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 1.67 2008/06/12 00:16:10 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);
162 static void hammer_cleanup_write_io(hammer_inode_t ip);
163 static void hammer_update_rsv_databufs(hammer_inode_t ip);
168 hammer_vop_vnoperate(struct vop_generic_args *)
170 return (VOCALL(&hammer_vnode_vops, ap));
175 * hammer_vop_fsync { vp, waitfor }
179 hammer_vop_fsync(struct vop_fsync_args *ap)
181 hammer_inode_t ip = VTOI(ap->a_vp);
183 vfsync(ap->a_vp, ap->a_waitfor, 1, NULL, NULL);
184 hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
185 if (ap->a_waitfor == MNT_WAIT)
186 hammer_wait_inode(ip);
191 * hammer_vop_read { vp, uio, ioflag, cred }
195 hammer_vop_read(struct vop_read_args *ap)
197 struct hammer_transaction trans;
206 if (ap->a_vp->v_type != VREG)
210 seqcount = ap->a_ioflag >> 16;
212 hammer_start_transaction(&trans, ip->hmp);
215 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
218 while (uio->uio_resid > 0 && uio->uio_offset < ip->ino_data.size) {
219 offset = uio->uio_offset & HAMMER_BUFMASK;
220 if (hammer_debug_cluster_enable) {
221 error = cluster_read(ap->a_vp, ip->ino_data.size,
222 uio->uio_offset - offset,
224 MAXBSIZE, seqcount, &bp);
226 error = bread(ap->a_vp, uio->uio_offset - offset,
227 HAMMER_BUFSIZE, &bp);
233 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
234 n = HAMMER_BUFSIZE - offset;
235 if (n > uio->uio_resid)
237 if (n > ip->ino_data.size - uio->uio_offset)
238 n = (int)(ip->ino_data.size - uio->uio_offset);
239 error = uiomove((char *)bp->b_data + offset, n, uio);
244 if ((ip->flags & HAMMER_INODE_RO) == 0 &&
245 (ip->hmp->mp->mnt_flag & MNT_NOATIME) == 0) {
246 ip->ino_leaf.atime = trans.time;
247 hammer_modify_inode(ip, HAMMER_INODE_ITIMES);
249 hammer_done_transaction(&trans);
254 * hammer_vop_write { vp, uio, ioflag, cred }
258 hammer_vop_write(struct vop_write_args *ap)
260 struct hammer_transaction trans;
261 struct hammer_inode *ip;
271 if (ap->a_vp->v_type != VREG)
276 if (ip->flags & HAMMER_INODE_RO)
280 * Create a transaction to cover the operations we perform.
282 hammer_start_transaction(&trans, ip->hmp);
288 if (ap->a_ioflag & IO_APPEND)
289 uio->uio_offset = ip->ino_data.size;
292 * Check for illegal write offsets. Valid range is 0...2^63-1.
294 * NOTE: the base_off assignment is required to work around what
295 * I consider to be a GCC-4 optimization bug.
297 if (uio->uio_offset < 0) {
298 hammer_done_transaction(&trans);
301 base_offset = uio->uio_offset + uio->uio_resid; /* work around gcc-4 */
302 if (uio->uio_resid > 0 && base_offset <= 0) {
303 hammer_done_transaction(&trans);
308 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
311 while (uio->uio_resid > 0) {
314 if ((error = hammer_checkspace(trans.hmp)) != 0)
318 * Do not allow HAMMER to blow out the buffer cache.
320 * Do not allow HAMMER to blow out system memory by
321 * accumulating too many records. Records are decoupled
322 * from the buffer cache.
324 * Always check at the beginning so separate writes are
325 * not able to bypass this code.
327 * WARNING: Cannot unlock vp when doing a NOCOPY write as
328 * part of a putpages operation. Doing so could cause us
329 * to deadlock against the VM system when we try to re-lock.
