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 1.83 2008/07/07 22:42:35 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;
725 * hammer_vop_nresolve { nch, dvp, cred }
727 * Locate the requested directory entry.
731 hammer_vop_nresolve(struct vop_nresolve_args *ap)
733 struct hammer_transaction trans;
734 struct namecache *ncp;
738 struct hammer_cursor cursor;
747 u_int32_t localization;
750 * Misc initialization, plus handle as-of name extensions. Look for
751 * the '@@' extension. Note that as-of files and directories cannot
754 dip = VTOI(ap->a_dvp);
755 ncp = ap->a_nch->ncp;
756 asof = dip->obj_asof;
761 hammer_simple_transaction(&trans, dip->hmp);
763 for (i = 0; i < nlen; ++i) {
764 if (ncp->nc_name[i] == '@' && ncp->nc_name[i+1] == '@') {
765 asof = hammer_str_to_tid(ncp->nc_name + i + 2);
766 flags |= HAMMER_INODE_RO;
773 * If there is no path component the time extension is relative to
777 ip = hammer_get_inode(&trans, dip, dip->obj_id,
778 asof, dip->obj_localization,
781 error = hammer_get_vnode(ip, &vp);
782 hammer_rel_inode(ip, 0);
788 cache_setvp(ap->a_nch, vp);
795 * Calculate the namekey and setup the key range for the scan. This
796 * works kinda like a chained hash table where the lower 32 bits
797 * of the namekey synthesize the chain.
799 * The key range is inclusive of both key_beg and key_end.
801 namekey = hammer_directory_namekey(ncp->nc_name, nlen);
803 error = hammer_init_cursor(&trans, &cursor, &dip->cache[1], dip);
804 cursor.key_beg.localization = dip->obj_localization +
805 HAMMER_LOCALIZE_MISC;
806 cursor.key_beg.obj_id = dip->obj_id;
807 cursor.key_beg.key = namekey;
808 cursor.key_beg.create_tid = 0;
809 cursor.key_beg.delete_tid = 0;
810 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
811 cursor.key_beg.obj_type = 0;
813 cursor.key_end = cursor.key_beg;
814 cursor.key_end.key |= 0xFFFFFFFFULL;
816 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
819 * Scan all matching records (the chain), locate the one matching
820 * the requested path component.
822 * The hammer_ip_*() functions merge in-memory records with on-disk
823 * records for the purposes of the search.
826 localization = HAMMER_DEF_LOCALIZATION;
829 error = hammer_ip_first(&cursor);
831 error = hammer_ip_resolve_data(&cursor);
834 if (nlen == cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF &&
835 bcmp(ncp->nc_name, cursor.data->entry.name, nlen) == 0) {
836 obj_id = cursor.data->entry.obj_id;
839 * Force relookups whenever a PFS root is
842 if (obj_id == HAMMER_OBJID_ROOT)
844 localization = cursor.data->entry.localization;
847 error = hammer_ip_next(&cursor);
850 hammer_done_cursor(&cursor);
852 ip = hammer_get_inode(&trans, dip, obj_id,
855 if (ispfs && asof > ip->pfsm->pfsd.sync_end_tid) {
856 asof = ip->pfsm->pfsd.sync_end_tid;
857 hammer_rel_inode(ip, 0);
858 ip = hammer_get_inode(&trans, dip, obj_id,
865 error = hammer_get_vnode(ip, &vp);
866 hammer_rel_inode(ip, 0);
872 cache_setvp(ap->a_nch, vp);
874 cache_settimeout(ap->a_nch, 0);
877 } else if (error == ENOENT) {
878 cache_setvp(ap->a_nch, NULL);
881 hammer_done_transaction(&trans);
886 * hammer_vop_nlookupdotdot { dvp, vpp, cred }
888 * Locate the parent directory of a directory vnode.
890 * dvp is referenced but not locked. *vpp must be returned referenced and
891 * locked. A parent_obj_id of 0 does not necessarily indicate that we are
892 * at the root, instead it could indicate that the directory we were in was
895 * NOTE: as-of sequences are not linked into the directory structure. If
896 * we are at the root with a different asof then the mount point, reload
897 * the same directory with the mount point's asof. I'm not sure what this
898 * will do to NFS. We encode ASOF stamps in NFS file handles so it might not
899 * get confused, but it hasn't been tested.
903 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
905 struct hammer_transaction trans;
906 struct hammer_inode *dip;
907 struct hammer_inode *ip;
908 int64_t parent_obj_id;
909 u_int32_t parent_obj_localization;
913 dip = VTOI(ap->a_dvp);
914 asof = dip->obj_asof;
917 * Whos are parent? This could be the root of a pseudo-filesystem
918 * whos parent is in another localization domain.
920 parent_obj_id = dip->ino_data.parent_obj_id;
921 if (dip->obj_id == HAMMER_OBJID_ROOT)
922 parent_obj_localization = dip->ino_data.ext.obj.parent_obj_localization;
924 parent_obj_localization = dip->obj_localization;
926 if (parent_obj_id == 0) {
927 if (dip->obj_id == HAMMER_OBJID_ROOT &&
928 asof != dip->hmp->asof) {
929 parent_obj_id = dip->obj_id;
930 asof = dip->hmp->asof;
931 *ap->a_fakename = kmalloc(19, M_TEMP, M_WAITOK);
932 ksnprintf(*ap->a_fakename, 19, "0x%016llx",
940 hammer_simple_transaction(&trans, dip->hmp);
942 ip = hammer_get_inode(&trans, dip, parent_obj_id,
943 asof, parent_obj_localization,
946 error = hammer_get_vnode(ip, ap->a_vpp);
947 hammer_rel_inode(ip, 0);
951 hammer_done_transaction(&trans);
956 * hammer_vop_nlink { nch, dvp, vp, cred }
960 hammer_vop_nlink(struct vop_nlink_args *ap)
962 struct hammer_transaction trans;
963 struct hammer_inode *dip;
964 struct hammer_inode *ip;
965 struct nchandle *nch;
969 dip = VTOI(ap->a_dvp);
972 if (dip->flags & HAMMER_INODE_RO)
974 if (ip->flags & HAMMER_INODE_RO)
976 if ((error = hammer_checkspace(dip->hmp, HAMMER_CHECKSPACE_SLOP_CREATE)) != 0)
980 * Create a transaction to cover the operations we perform.
982 hammer_start_transaction(&trans, dip->hmp);
985 * Add the filesystem object to the directory. Note that neither
986 * dip nor ip are referenced or locked, but their vnodes are
987 * referenced. This function will bump the inode's link count.
