx - %016llx\n",
2407 "", base_disk_offset, last_disk_offset);
2411 hammer_cache_node(&ip->cache[1], cursor.node);
2413 kprintf("bmap_end2 %016llx ip->cache %p\n", ap->a_loffset, ip->cache[1]);
2416 hammer_done_cursor(&cursor);
2417 hammer_done_transaction(&trans);
2420 * If we couldn't find any records or the records we did find were
2421 * all behind the requested offset, return failure. A forward
2422 * truncation can leave a hole w/ no on-disk records.
2424 if (last_offset == 0 || last_offset < ap->a_loffset)
2425 return (EOPNOTSUPP);
2428 * Figure out the block size at the requested offset and adjust
2429 * our limits so the cluster_read() does not create inappropriately
2430 * sized buffer cache buffers.
2432 blksize = hammer_blocksize(ap->a_loffset);
2433 if (hammer_blocksize(base_offset) != blksize) {
2434 base_offset = hammer_blockdemarc(base_offset, ap->a_loffset);
2436 if (last_offset != ap->a_loffset &&
2437 hammer_blocksize(last_offset - 1) != blksize) {
2438 last_offset = hammer_blockdemarc(ap->a_loffset,
2443 * Returning EOPNOTSUPP simply prevents the direct-IO optimization
2446 disk_offset = base_disk_offset + (ap->a_loffset - base_offset);
2449 * If doffsetp is not aligned or the forward run size does
2450 * not cover a whole buffer, disallow the direct I/O.
2452 if ((disk_offset & HAMMER_BUFMASK) ||
2453 (last_offset - ap->a_loffset) < blksize) {
2456 *ap->a_doffsetp = disk_offset;
2458 *ap->a_runb = ap->a_loffset - base_offset;
2459 KKASSERT(*ap->a_runb >= 0);
2462 *ap->a_runp = last_offset - ap->a_loffset;
2463 KKASSERT(*ap->a_runp >= 0);
2471 * Write to a regular file. Because this is a strategy call the OS is
2472 * trying to actually get data onto the media.
2476 hammer_vop_strategy_write(struct vop_strategy_args *ap)
2478 hammer_record_t record;
2489 ip = ap->a_vp->v_data;
2492 blksize = hammer_blocksize(bio->bio_offset);
2493 KKASSERT(bp->b_bufsize == blksize);
2495 if (ip->flags & HAMMER_INODE_RO) {
2496 bp->b_error = EROFS;
2497 bp->b_flags |= B_ERROR;
2503 * Interlock with inode destruction (no in-kernel or directory
2504 * topology visibility). If we queue new IO while trying to
2505 * destroy the inode we can deadlock the vtrunc call in
2506 * hammer_inode_unloadable_check().
2508 if (ip->flags & (HAMMER_INODE_DELETING|HAMMER_INODE_DELETED)) {
2515 * Reserve space and issue a direct-write from the front-end.
2516 * NOTE: The direct_io code will hammer_bread/bcopy smaller
2519 * An in-memory record will be installed to reference the storage
2520 * until the flusher can get to it.
2522 * Since we own the high level bio the front-end will not try to
2523 * do a direct-read until the write completes.
2525 * NOTE: The only time we do not reserve a full-sized buffers
2526 * worth of data is if the file is small. We do not try to
2527 * allocate a fragment (from the small-data zone) at the end of
2528 * an otherwise large file as this can lead to wildly separated
2531 KKASSERT((bio->bio_offset & HAMMER_BUFMASK) == 0);
2532 KKASSERT(bio->bio_offset < ip->ino_data.size);
2533 if (bio->bio_offset || ip->ino_data.size > HAMMER_BUFSIZE / 2)
2534 bytes = bp->b_bufsize;
2536 bytes = ((int)ip->ino_data.size + 15) & ~15;
2538 record = hammer_ip_add_bulk(ip, bio->bio_offset, bp->b_data,
2541 hammer_io_direct_write(hmp, &record->leaf, bio);
2542 hammer_rel_mem_record(record);
2543 if (ip->rsv_recs > 1 && hmp->rsv_recs > hammer_limit_recs)
2544 hammer_flush_inode(ip, 0);
2546 bp->b_bio2.bio_offset = NOOFFSET;
2547 bp->b_error = error;
2548 bp->b_flags |= B_ERROR;
2555 * dounlink - disconnect a directory entry
2557 * XXX whiteout support not really in yet
2560 hammer_dounlink(hammer_transaction_t trans, struct nchandle *nch,
2561 struct vnode *dvp, struct ucred *cred, int flags)
2563 struct namecache *ncp;
2566 struct hammer_cursor cursor;
2571 * Calculate the namekey and setup the key range for the scan. This
2572 * works kinda like a chained hash table where the lower 32 bits
2573 * of the namekey synthesize the chain.