331 if ((count++ & 15) == 0) {
332 if (uio->uio_segflg != UIO_NOCOPY) {
334 if ((ap->a_ioflag & IO_NOBWILL) == 0)
337 if (ip->rsv_recs > hammer_limit_irecs)
338 hammer_wait_inode_recs(ip);
339 if (uio->uio_segflg != UIO_NOCOPY)
340 vn_lock(ap->a_vp, LK_EXCLUSIVE|LK_RETRY);
343 rel_offset = (int)(uio->uio_offset & HAMMER_BUFMASK);
344 base_offset = uio->uio_offset & ~HAMMER_BUFMASK64;
345 n = HAMMER_BUFSIZE - rel_offset;
346 if (n > uio->uio_resid)
348 if (uio->uio_offset + n > ip->ino_data.size) {
349 vnode_pager_setsize(ap->a_vp, uio->uio_offset + n);
353 if (uio->uio_segflg == UIO_NOCOPY) {
355 * Issuing a write with the same data backing the
356 * buffer. Instantiate the buffer to collect the
357 * backing vm pages, then read-in any missing bits.
359 * This case is used by vop_stdputpages().
361 bp = getblk(ap->a_vp, base_offset,
362 HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
363 if ((bp->b_flags & B_CACHE) == 0) {
365 error = bread(ap->a_vp, base_offset,
366 HAMMER_BUFSIZE, &bp);
368 } else if (rel_offset == 0 && uio->uio_resid >= HAMMER_BUFSIZE) {
370 * Even though we are entirely overwriting the buffer
371 * we may still have to zero it out to avoid a
372 * mmap/write visibility issue.
374 bp = getblk(ap->a_vp, base_offset,
375 HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
376 if ((bp->b_flags & B_CACHE) == 0)
378 } else if (base_offset >= ip->ino_data.size) {
380 * If the base offset of the buffer is beyond the
381 * file EOF, we don't have to issue a read.
383 bp = getblk(ap->a_vp, base_offset,
384 HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
388 * Partial overwrite, read in any missing bits then
389 * replace the portion being written.
391 error = bread(ap->a_vp, base_offset,
392 HAMMER_BUFSIZE, &bp);
397 error = uiomove((char *)bp->b_data + rel_offset,
402 * If we screwed up we have to undo any VM size changes we
408 vtruncbuf(ap->a_vp, ip->ino_data.size,
413 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
414 if (ip->ino_data.size < uio->uio_offset) {
415 ip->ino_data.size = uio->uio_offset;
416 flags = HAMMER_INODE_DDIRTY;
417 vnode_pager_setsize(ap->a_vp, ip->ino_data.size);
421 ip->ino_data.mtime = trans.time;
422 flags |= HAMMER_INODE_ITIMES | HAMMER_INODE_BUFS;
423 flags |= HAMMER_INODE_DDIRTY; /* XXX mtime */
424 hammer_modify_inode(ip, flags);
427 * Try to keep track of cached dirty data.
429 if ((bp->b_flags & B_DIRTY) == 0) {
431 ++ip->hmp->rsv_databufs;
435 * Final buffer disposition.
437 if (ap->a_ioflag & IO_SYNC) {
439 } else if (ap->a_ioflag & IO_DIRECT) {
442 } else if ((ap->a_ioflag >> 16) == IO_SEQMAX &&
443 (uio->uio_offset & HAMMER_BUFMASK) == 0) {
445 * If seqcount indicates sequential operation and
446 * we just finished filling a buffer, push it out
447 * now to prevent the buffer cache from becoming
448 * too full, which would trigger non-optimal
457 hammer_done_transaction(&trans);
462 * hammer_vop_access { vp, mode, cred }
466 hammer_vop_access(struct vop_access_args *ap)
468 struct hammer_inode *ip = VTOI(ap->a_vp);
473 uid = hammer_to_unix_xid(&ip->ino_data.uid);
474 gid = hammer_to_unix_xid(&ip->ino_data.gid);
476 error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
477 ip->ino_data.uflags);
482 * hammer_vop_advlock { vp, id, op, fl, flags }
486 hammer_vop_advlock(struct vop_advlock_args *ap)
488 struct hammer_inode *ip = VTOI(ap->a_vp);
490 return (lf_advlock(ap, &ip->advlock, ip->ino_data.size));
494 * hammer_vop_close { vp, fflag }
498 hammer_vop_close(struct vop_close_args *ap)
500 struct hammer_inode *ip = VTOI(ap->a_vp);
502 if (ap->a_vp->v_opencount == 1)
503 hammer_inode_waitreclaims(ip);
505 return (vop_stdclose(ap));
509 * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
511 * The operating system has already ensured that the directory entry
512 * does not exist and done all appropriate namespace locking.