989 error = hammer_ip_add_directory(&trans, dip,
990 nch->ncp->nc_name, nch->ncp->nc_nlen,
997 cache_setunresolved(nch);
998 cache_setvp(nch, ap->a_vp);
1000 hammer_done_transaction(&trans);
1005 * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
1007 * The operating system has already ensured that the directory entry
1008 * does not exist and done all appropriate namespace locking.
1012 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
1014 struct hammer_transaction trans;
1015 struct hammer_inode *dip;
1016 struct hammer_inode *nip;
1017 struct nchandle *nch;
1021 dip = VTOI(ap->a_dvp);
1023 if (dip->flags & HAMMER_INODE_RO)
1025 if ((error = hammer_checkspace(dip->hmp, HAMMER_CHECKSPACE_SLOP_CREATE)) != 0)
1029 * Create a transaction to cover the operations we perform.
1031 hammer_start_transaction(&trans, dip->hmp);
1034 * Create a new filesystem object of the requested type. The
1035 * returned inode will be referenced but not locked.
1037 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred,
1040 hkprintf("hammer_mkdir error %d\n", error);
1041 hammer_done_transaction(&trans);
1046 * Add the new filesystem object to the directory. This will also
1047 * bump the inode's link count.
1049 error = hammer_ip_add_directory(&trans, dip,
1050 nch->ncp->nc_name, nch->ncp->nc_nlen,
1053 hkprintf("hammer_mkdir (add) error %d\n", error);
1059 hammer_rel_inode(nip, 0);
1062 error = hammer_get_vnode(nip, ap->a_vpp);
1063 hammer_rel_inode(nip, 0);
1065 cache_setunresolved(ap->a_nch);
1066 cache_setvp(ap->a_nch, *ap->a_vpp);
1069 hammer_done_transaction(&trans);
1074 * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
1076 * The operating system has already ensured that the directory entry
1077 * does not exist and done all appropriate namespace locking.
1081 hammer_vop_nmknod(struct vop_nmknod_args *ap)
1083 struct hammer_transaction trans;
1084 struct hammer_inode *dip;
1085 struct hammer_inode *nip;
1086 struct nchandle *nch;
1091 dip = VTOI(ap->a_dvp);
1093 if (dip->flags & HAMMER_INODE_RO)
1095 if ((error = hammer_checkspace(dip->hmp, HAMMER_CHECKSPACE_SLOP_CREATE)) != 0)
1099 * Create a transaction to cover the operations we perform.
1101 hammer_start_transaction(&trans, dip->hmp);
1104 * Create a new filesystem object of the requested type. The
1105 * returned inode will be referenced but not locked.
1107 * If mknod specifies a directory a pseudo-fs is created.
1109 pseudofs = (ap->a_vap->va_type == VDIR);
1110 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred,
1111 dip, pseudofs, &nip);
1113 hammer_done_transaction(&trans);
1119 * Add the new filesystem object to the directory. This will also
1120 * bump the inode's link count.
1122 error = hammer_ip_add_directory(&trans, dip,
1123 nch->ncp->nc_name, nch->ncp->nc_nlen,
1130 hammer_rel_inode(nip, 0);
1133 error = hammer_get_vnode(nip, ap->a_vpp);
1134 hammer_rel_inode(nip, 0);
1136 cache_setunresolved(ap->a_nch);
1137 cache_setvp(ap->a_nch, *ap->a_vpp);
1140 hammer_done_transaction(&trans);
1145 * hammer_vop_open { vp, mode, cred, fp }
1149 hammer_vop_open(struct vop_open_args *ap)
1153 ip = VTOI(ap->a_vp);
1155 if ((ap->a_mode & FWRITE) && (ip->flags & HAMMER_INODE_RO))
1157 return(vop_stdopen(ap));
1161 * hammer_vop_pathconf { vp, name, retval }
1165 hammer_vop_pathconf(struct vop_pathconf_args *ap)
1171 * hammer_vop_print { vp }
1175 hammer_vop_print(struct vop_print_args *ap)
1181 * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
1185 hammer_vop_readdir(struct vop_readdir_args *ap)
1187 struct hammer_transaction trans;
1188 struct hammer_cursor cursor;
1189 struct hammer_inode *ip;
1191 hammer_base_elm_t base;
1199 ip = VTOI(ap->a_vp);
1201 saveoff = uio->uio_offset;
1203 if (ap->a_ncookies) {
1204 ncookies = uio->uio_resid / 16 + 1;
1205 if (ncookies > 1024)
1207 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1215 hammer_simple_transaction(&trans, ip->hmp);
1218 * Handle artificial entries
1222 r = vop_write_dirent(&error, uio, ip->obj_id, DT_DIR, 1, ".");
1226 cookies[cookie_index] = saveoff;
1229 if (cookie_index == ncookies)
1233 if (ip->ino_data.parent_obj_id) {
1234 r = vop_write_dirent(&error, uio,
1235 ip->ino_data.parent_obj_id,
1238 r = vop_write_dirent(&error, uio,
1239 ip->obj_id, DT_DIR, 2, "..");
1244 cookies[cookie_index] = saveoff;
1247 if (cookie_index == ncookies)
1252 * Key range (begin and end inclusive) to scan. Directory keys
1253 * directly translate to a 64 bit 'seek' position.
1255 hammer_init_cursor(&trans, &cursor, &ip->cache[1], ip);
1256 cursor.key_beg.localization = ip->obj_localization +
1257 HAMMER_LOCALIZE_MISC;
1258 cursor.key_beg.obj_id = ip->obj_id;
1259 cursor.key_beg.create_tid = 0;
1260 cursor.key_beg.delete_tid = 0;
1261 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1262 cursor.key_beg.obj_type = 0;
1263 cursor.key_beg.key = saveoff;
1265 cursor.key_end = cursor.key_beg;
1266 cursor.key_end.key = HAMMER_MAX_KEY;
1267 cursor.asof = ip->obj_asof;
1268 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1270 error = hammer_ip_first(&cursor);
1272 while (error == 0) {
1273 error = hammer_ip_resolve_data(&cursor);
1276 base = &cursor.leaf->base;
1277 saveoff = base->key;
1278 KKASSERT(cursor.leaf->data_len > HAMMER_ENTRY_NAME_OFF);
1280 if (base->obj_id != ip->obj_id)
1281 panic("readdir: bad record at %p", cursor.node);
1283 r = vop_write_dirent(
1284 &error, uio, cursor.data->entry.obj_id,
1285 hammer_get_dtype(cursor.leaf->base.obj_type),
1286 cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF ,
1287 (void *)cursor.data->entry.name);
1292 cookies[cookie_index] = base->key;
1294 if (cookie_index == ncookies)
1296 error = hammer_ip_next(&cursor);
1298 hammer_done_cursor(&cursor);
1301 hammer_done_transaction(&trans);
1304 *ap->a_eofflag = (error == ENOENT);
1305 uio->uio_offset = saveoff;
1306 if (error && cookie_index == 0) {
1307 if (error == ENOENT)
1310 kfree(cookies, M_TEMP);
1311 *ap->a_ncookies = 0;
1312 *ap->a_cookies = NULL;
1315 if (error == ENOENT)
1318 *ap->a_ncookies = cookie_index;
1319 *ap->a_cookies = cookies;
1326 * hammer_vop_readlink { vp, uio, cred }
1330 hammer_vop_readlink(struct vop_readlink_args *ap)
1332 struct hammer_transaction trans;
1333 struct hammer_cursor cursor;
1334 struct hammer_inode *ip;
1337 ip = VTOI(ap->a_vp);
1340 * Shortcut if the symlink data was stuffed into ino_data.