2575 * The key range is inclusive of both key_beg and key_end.
2580 if (dip->flags & HAMMER_INODE_RO)
2583 namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
2585 hammer_init_cursor(trans, &cursor, &dip->cache[1], dip);
2586 cursor.key_beg.localization = dip->obj_localization +
2587 HAMMER_LOCALIZE_MISC;
2588 cursor.key_beg.obj_id = dip->obj_id;
2589 cursor.key_beg.key = namekey;
2590 cursor.key_beg.create_tid = 0;
2591 cursor.key_beg.delete_tid = 0;
2592 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
2593 cursor.key_beg.obj_type = 0;
2595 cursor.key_end = cursor.key_beg;
2596 cursor.key_end.key |= 0xFFFFFFFFULL;
2597 cursor.asof = dip->obj_asof;
2598 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
2601 * Scan all matching records (the chain), locate the one matching
2602 * the requested path component. info->last_error contains the
2603 * error code on search termination and could be 0, ENOENT, or
2606 * The hammer_ip_*() functions merge in-memory records with on-disk
2607 * records for the purposes of the search.
2609 error = hammer_ip_first(&cursor);
2611 while (error == 0) {
2612 error = hammer_ip_resolve_data(&cursor);
2615 nlen = cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF;
2617 if (ncp->nc_nlen == nlen &&
2618 bcmp(ncp->nc_name, cursor.data->entry.name, nlen) == 0) {
2621 error = hammer_ip_next(&cursor);
2625 * If all is ok we have to get the inode so we can adjust nlinks.
2626 * To avoid a deadlock with the flusher we must release the inode
2627 * lock on the directory when acquiring the inode for the entry.
2629 * If the target is a directory, it must be empty.
2632 hammer_unlock(&cursor.ip->lock);
2633 ip = hammer_get_inode(trans, dip, cursor.data->entry.obj_id,
2635 cursor.data->entry.localization,
2637 hammer_lock_sh(&cursor.ip->lock);
2638 if (error == ENOENT) {
2639 kprintf("obj_id %016llx\n", cursor.data->entry.obj_id);
2640 Debugger("ENOENT unlinking object that should exist");
2644 * If we are trying to remove a directory the directory must
2647 * WARNING: hammer_ip_check_directory_empty() may have to
2648 * terminate the cursor to avoid a deadlock. It is ok to
2649 * call hammer_done_cursor() twice.
2651 if (error == 0 && ip->ino_data.obj_type ==
2652 HAMMER_OBJTYPE_DIRECTORY) {
2653 error = hammer_ip_check_directory_empty(trans, ip);
2657 * Delete the directory entry.
2659 * WARNING: hammer_ip_del_directory() may have to terminate
2660 * the cursor to avoid a deadlock. It is ok to call
2661 * hammer_done_cursor() twice.
2664 error = hammer_ip_del_directory(trans, &cursor,
2667 hammer_done_cursor(&cursor);
2669 cache_setunresolved(nch);
2670 cache_setvp(nch, NULL);
2673 cache_inval_vp(ip->vp, CINV_DESTROY);
2676 hammer_rel_inode(ip, 0);
2678 hammer_done_cursor(&cursor);
2680 if (error == EDEADLK)
2686 /************************************************************************
2687 * FIFO AND SPECFS OPS *
2688 ************************************************************************
2693 hammer_vop_fifoclose (struct vop_close_args *ap)
2695 /* XXX update itimes */
2696 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
2700 hammer_vop_fiforead (struct vop_read_args *ap)
2704 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2705 /* XXX update access time */
2710 hammer_vop_fifowrite (struct vop_write_args *ap)
2714 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2715 /* XXX update access time */
2720 hammer_vop_specclose (struct vop_close_args *ap)
2722 /* XXX update itimes */
2723 return (VOCALL(&spec_vnode_vops, &ap->a_head));
2727 hammer_vop_specread (struct vop_read_args *ap)
2729 /* XXX update access time */
2730 return (VOCALL(&spec_vnode_vops, &ap->a_head));
2734 hammer_vop_specwrite (struct vop_write_args *ap)
2736 /* XXX update last change time */
2737 return (VOCALL(&spec_vnode_vops, &ap->a_head));