516 hammer_vop_ncreate(struct vop_ncreate_args *ap)
518 struct hammer_transaction trans;
519 struct hammer_inode *dip;
520 struct hammer_inode *nip;
521 struct nchandle *nch;
525 dip = VTOI(ap->a_dvp);
527 if (dip->flags & HAMMER_INODE_RO)
529 if ((error = hammer_checkspace(dip->hmp)) != 0)
533 * Create a transaction to cover the operations we perform.
535 hammer_start_transaction(&trans, dip->hmp);
538 * Create a new filesystem object of the requested type. The
539 * returned inode will be referenced and shared-locked to prevent
540 * it from being moved to the flusher.
543 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
545 hkprintf("hammer_create_inode error %d\n", error);
546 hammer_done_transaction(&trans);
552 * Add the new filesystem object to the directory. This will also
553 * bump the inode's link count.
555 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
557 hkprintf("hammer_ip_add_directory error %d\n", error);
563 hammer_rel_inode(nip, 0);
564 hammer_done_transaction(&trans);
567 error = hammer_get_vnode(nip, ap->a_vpp);
568 hammer_done_transaction(&trans);
569 hammer_rel_inode(nip, 0);
571 cache_setunresolved(ap->a_nch);
572 cache_setvp(ap->a_nch, *ap->a_vpp);
579 * hammer_vop_getattr { vp, vap }
581 * Retrieve an inode's attribute information. When accessing inodes
582 * historically we fake the atime field to ensure consistent results.
583 * The atime field is stored in the B-Tree element and allowed to be
584 * updated without cycling the element.
588 hammer_vop_getattr(struct vop_getattr_args *ap)
590 struct hammer_inode *ip = VTOI(ap->a_vp);
591 struct vattr *vap = ap->a_vap;
594 if (cache_check_fsmid_vp(ap->a_vp, &ip->fsmid) &&
595 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 &&
600 hammer_itimes(ap->a_vp);
603 vap->va_fsid = ip->hmp->fsid_udev;
604 vap->va_fileid = ip->ino_leaf.base.obj_id;
605 vap->va_mode = ip->ino_data.mode;
606 vap->va_nlink = ip->ino_data.nlinks;
607 vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
608 vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
611 vap->va_size = ip->ino_data.size;
612 if (ip->flags & HAMMER_INODE_RO)
613 hammer_to_timespec(ip->ino_data.mtime, &vap->va_atime);
615 hammer_to_timespec(ip->ino_leaf.atime, &vap->va_atime);
616 hammer_to_timespec(ip->ino_data.mtime, &vap->va_mtime);
617 hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
618 vap->va_flags = ip->ino_data.uflags;
619 vap->va_gen = 1; /* hammer inums are unique for all time */
620 vap->va_blocksize = HAMMER_BUFSIZE;
621 vap->va_bytes = (ip->ino_data.size + 63) & ~63;
622 vap->va_type = hammer_get_vnode_type(ip->ino_data.obj_type);
623 vap->va_filerev = 0; /* XXX */
624 /* mtime uniquely identifies any adjustments made to the file */
625 vap->va_fsmid = ip->ino_data.mtime;
626 vap->va_uid_uuid = ip->ino_data.uid;
627 vap->va_gid_uuid = ip->ino_data.gid;
628 vap->va_fsid_uuid = ip->hmp->fsid;
629 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
632 switch (ip->ino_data.obj_type) {
633 case HAMMER_OBJTYPE_CDEV:
634 case HAMMER_OBJTYPE_BDEV:
635 vap->va_rmajor = ip->ino_data.rmajor;
636 vap->va_rminor = ip->ino_data.rminor;
646 * hammer_vop_nresolve { nch, dvp, cred }
648 * Locate the requested directory entry.