1342 if (ip->ino_data.size <= HAMMER_INODE_BASESYMLEN) {
1343 error = uiomove(ip->ino_data.ext.symlink,
1344 ip->ino_data.size, ap->a_uio);
1351 hammer_simple_transaction(&trans, ip->hmp);
1352 hammer_init_cursor(&trans, &cursor, &ip->cache[1], ip);
1355 * Key range (begin and end inclusive) to scan. Directory keys
1356 * directly translate to a 64 bit 'seek' position.
1358 cursor.key_beg.localization = ip->obj_localization +
1359 HAMMER_LOCALIZE_MISC;
1360 cursor.key_beg.obj_id = ip->obj_id;
1361 cursor.key_beg.create_tid = 0;
1362 cursor.key_beg.delete_tid = 0;
1363 cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
1364 cursor.key_beg.obj_type = 0;
1365 cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
1366 cursor.asof = ip->obj_asof;
1367 cursor.flags |= HAMMER_CURSOR_ASOF;
1369 error = hammer_ip_lookup(&cursor);
1371 error = hammer_ip_resolve_data(&cursor);
1373 KKASSERT(cursor.leaf->data_len >=
1374 HAMMER_SYMLINK_NAME_OFF);
1375 error = uiomove(cursor.data->symlink.name,
1376 cursor.leaf->data_len -
1377 HAMMER_SYMLINK_NAME_OFF,
1381 hammer_done_cursor(&cursor);
1382 hammer_done_transaction(&trans);
1387 * hammer_vop_nremove { nch, dvp, cred }
1391 hammer_vop_nremove(struct vop_nremove_args *ap)
1393 struct hammer_transaction trans;
1394 struct hammer_inode *dip;
1397 dip = VTOI(ap->a_dvp);
1399 if (hammer_nohistory(dip) == 0 &&
1400 (error = hammer_checkspace(dip->hmp, HAMMER_CHECKSPACE_SLOP_REMOVE)) != 0) {
1404 hammer_start_transaction(&trans, dip->hmp);
1405 error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1406 hammer_done_transaction(&trans);
1412 * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1416 hammer_vop_nrename(struct vop_nrename_args *ap)
1418 struct hammer_transaction trans;
1419 struct namecache *fncp;
1420 struct namecache *tncp;
1421 struct hammer_inode *fdip;
1422 struct hammer_inode *tdip;
1423 struct hammer_inode *ip;
1424 struct hammer_cursor cursor;
1428 fdip = VTOI(ap->a_fdvp);
1429 tdip = VTOI(ap->a_tdvp);
1430 fncp = ap->a_fnch->ncp;
1431 tncp = ap->a_tnch->ncp;
1432 ip = VTOI(fncp->nc_vp);
1433 KKASSERT(ip != NULL);
1435 if (fdip->flags & HAMMER_INODE_RO)
1437 if (tdip->flags & HAMMER_INODE_RO)
1439 if (ip->flags & HAMMER_INODE_RO)
1441 if ((error = hammer_checkspace(fdip->hmp, HAMMER_CHECKSPACE_SLOP_CREATE)) != 0)
1444 hammer_start_transaction(&trans, fdip->hmp);
1447 * Remove tncp from the target directory and then link ip as
1448 * tncp. XXX pass trans to dounlink
1450 * Force the inode sync-time to match the transaction so it is
1451 * in-sync with the creation of the target directory entry.
1453 error = hammer_dounlink(&trans, ap->a_tnch, ap->a_tdvp, ap->a_cred, 0);
1454 if (error == 0 || error == ENOENT) {
1455 error = hammer_ip_add_directory(&trans, tdip,
1456 tncp->nc_name, tncp->nc_nlen,
1459 ip->ino_data.parent_obj_id = tdip->obj_id;
1460 hammer_modify_inode(ip, HAMMER_INODE_DDIRTY);
1464 goto failed; /* XXX */
1467 * Locate the record in the originating directory and remove it.
1469 * Calculate the namekey and setup the key range for the scan. This
1470 * works kinda like a chained hash table where the lower 32 bits
1471 * of the namekey synthesize the chain.
1473 * The key range is inclusive of both key_beg and key_end.
1475 namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
1477 hammer_init_cursor(&trans, &cursor, &fdip->cache[1], fdip);
1478 cursor.key_beg.localization = fdip->obj_localization +
1479 HAMMER_LOCALIZE_MISC;
1480 cursor.key_beg.obj_id = fdip->obj_id;
1481 cursor.key_beg.key = namekey;
1482 cursor.key_beg.create_tid = 0;
1483 cursor.key_beg.delete_tid = 0;
1484 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1485 cursor.key_beg.obj_type = 0;
1487 cursor.key_end = cursor.key_beg;
1488 cursor.key_end.key |= 0xFFFFFFFFULL;
1489 cursor.asof = fdip->obj_asof;
1490 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1493 * Scan all matching records (the chain), locate the one matching
1494 * the requested path component.
1496 * The hammer_ip_*() functions merge in-memory records with on-disk
1497 * records for the purposes of the search.
1499 error = hammer_ip_first(&cursor);
1500 while (error == 0) {
1501 if (hammer_ip_resolve_data(&cursor) != 0)
1503 nlen = cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF;
1505 if (fncp->nc_nlen == nlen &&
1506 bcmp(fncp->nc_name, cursor.data->entry.name, nlen) == 0) {
1509 error = hammer_ip_next(&cursor);
1513 * If all is ok we have to get the inode so we can adjust nlinks.