652 hammer_vop_nresolve(struct vop_nresolve_args *ap)
654 struct hammer_transaction trans;
655 struct namecache *ncp;
659 struct hammer_cursor cursor;
669 * Misc initialization, plus handle as-of name extensions. Look for
670 * the '@@' extension. Note that as-of files and directories cannot
673 dip = VTOI(ap->a_dvp);
674 ncp = ap->a_nch->ncp;
675 asof = dip->obj_asof;
679 hammer_simple_transaction(&trans, dip->hmp);
681 for (i = 0; i < nlen; ++i) {
682 if (ncp->nc_name[i] == '@' && ncp->nc_name[i+1] == '@') {
683 asof = hammer_str_to_tid(ncp->nc_name + i + 2);
684 flags |= HAMMER_INODE_RO;
691 * If there is no path component the time extension is relative to
695 ip = hammer_get_inode(&trans, &dip->cache[1], dip->obj_id,
696 asof, flags, &error);
698 error = hammer_get_vnode(ip, &vp);
699 hammer_rel_inode(ip, 0);
705 cache_setvp(ap->a_nch, vp);
712 * Calculate the namekey and setup the key range for the scan. This
713 * works kinda like a chained hash table where the lower 32 bits
714 * of the namekey synthesize the chain.
716 * The key range is inclusive of both key_beg and key_end.
718 namekey = hammer_directory_namekey(ncp->nc_name, nlen);
720 error = hammer_init_cursor(&trans, &cursor, &dip->cache[0], dip);
721 cursor.key_beg.localization = HAMMER_LOCALIZE_MISC;
722 cursor.key_beg.obj_id = dip->obj_id;
723 cursor.key_beg.key = namekey;
724 cursor.key_beg.create_tid = 0;
725 cursor.key_beg.delete_tid = 0;
726 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
727 cursor.key_beg.obj_type = 0;
729 cursor.key_end = cursor.key_beg;
730 cursor.key_end.key |= 0xFFFFFFFFULL;
732 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
735 * Scan all matching records (the chain), locate the one matching
736 * the requested path component.
738 * The hammer_ip_*() functions merge in-memory records with on-disk
739 * records for the purposes of the search.
744 error = hammer_ip_first(&cursor);
746 error = hammer_ip_resolve_data(&cursor);
749 if (nlen == cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF &&
750 bcmp(ncp->nc_name, cursor.data->entry.name, nlen) == 0) {
751 obj_id = cursor.data->entry.obj_id;
754 error = hammer_ip_next(&cursor);
757 hammer_done_cursor(&cursor);
759 ip = hammer_get_inode(&trans, &dip->cache[1],
760 obj_id, asof, flags, &error);
762 error = hammer_get_vnode(ip, &vp);
763 hammer_rel_inode(ip, 0);
769 cache_setvp(ap->a_nch, vp);
772 } else if (error == ENOENT) {
773 cache_setvp(ap->a_nch, NULL);
776 hammer_done_transaction(&trans);
781 * hammer_vop_nlookupdotdot { dvp, vpp, cred }
783 * Locate the parent directory of a directory vnode.
785 * dvp is referenced but not locked. *vpp must be returned referenced and
786 * locked. A parent_obj_id of 0 does not necessarily indicate that we are
787 * at the root, instead it could indicate that the directory we were in was
790 * NOTE: as-of sequences are not linked into the directory structure. If
791 * we are at the root with a different asof then the mount point, reload
792 * the same directory with the mount point's asof. I'm not sure what this
793 * will do to NFS. We encode ASOF stamps in NFS file handles so it might not
794 * get confused, but it hasn't been tested.
798 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
800 struct hammer_transaction trans;
801 struct hammer_inode *dip;
802 struct hammer_inode *ip;
803 int64_t parent_obj_id;
807 dip = VTOI(ap->a_dvp);
808 asof = dip->obj_asof;
809 parent_obj_id = dip->ino_data.parent_obj_id;
811 if (parent_obj_id == 0) {
812 if (dip->obj_id == HAMMER_OBJID_ROOT &&
813 asof != dip->hmp->asof) {
814 parent_obj_id = dip->obj_id;
815 asof = dip->hmp->asof;
816 *ap->a_fakename = kmalloc(19, M_TEMP, M_WAITOK);
817 ksnprintf(*ap->a_fakename, 19, "0x%016llx",
825 hammer_simple_transaction(&trans, dip->hmp);
827 ip = hammer_get_inode(&trans, &dip->cache[1], parent_obj_id,
828 asof, dip->flags, &error);
830 error = hammer_get_vnode(ip, ap->a_vpp);
831 hammer_rel_inode(ip, 0);
835 hammer_done_transaction(&trans);
840 * hammer_vop_nlink { nch, dvp, vp, cred }
844 hammer_vop_nlink(struct vop_nlink_args *ap)
846 struct hammer_transaction trans;
847 struct hammer_inode *dip;
848 struct hammer_inode *ip;
849 struct nchandle *nch;
853 dip = VTOI(ap->a_dvp);
856 if (dip->flags & HAMMER_INODE_RO)
858 if (ip->flags & HAMMER_INODE_RO)
860 if ((error = hammer_checkspace(dip->hmp)) != 0)
864 * Create a transaction to cover the operations we perform.
866 hammer_start_transaction(&trans, dip->hmp);
869 * Add the filesystem object to the directory. Note that neither
870 * dip nor ip are referenced or locked, but their vnodes are
871 * referenced. This function will bump the inode's link count.
873 error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
879 cache_setunresolved(nch);
880 cache_setvp(nch, ap->a_vp);
882 hammer_done_transaction(&trans);
887 * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
889 * The operating system has already ensured that the directory entry
890 * does not exist and done all appropriate namespace locking.
894 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
896 struct hammer_transaction trans;
897 struct hammer_inode *dip;
898 struct hammer_inode *nip;
899 struct nchandle *nch;
903 dip = VTOI(ap->a_dvp);
905 if (dip->flags & HAMMER_INODE_RO)
907 if ((error = hammer_checkspace(dip->hmp)) != 0)
911 * Create a transaction to cover the operations we perform.
913 hammer_start_transaction(&trans, dip->hmp);
916 * Create a new filesystem object of the requested type. The
917 * returned inode will be referenced but not locked.
919 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
921 hkprintf("hammer_mkdir error %d\n", error);
922 hammer_done_transaction(&trans);
927 * Add the new filesystem object to the directory. This will also
928 * bump the inode's link count.
930 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
932 hkprintf("hammer_mkdir (add) error %d\n", error);
938 hammer_rel_inode(nip, 0);
941 error = hammer_get_vnode(nip, ap->a_vpp);
942 hammer_rel_inode(nip, 0);
944 cache_setunresolved(ap->a_nch);
945 cache_setvp(ap->a_nch, *ap->a_vpp);
948 hammer_done_transaction(&trans);
953 * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
955 * The operating system has already ensured that the directory entry
956 * does not exist and done all appropriate namespace locking.
960 hammer_vop_nmknod(struct vop_nmknod_args *ap)
962 struct hammer_transaction trans;
963 struct hammer_inode *dip;
964 struct hammer_inode *nip;
965 struct nchandle *nch;
969 dip = VTOI(ap->a_dvp);
971 if (dip->flags & HAMMER_INODE_RO)
973 if ((error = hammer_checkspace(dip->hmp)) != 0)
977 * Create a transaction to cover the operations we perform.
979 hammer_start_transaction(&trans, dip->hmp);
982 * Create a new filesystem object of the requested type. The
983 * returned inode will be referenced but not locked.
985 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
987 hammer_done_transaction(&trans);
993 * Add the new filesystem object to the directory. This will also
994 * bump the inode's link count.