1515 * WARNING: hammer_ip_del_directory() may have to terminate the
1516 * cursor to avoid a recursion. It's ok to call hammer_done_cursor()
1520 error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
1523 * XXX A deadlock here will break rename's atomicy for the purposes
1524 * of crash recovery.
1526 if (error == EDEADLK) {
1527 hammer_done_cursor(&cursor);
1532 * Cleanup and tell the kernel that the rename succeeded.
1534 hammer_done_cursor(&cursor);
1536 cache_rename(ap->a_fnch, ap->a_tnch);
1539 hammer_done_transaction(&trans);
1544 * hammer_vop_nrmdir { nch, dvp, cred }
1548 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1550 struct hammer_transaction trans;
1551 struct hammer_inode *dip;
1554 dip = VTOI(ap->a_dvp);
1556 if (hammer_nohistory(dip) == 0 &&
1557 (error = hammer_checkspace(dip->hmp, HAMMER_CHECKSPACE_SLOP_REMOVE)) != 0) {
1561 hammer_start_transaction(&trans, dip->hmp);
1562 error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1563 hammer_done_transaction(&trans);
1569 * hammer_vop_setattr { vp, vap, cred }
1573 hammer_vop_setattr(struct vop_setattr_args *ap)
1575 struct hammer_transaction trans;
1577 struct hammer_inode *ip;
1582 int64_t aligned_size;
1586 ip = ap->a_vp->v_data;
1589 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1591 if (ip->flags & HAMMER_INODE_RO)
1593 if (hammer_nohistory(ip) == 0 &&
1594 (error = hammer_checkspace(ip->hmp, HAMMER_CHECKSPACE_SLOP_REMOVE)) != 0) {
1598 hammer_start_transaction(&trans, ip->hmp);
1601 if (vap->va_flags != VNOVAL) {
1602 flags = ip->ino_data.uflags;
1603 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1604 hammer_to_unix_xid(&ip->ino_data.uid),
1607 if (ip->ino_data.uflags != flags) {
1608 ip->ino_data.uflags = flags;
1609 modflags |= HAMMER_INODE_DDIRTY;
1611 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1618 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1622 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
1623 mode_t cur_mode = ip->ino_data.mode;
1624 uid_t cur_uid = hammer_to_unix_xid(&ip->ino_data.uid);
1625 gid_t cur_gid = hammer_to_unix_xid(&ip->ino_data.gid);
1629 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
1631 &cur_uid, &cur_gid, &cur_mode);
1633 hammer_guid_to_uuid(&uuid_uid, cur_uid);
1634 hammer_guid_to_uuid(&uuid_gid, cur_gid);
1635 if (bcmp(&uuid_uid, &ip->ino_data.uid,
1636 sizeof(uuid_uid)) ||
1637 bcmp(&uuid_gid, &ip->ino_data.gid,
1638 sizeof(uuid_gid)) ||
1639 ip->ino_data.mode != cur_mode
1641 ip->ino_data.uid = uuid_uid;
1642 ip->ino_data.gid = uuid_gid;
1643 ip->ino_data.mode = cur_mode;
1645 modflags |= HAMMER_INODE_DDIRTY;
1648 while (vap->va_size != VNOVAL && ip->ino_data.size != vap->va_size) {
1649 switch(ap->a_vp->v_type) {
1651 if (vap->va_size == ip->ino_data.size)
1654 * XXX break atomicy, we can deadlock the backend
1655 * if we do not release the lock. Probably not a
1658 blksize = hammer_blocksize(vap->va_size);
1659 if (vap->va_size < ip->ino_data.size) {
1660 vtruncbuf(ap->a_vp, vap->va_size, blksize);
1663 vnode_pager_setsize(ap->a_vp, vap->va_size);
1666 ip->ino_data.size = vap->va_size;
1667 modflags |= HAMMER_INODE_DDIRTY;
1670 * on-media truncation is cached in the inode until
1671 * the inode is synchronized.
1674 hammer_ip_frontend_trunc(ip, vap->va_size);
1675 #ifdef DEBUG_TRUNCATE
1676 if (HammerTruncIp == NULL)
1679 if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1680 ip->flags |= HAMMER_INODE_TRUNCATED;
1681 ip->trunc_off = vap->va_size;
1682 #ifdef DEBUG_TRUNCATE
1683 if (ip == HammerTruncIp)
1684 kprintf("truncate1 %016llx\n", ip->trunc_off);
1686 } else if (ip->trunc_off > vap->va_size) {
1687 ip->trunc_off = vap->va_size;
1688 #ifdef DEBUG_TRUNCATE
1689 if (ip == HammerTruncIp)
1690 kprintf("truncate2 %016llx\n", ip->trunc_off);
1693 #ifdef DEBUG_TRUNCATE
1694 if (ip == HammerTruncIp)
1695 kprintf("truncate3 %016llx (ignored)\n", vap->va_size);
1701 * If truncating we have to clean out a portion of
1702 * the last block on-disk. We do this in the
1703 * front-end buffer cache.
1705 aligned_size = (vap->va_size + (blksize - 1)) &
1706 ~(int64_t)(blksize - 1);
1707 if (truncating && vap->va_size < aligned_size) {
1711 aligned_size -= blksize;
1713 offset = (int)vap->va_size & (blksize - 1);
1714 error = bread(ap->a_vp, aligned_size,
1716 hammer_ip_frontend_trunc(ip, aligned_size);
1718 bzero(bp->b_data + offset,
1722 kprintf("ERROR %d\n", error);
1728 if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1729 ip->flags |= HAMMER_INODE_TRUNCATED;
1730 ip->trunc_off = vap->va_size;
1731 } else if (ip->trunc_off > vap->va_size) {
1732 ip->trunc_off = vap->va_size;
1734 hammer_ip_frontend_trunc(ip, vap->va_size);
1735 ip->ino_data.size = vap->va_size;
1736 modflags |= HAMMER_INODE_DDIRTY;
1744 if (vap->va_atime.tv_sec != VNOVAL) {
1745 ip->ino_data.atime =
1746 hammer_timespec_to_time(&vap->va_atime);
1747 modflags |= HAMMER_INODE_ATIME;
1749 if (vap->va_mtime.tv_sec != VNOVAL) {
1750 ip->ino_data.mtime =
1751 hammer_timespec_to_time(&vap->va_mtime);
1752 modflags |= HAMMER_INODE_MTIME;
1754 if (vap->va_mode != (mode_t)VNOVAL) {
1755 mode_t cur_mode = ip->ino_data.mode;
1756 uid_t cur_uid = hammer_to_unix_xid(&ip->ino_data.uid);
1757 gid_t cur_gid = hammer_to_unix_xid(&ip->ino_data.gid);
1759 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
1760 cur_uid, cur_gid, &cur_mode);
1761 if (error == 0 && ip->ino_data.mode != cur_mode) {
1762 ip->ino_data.mode = cur_mode;
1763 modflags |= HAMMER_INODE_DDIRTY;
1768 hammer_modify_inode(ip, modflags);
1769 hammer_done_transaction(&trans);
1774 * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1778 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1780 struct hammer_transaction trans;
1781 struct hammer_inode *dip;
1782 struct hammer_inode *nip;
1783 struct nchandle *nch;
1784 hammer_record_t record;
1788 ap->a_vap->va_type = VLNK;
1791 dip = VTOI(ap->a_dvp);
1793 if (dip->flags & HAMMER_INODE_RO)
1795 if ((error = hammer_checkspace(dip->hmp, HAMMER_CHECKSPACE_SLOP_CREATE)) != 0)
1799 * Create a transaction to cover the operations we perform.
1801 hammer_start_transaction(&trans, dip->hmp);
1804 * Create a new filesystem object of the requested type. The
1805 * returned inode will be referenced but not locked.
1808 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred,
1811 hammer_done_transaction(&trans);
1817 * Add a record representing the symlink. symlink stores the link
1818 * as pure data, not a string, and is no \0 terminated.
1821 bytes = strlen(ap->a_target);
1823 if (bytes <= HAMMER_INODE_BASESYMLEN) {
1824 bcopy(ap->a_target, nip->ino_data.ext.symlink, bytes);
1826 record = hammer_alloc_mem_record(nip, bytes);
1827 record->type = HAMMER_MEM_RECORD_GENERAL;
1829 record->leaf.base.localization = nip->obj_localization +
1830 HAMMER_LOCALIZE_MISC;
1831 record->leaf.base.key = HAMMER_FIXKEY_SYMLINK;
1832 record->leaf.base.rec_type = HAMMER_RECTYPE_FIX;
1833 record->leaf.data_len = bytes;
1834 KKASSERT(HAMMER_SYMLINK_NAME_OFF == 0);
1835 bcopy(ap->a_target, record->data->symlink.name, bytes);
1836 error = hammer_ip_add_record(&trans, record);
1840 * Set the file size to the length of the link.
1843 nip->ino_data.size = bytes;
1844 hammer_modify_inode(nip, HAMMER_INODE_DDIRTY);
1848 error = hammer_ip_add_directory(&trans, dip, nch->ncp->nc_name,
1849 nch->ncp->nc_nlen, nip);
1855 hammer_rel_inode(nip, 0);
1858 error = hammer_get_vnode(nip, ap->a_vpp);
1859 hammer_rel_inode(nip, 0);
1861 cache_setunresolved(ap->a_nch);
1862 cache_setvp(ap->a_nch, *ap->a_vpp);
1865 hammer_done_transaction(&trans);
1870 * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1874 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1876 struct hammer_transaction trans;
1877 struct hammer_inode *dip;
1880 dip = VTOI(ap->a_dvp);
1882 if (hammer_nohistory(dip) == 0 &&
1883 (error = hammer_checkspace(dip->hmp, HAMMER_CHECKSPACE_SLOP_CREATE)) != 0) {
1887 hammer_start_transaction(&trans, dip->hmp);
1888 error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp,
1889 ap->a_cred, ap->a_flags);
1890 hammer_done_transaction(&trans);
1896 * hammer_vop_ioctl { vp, command, data, fflag, cred }
1900 hammer_vop_ioctl(struct vop_ioctl_args *ap)
1902 struct hammer_inode *ip = ap->a_vp->v_data;
1904 return(hammer_ioctl(ip, ap->a_command, ap->a_data,
1905 ap->a_fflag, ap->a_cred));
1910 hammer_vop_mountctl(struct vop_mountctl_args *ap)
1915 mp = ap->a_head.a_ops->head.vv_mount;
1918 case MOUNTCTL_SET_EXPORT:
1919 if (ap->a_ctllen != sizeof(struct export_args))
1921 error = hammer_vfs_export(mp, ap->a_op,
1922 (const struct export_args *)ap->a_ctl);
1925 error = journal_mountctl(ap);
1932 * hammer_vop_strategy { vp, bio }
1934 * Strategy call, used for regular file read & write only. Note that the
1935 * bp may represent a cluster.
1937 * To simplify operation and allow better optimizations in the future,
1938 * this code does not make any assumptions with regards to buffer alignment
1943 hammer_vop_strategy(struct vop_strategy_args *ap)
1948 bp = ap->a_bio->bio_buf;
1952 error = hammer_vop_strategy_read(ap);
1955 error = hammer_vop_strategy_write(ap);
1958 bp->b_error = error = EINVAL;
1959 bp->b_flags |= B_ERROR;
1967 * Read from a regular file. Iterate the related records and fill in the
1968 * BIO/BUF. Gaps are zero-filled.
1970 * The support code in hammer_object.c should be used to deal with mixed
1971 * in-memory and on-disk records.
1973 * NOTE: Can be called from the cluster code with an oversized buf.
1979 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1981 struct hammer_transaction trans;
1982 struct hammer_inode *ip;
1983 struct hammer_cursor cursor;
1984 hammer_base_elm_t base;
1985 hammer_off_t disk_offset;
1999 ip = ap->a_vp->v_data;
2002 * The zone-2 disk offset may have been set by the cluster code via
2003 * a BMAP operation, or else should be NOOFFSET.
2005 * Checking the high bits for a match against zone-2 should suffice.
2007 nbio = push_bio(bio);
2008 if ((nbio->bio_offset & HAMMER_OFF_ZONE_MASK) ==
2009 HAMMER_ZONE_RAW_BUFFER) {
2010 error = hammer_io_direct_read(ip->hmp, nbio);
2015 * Well, that sucked. Do it the hard way. If all the stars are
2016 * aligned we may still be able to issue a direct-read.
2018 hammer_simple_transaction(&trans, ip->hmp);
2019 hammer_init_cursor(&trans, &cursor, &ip->cache[1], ip);
2022 * Key range (begin and end inclusive) to scan. Note that the key's
2023 * stored in the actual records represent BASE+LEN, not BASE. The
2024 * first record containing bio_offset will have a key > bio_offset.