996 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
1002 hammer_rel_inode(nip, 0);
1005 error = hammer_get_vnode(nip, ap->a_vpp);
1006 hammer_rel_inode(nip, 0);
1008 cache_setunresolved(ap->a_nch);
1009 cache_setvp(ap->a_nch, *ap->a_vpp);
1012 hammer_done_transaction(&trans);
1017 * hammer_vop_open { vp, mode, cred, fp }
1021 hammer_vop_open(struct vop_open_args *ap)
1025 ip = VTOI(ap->a_vp);
1027 if ((ap->a_mode & FWRITE) && (ip->flags & HAMMER_INODE_RO))
1029 return(vop_stdopen(ap));
1033 * hammer_vop_pathconf { vp, name, retval }
1037 hammer_vop_pathconf(struct vop_pathconf_args *ap)
1043 * hammer_vop_print { vp }
1047 hammer_vop_print(struct vop_print_args *ap)
1053 * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
1057 hammer_vop_readdir(struct vop_readdir_args *ap)
1059 struct hammer_transaction trans;
1060 struct hammer_cursor cursor;
1061 struct hammer_inode *ip;
1063 hammer_base_elm_t base;
1071 ip = VTOI(ap->a_vp);
1073 saveoff = uio->uio_offset;
1075 if (ap->a_ncookies) {
1076 ncookies = uio->uio_resid / 16 + 1;
1077 if (ncookies > 1024)
1079 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1087 hammer_simple_transaction(&trans, ip->hmp);
1090 * Handle artificial entries
1094 r = vop_write_dirent(&error, uio, ip->obj_id, DT_DIR, 1, ".");
1098 cookies[cookie_index] = saveoff;
1101 if (cookie_index == ncookies)
1105 if (ip->ino_data.parent_obj_id) {
1106 r = vop_write_dirent(&error, uio,
1107 ip->ino_data.parent_obj_id,
1110 r = vop_write_dirent(&error, uio,
1111 ip->obj_id, DT_DIR, 2, "..");
1116 cookies[cookie_index] = saveoff;
1119 if (cookie_index == ncookies)
1124 * Key range (begin and end inclusive) to scan. Directory keys
1125 * directly translate to a 64 bit 'seek' position.
1127 hammer_init_cursor(&trans, &cursor, &ip->cache[0], ip);
1128 cursor.key_beg.localization = HAMMER_LOCALIZE_MISC;
1129 cursor.key_beg.obj_id = ip->obj_id;
1130 cursor.key_beg.create_tid = 0;
1131 cursor.key_beg.delete_tid = 0;
1132 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1133 cursor.key_beg.obj_type = 0;
1134 cursor.key_beg.key = saveoff;
1136 cursor.key_end = cursor.key_beg;
1137 cursor.key_end.key = HAMMER_MAX_KEY;
1138 cursor.asof = ip->obj_asof;
1139 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1141 error = hammer_ip_first(&cursor);
1143 while (error == 0) {
1144 error = hammer_ip_resolve_data(&cursor);
1147 base = &cursor.leaf->base;
1148 saveoff = base->key;
1149 KKASSERT(cursor.leaf->data_len > HAMMER_ENTRY_NAME_OFF);
1151 if (base->obj_id != ip->obj_id)
1152 panic("readdir: bad record at %p", cursor.node);
1154 r = vop_write_dirent(
1155 &error, uio, cursor.data->entry.obj_id,
1156 hammer_get_dtype(cursor.leaf->base.obj_type),
1157 cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF ,
1158 (void *)cursor.data->entry.name);
1163 cookies[cookie_index] = base->key;
1165 if (cookie_index == ncookies)
1167 error = hammer_ip_next(&cursor);
1169 hammer_done_cursor(&cursor);
1172 hammer_done_transaction(&trans);
1175 *ap->a_eofflag = (error == ENOENT);
1176 uio->uio_offset = saveoff;
1177 if (error && cookie_index == 0) {
1178 if (error == ENOENT)
1181 kfree(cookies, M_TEMP);
1182 *ap->a_ncookies = 0;
1183 *ap->a_cookies = NULL;
1186 if (error == ENOENT)
1189 *ap->a_ncookies = cookie_index;
1190 *ap->a_cookies = cookies;
1197 * hammer_vop_readlink { vp, uio, cred }
1201 hammer_vop_readlink(struct vop_readlink_args *ap)
1203 struct hammer_transaction trans;
1204 struct hammer_cursor cursor;
1205 struct hammer_inode *ip;
1208 ip = VTOI(ap->a_vp);
1211 * Shortcut if the symlink data was stuffed into ino_data.