2026 cursor.key_beg.localization = ip->obj_localization +
2027 HAMMER_LOCALIZE_MISC;
2028 cursor.key_beg.obj_id = ip->obj_id;
2029 cursor.key_beg.create_tid = 0;
2030 cursor.key_beg.delete_tid = 0;
2031 cursor.key_beg.obj_type = 0;
2032 cursor.key_beg.key = bio->bio_offset + 1;
2033 cursor.asof = ip->obj_asof;
2034 cursor.flags |= HAMMER_CURSOR_ASOF;
2036 cursor.key_end = cursor.key_beg;
2037 KKASSERT(ip->ino_data.obj_type == HAMMER_OBJTYPE_REGFILE);
2039 if (ip->ino_data.obj_type == HAMMER_OBJTYPE_DBFILE) {
2040 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
2041 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
2042 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
2046 ran_end = bio->bio_offset + bp->b_bufsize;
2047 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
2048 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
2049 tmp64 = ran_end + MAXPHYS + 1; /* work-around GCC-4 bug */
2050 if (tmp64 < ran_end)
2051 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
2053 cursor.key_end.key = ran_end + MAXPHYS + 1;
2055 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
2057 error = hammer_ip_first(&cursor);
2060 while (error == 0) {
2062 * Get the base file offset of the record. The key for
2063 * data records is (base + bytes) rather then (base).
2065 base = &cursor.leaf->base;
2066 rec_offset = base->key - cursor.leaf->data_len;
2069 * Calculate the gap, if any, and zero-fill it.
2071 * n is the offset of the start of the record verses our
2072 * current seek offset in the bio.
2074 n = (int)(rec_offset - (bio->bio_offset + boff));
2076 if (n > bp->b_bufsize - boff)
2077 n = bp->b_bufsize - boff;
2078 bzero((char *)bp->b_data + boff, n);
2084 * Calculate the data offset in the record and the number
2085 * of bytes we can copy.
2087 * There are two degenerate cases. First, boff may already
2088 * be at bp->b_bufsize. Secondly, the data offset within
2089 * the record may exceed the record's size.
2093 n = cursor.leaf->data_len - roff;
2095 kprintf("strategy_read: bad n=%d roff=%d\n", n, roff);
2097 } else if (n > bp->b_bufsize - boff) {
2098 n = bp->b_bufsize - boff;
2102 * Deal with cached truncations. This cool bit of code
2103 * allows truncate()/ftruncate() to avoid having to sync
2106 * If the frontend is truncated then all backend records are
2107 * subject to the frontend's truncation.
2109 * If the backend is truncated then backend records on-disk
2110 * (but not in-memory) are subject to the backend's
2111 * truncation. In-memory records owned by the backend
2112 * represent data written after the truncation point on the
2113 * backend and must not be truncated.
2115 * Truncate operations deal with frontend buffer cache
2116 * buffers and frontend-owned in-memory records synchronously.
2118 if (ip->flags & HAMMER_INODE_TRUNCATED) {
2119 if (hammer_cursor_ondisk(&cursor) ||
2120 cursor.iprec->flush_state == HAMMER_FST_FLUSH) {
2121 if (ip->trunc_off <= rec_offset)
2123 else if (ip->trunc_off < rec_offset + n)
2124 n = (int)(ip->trunc_off - rec_offset);
2127 if (ip->sync_flags & HAMMER_INODE_TRUNCATED) {
2128 if (hammer_cursor_ondisk(&cursor)) {
2129 if (ip->sync_trunc_off <= rec_offset)
2131 else if (ip->sync_trunc_off < rec_offset + n)
2132 n = (int)(ip->sync_trunc_off - rec_offset);
2137 * Try to issue a direct read into our bio if possible,
2138 * otherwise resolve the element data into a hammer_buffer
2141 * The buffer on-disk should be zerod past any real
2142 * truncation point, but may not be for any synthesized
2143 * truncation point from above.
2145 if (boff == 0 && n == bp->b_bufsize &&
2146 ((cursor.leaf->data_offset + roff) & HAMMER_BUFMASK) == 0) {
2147 disk_offset = hammer_blockmap_lookup(
2149 cursor.leaf->data_offset + roff,
2153 nbio->bio_offset = disk_offset;
2154 error = hammer_io_direct_read(trans.hmp, nbio);
2157 error = hammer_ip_resolve_data(&cursor);
2159 bcopy((char *)cursor.data + roff,
2160 (char *)bp->b_data + boff, n);
2167 * Iterate until we have filled the request.
2170 if (boff == bp->b_bufsize)
2172 error = hammer_ip_next(&cursor);
2176 * There may have been a gap after the last record
2178 if (error == ENOENT)
2180 if (error == 0 && boff != bp->b_bufsize) {
2181 KKASSERT(boff < bp->b_bufsize);
2182 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
2183 /* boff = bp->b_bufsize; */
2186 bp->b_error = error;
2188 bp->b_flags |= B_ERROR;
2193 hammer_cache_node(&ip->cache[1], cursor.node);
2194 hammer_done_cursor(&cursor);
2195 hammer_done_transaction(&trans);
2200 * BMAP operation - used to support cluster_read() only.
2202 * (struct vnode *vp, off_t loffset, off_t *doffsetp, int *runp, int *runb)
2204 * This routine may return EOPNOTSUPP if the opration is not supported for
2205 * the specified offset. The contents of the pointer arguments do not
2206 * need to be initialized in that case.
2208 * If a disk address is available and properly aligned return 0 with
2209 * *doffsetp set to the zone-2 address, and *runp / *runb set appropriately
2210 * to the run-length relative to that offset. Callers may assume that
2211 * *doffsetp is valid if 0 is returned, even if *runp is not sufficiently
2212 * large, so return EOPNOTSUPP if it is not sufficiently large.
2216 hammer_vop_bmap(struct vop_bmap_args *ap)
2218 struct hammer_transaction trans;
2219 struct hammer_inode *ip;
2220 struct hammer_cursor cursor;
2221 hammer_base_elm_t base;
2225 int64_t base_offset;
2226 int64_t base_disk_offset;
2227 int64_t last_offset;
2228 hammer_off_t last_disk_offset;
2229 hammer_off_t disk_offset;
2234 ip = ap->a_vp->v_data;
2237 * We can only BMAP regular files. We can't BMAP database files,
2240 if (ip->ino_data.obj_type != HAMMER_OBJTYPE_REGFILE)
2244 * bmap is typically called with runp/runb both NULL when used
2245 * for writing. We do not support BMAP for writing atm.
2247 if (ap->a_cmd != BUF_CMD_READ)
2251 * Scan the B-Tree to acquire blockmap addresses, then translate
2254 hammer_simple_transaction(&trans, ip->hmp);
2256 kprintf("bmap_beg %016llx ip->cache %p\n", ap->a_loffset, ip->cache[1]);
2258 hammer_init_cursor(&trans, &cursor, &ip->cache[1], ip);
2261 * Key range (begin and end inclusive) to scan. Note that the key's
2262 * stored in the actual records represent BASE+LEN, not BASE. The
2263 * first record containing bio_offset will have a key > bio_offset.
2265 cursor.key_beg.localization = ip->obj_localization +
2266 HAMMER_LOCALIZE_MISC;
2267 cursor.key_beg.obj_id = ip->obj_id;
2268 cursor.key_beg.create_tid = 0;
2269 cursor.key_beg.delete_tid = 0;
2270 cursor.key_beg.obj_type = 0;
2272 cursor.key_beg.key = ap->a_loffset - MAXPHYS + 1;
2274 cursor.key_beg.key = ap->a_loffset + 1;
2275 if (cursor.key_beg.key < 0)
2276 cursor.key_beg.key = 0;
2277 cursor.asof = ip->obj_asof;
2278 cursor.flags |= HAMMER_CURSOR_ASOF;
2280 cursor.key_end = cursor.key_beg;
2281 KKASSERT(ip->ino_data.obj_type == HAMMER_OBJTYPE_REGFILE);
2283 ran_end = ap->a_loffset + MAXPHYS;
2284 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
2285 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
2286 tmp64 = ran_end + MAXPHYS + 1; /* work-around GCC-4 bug */
2287 if (tmp64 < ran_end)
2288 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
2290 cursor.key_end.key = ran_end + MAXPHYS + 1;
2292 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
2294 error = hammer_ip_first(&cursor);
2295 base_offset = last_offset = 0;
2296 base_disk_offset = last_disk_offset = 0;
2298 while (error == 0) {
2300 * Get the base file offset of the record. The key for
2301 * data records is (base + bytes) rather then (base).
2303 * NOTE: rec_offset + rec_len may exceed the end-of-file.
2304 * The extra bytes should be zero on-disk and the BMAP op
2305 * should still be ok.
2307 base = &cursor.leaf->base;
2308 rec_offset = base->key - cursor.leaf->data_len;
2309 rec_len = cursor.leaf->data_len;
2312 * Incorporate any cached truncation.
2314 * NOTE: Modifications to rec_len based on synthesized
2315 * truncation points remove the guarantee that any extended
2316 * data on disk is zero (since the truncations may not have
2317 * taken place on-media yet).
2319 if (ip->flags & HAMMER_INODE_TRUNCATED) {
2320 if (hammer_cursor_ondisk(&cursor) ||
2321 cursor.iprec->flush_state == HAMMER_FST_FLUSH) {
2322 if (ip->trunc_off <= rec_offset)
2324 else if (ip->trunc_off < rec_offset + rec_len)
2325 rec_len = (int)(ip->trunc_off - rec_offset);
2328 if (ip->sync_flags & HAMMER_INODE_TRUNCATED) {
2329 if (hammer_cursor_ondisk(&cursor)) {
2330 if (ip->sync_trunc_off <= rec_offset)
2332 else if (ip->sync_trunc_off < rec_offset + rec_len)
2333 rec_len = (int)(ip->sync_trunc_off - rec_offset);
2338 * Accumulate information. If we have hit a discontiguous
2339 * block reset base_offset unless we are already beyond the
2340 * requested offset. If we are, that's it, we stop.
2342 disk_offset = hammer_blockmap_lookup(trans.hmp,
2343 cursor.leaf->data_offset,
2347 if (rec_offset != last_offset ||
2348 disk_offset != last_disk_offset) {
2349 if (rec_offset > ap->a_loffset)
2351 base_offset = rec_offset;
2352 base_disk_offset = disk_offset;
2354 last_offset = rec_offset + rec_len;
2355 last_disk_offset = disk_offset + rec_len;
2357 error = hammer_ip_next(&cursor);
2361 kprintf("BMAP %016llx: %016llx - %016llx\n",
2362 ap->a_loffset, base_offset, last_offset);
2363 kprintf("BMAP %16s: %016llx - %016llx\n",
2364 "", base_disk_offset, last_disk_offset);
2368 hammer_cache_node(&ip->cache[1], cursor.node);
2370 kprintf("bmap_end2 %016llx ip->cache %p\n", ap->a_loffset, ip->cache[1]);
2373 hammer_done_cursor(&cursor);
2374 hammer_done_transaction(&trans);
2377 * If we couldn't find any records or the records we did find were
2378 * all behind the requested offset, return failure. A forward
2379 * truncation can leave a hole w/ no on-disk records.
2381 if (last_offset == 0 || last_offset < ap->a_loffset)
2382 return (EOPNOTSUPP);
2385 * Figure out the block size at the requested offset and adjust
2386 * our limits so the cluster_read() does not create inappropriately
2387 * sized buffer cache buffers.
2389 blksize = hammer_blocksize(ap->a_loffset);
2390 if (hammer_blocksize(base_offset) != blksize) {
2391 base_offset = hammer_blockdemarc(base_offset, ap->a_loffset);
2393 if (last_offset != ap->a_loffset &&
2394 hammer_blocksize(last_offset - 1) != blksize) {
2395 last_offset = hammer_blockdemarc(ap->a_loffset,
2400 * Returning EOPNOTSUPP simply prevents the direct-IO optimization
2403 disk_offset = base_disk_offset + (ap->a_loffset - base_offset);
2406 * If doffsetp is not aligned or the forward run size does
2407 * not cover a whole buffer, disallow the direct I/O.
2409 if ((disk_offset & HAMMER_BUFMASK) ||
2410 (last_offset - ap->a_loffset) < blksize) {
2413 *ap->a_doffsetp = disk_offset;
2415 *ap->a_runb = ap->a_loffset - base_offset;
2416 KKASSERT(*ap->a_runb >= 0);
2419 *ap->a_runp = last_offset - ap->a_loffset;
2420 KKASSERT(*ap->a_runp >= 0);
2428 * Write to a regular file. Because this is a strategy call the OS is
2429 * trying to actually get data onto the media.
2433 hammer_vop_strategy_write(struct vop_strategy_args *ap)
2435 hammer_record_t record;
2446 ip = ap->a_vp->v_data;
2449 blksize = hammer_blocksize(bio->bio_offset);
2450 KKASSERT(bp->b_bufsize == blksize);
2452 if (ip->flags & HAMMER_INODE_RO) {
2453 bp->b_error = EROFS;
2454 bp->b_flags |= B_ERROR;
2460 * Interlock with inode destruction (no in-kernel or directory
2461 * topology visibility). If we queue new IO while trying to
2462 * destroy the inode we can deadlock the vtrunc call in
2463 * hammer_inode_unloadable_check().
2465 if (ip->flags & (HAMMER_INODE_DELETING|HAMMER_INODE_DELETED)) {
2472 * Reserve space and issue a direct-write from the front-end.
2473 * NOTE: The direct_io code will hammer_bread/bcopy smaller
2476 * An in-memory record will be installed to reference the storage
2477 * until the flusher can get to it.
2479 * Since we own the high level bio the front-end will not try to
2480 * do a direct-read until the write completes.
2482 * NOTE: The only time we do not reserve a full-sized buffers
2483 * worth of data is if the file is small. We do not try to
2484 * allocate a fragment (from the small-data zone) at the end of
2485 * an otherwise large file as this can lead to wildly separated
2488 KKASSERT((bio->bio_offset & HAMMER_BUFMASK) == 0);
2489 KKASSERT(bio->bio_offset < ip->ino_data.size);
2490 if (bio->bio_offset || ip->ino_data.size > HAMMER_BUFSIZE / 2)
2491 bytes = bp->b_bufsize;
2493 bytes = ((int)ip->ino_data.size + 15) & ~15;
2495 record = hammer_ip_add_bulk(ip, bio->bio_offset, bp->b_data,
2498 hammer_io_direct_write(hmp, &record->leaf, bio);
2499 hammer_rel_mem_record(record);
2500 if (ip->rsv_recs > 1 && hmp->rsv_recs > hammer_limit_recs)
2501 hammer_flush_inode(ip, 0);
2503 bp->b_bio2.bio_offset = NOOFFSET;
2504 bp->b_error = error;
2505 bp->b_flags |= B_ERROR;
2512 * dounlink - disconnect a directory entry
2514 * XXX whiteout support not really in yet
2517 hammer_dounlink(hammer_transaction_t trans, struct nchandle *nch,
2518 struct vnode *dvp, struct ucred *cred, int flags)
2520 struct namecache *ncp;
2523 struct hammer_cursor cursor;
2528 * Calculate the namekey and setup the key range for the scan. This
2529 * works kinda like a chained hash table where the lower 32 bits
2530 * of the namekey synthesize the chain.
2532 * The key range is inclusive of both key_beg and key_end.
2537 if (dip->flags & HAMMER_INODE_RO)
2540 namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
2542 hammer_init_cursor(trans, &cursor, &dip->cache[1], dip);
2543 cursor.key_beg.localization = dip->obj_localization +
2544 HAMMER_LOCALIZE_MISC;
2545 cursor.key_beg.obj_id = dip->obj_id;
2546 cursor.key_beg.key = namekey;
2547 cursor.key_beg.create_tid = 0;
2548 cursor.key_beg.delete_tid = 0;
2549 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
2550 cursor.key_beg.obj_type = 0;
2552 cursor.key_end = cursor.key_beg;
2553 cursor.key_end.key |= 0xFFFFFFFFULL;
2554 cursor.asof = dip->obj_asof;
2555 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
2558 * Scan all matching records (the chain), locate the one matching
2559 * the requested path component. info->last_error contains the
2560 * error code on search termination and could be 0, ENOENT, or
2563 * The hammer_ip_*() functions merge in-memory records with on-disk
2564 * records for the purposes of the search.
2566 error = hammer_ip_first(&cursor);
2568 while (error == 0) {
2569 error = hammer_ip_resolve_data(&cursor);
2572 nlen = cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF;
2574 if (ncp->nc_nlen == nlen &&
2575 bcmp(ncp->nc_name, cursor.data->entry.name, nlen) == 0) {
2578 error = hammer_ip_next(&cursor);
2582 * If all is ok we have to get the inode so we can adjust nlinks.
2583 * To avoid a deadlock with the flusher we must release the inode
2584 * lock on the directory when acquiring the inode for the entry.
2586 * If the target is a directory, it must be empty.
2589 hammer_unlock(&cursor.ip->lock);
2590 ip = hammer_get_inode(trans, dip, cursor.data->entry.obj_id,
2592 cursor.data->entry.localization,
2594 hammer_lock_sh(&cursor.ip->lock);
2595 if (error == ENOENT) {
2596 kprintf("obj_id %016llx\n", cursor.data->entry.obj_id);
2597 Debugger("ENOENT unlinking object that should exist");
2601 * If we are trying to remove a directory the directory must
2604 * WARNING: hammer_ip_check_directory_empty() may have to
2605 * terminate the cursor to avoid a deadlock. It is ok to
2606 * call hammer_done_cursor() twice.
2608 if (error == 0 && ip->ino_data.obj_type ==
2609 HAMMER_OBJTYPE_DIRECTORY) {
2610 error = hammer_ip_check_directory_empty(trans, ip);
2614 * Delete the directory entry.
2616 * WARNING: hammer_ip_del_directory() may have to terminate
2617 * the cursor to avoid a deadlock. It is ok to call
2618 * hammer_done_cursor() twice.
2621 error = hammer_ip_del_directory(trans, &cursor,
2624 hammer_done_cursor(&cursor);
2626 cache_setunresolved(nch);
2627 cache_setvp(nch, NULL);
2630 cache_inval_vp(ip->vp, CINV_DESTROY);
2633 hammer_rel_inode(ip, 0);
2635 hammer_done_cursor(&cursor);
2637 if (error == EDEADLK)
2643 /************************************************************************
2644 * FIFO AND SPECFS OPS *
2645 ************************************************************************
2650 hammer_vop_fifoclose (struct vop_close_args *ap)
2652 /* XXX update itimes */
2653 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
2657 hammer_vop_fiforead (struct vop_read_args *ap)
2661 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2662 /* XXX update access time */
2667 hammer_vop_fifowrite (struct vop_write_args *ap)
2671 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2672 /* XXX update access time */
2677 hammer_vop_specclose (struct vop_close_args *ap)
2679 /* XXX update itimes */
2680 return (VOCALL(&spec_vnode_vops, &ap->a_head));
2684 hammer_vop_specread (struct vop_read_args *ap)
2686 /* XXX update access time */
2687 return (VOCALL(&spec_vnode_vops, &ap->a_head));
2691 hammer_vop_specwrite (struct vop_write_args *ap)
2693 /* XXX update last change time */
2694 return (VOCALL(&spec_vnode_vops, &ap->a_head));