1213 if (ip->ino_data.size <= HAMMER_INODE_BASESYMLEN) {
1214 error = uiomove(ip->ino_data.ext.symlink,
1215 ip->ino_data.size, ap->a_uio);
1222 hammer_simple_transaction(&trans, ip->hmp);
1223 hammer_init_cursor(&trans, &cursor, &ip->cache[0], ip);
1226 * Key range (begin and end inclusive) to scan. Directory keys
1227 * directly translate to a 64 bit 'seek' position.
1229 cursor.key_beg.localization = HAMMER_LOCALIZE_MISC; /* XXX */
1230 cursor.key_beg.obj_id = ip->obj_id;
1231 cursor.key_beg.create_tid = 0;
1232 cursor.key_beg.delete_tid = 0;
1233 cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
1234 cursor.key_beg.obj_type = 0;
1235 cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
1236 cursor.asof = ip->obj_asof;
1237 cursor.flags |= HAMMER_CURSOR_ASOF;
1239 error = hammer_ip_lookup(&cursor);
1241 error = hammer_ip_resolve_data(&cursor);
1243 KKASSERT(cursor.leaf->data_len >=
1244 HAMMER_SYMLINK_NAME_OFF);
1245 error = uiomove(cursor.data->symlink.name,
1246 cursor.leaf->data_len -
1247 HAMMER_SYMLINK_NAME_OFF,
1251 hammer_done_cursor(&cursor);
1252 hammer_done_transaction(&trans);
1257 * hammer_vop_nremove { nch, dvp, cred }
1261 hammer_vop_nremove(struct vop_nremove_args *ap)
1263 struct hammer_transaction trans;
1264 struct hammer_inode *dip;
1267 dip = VTOI(ap->a_dvp);
1269 if (hammer_nohistory(dip) == 0 &&
1270 (error = hammer_checkspace(dip->hmp)) != 0) {
1274 hammer_start_transaction(&trans, dip->hmp);
1275 error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1276 hammer_done_transaction(&trans);
1282 * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1286 hammer_vop_nrename(struct vop_nrename_args *ap)
1288 struct hammer_transaction trans;
1289 struct namecache *fncp;
1290 struct namecache *tncp;
1291 struct hammer_inode *fdip;
1292 struct hammer_inode *tdip;
1293 struct hammer_inode *ip;
1294 struct hammer_cursor cursor;
1298 fdip = VTOI(ap->a_fdvp);
1299 tdip = VTOI(ap->a_tdvp);
1300 fncp = ap->a_fnch->ncp;
1301 tncp = ap->a_tnch->ncp;
1302 ip = VTOI(fncp->nc_vp);
1303 KKASSERT(ip != NULL);
1305 if (fdip->flags & HAMMER_INODE_RO)
1307 if (tdip->flags & HAMMER_INODE_RO)
1309 if (ip->flags & HAMMER_INODE_RO)
1311 if ((error = hammer_checkspace(fdip->hmp)) != 0)
1314 hammer_start_transaction(&trans, fdip->hmp);
1317 * Remove tncp from the target directory and then link ip as
1318 * tncp. XXX pass trans to dounlink
1320 * Force the inode sync-time to match the transaction so it is
1321 * in-sync with the creation of the target directory entry.
1323 error = hammer_dounlink(&trans, ap->a_tnch, ap->a_tdvp, ap->a_cred, 0);
1324 if (error == 0 || error == ENOENT) {
1325 error = hammer_ip_add_directory(&trans, tdip, tncp, ip);
1327 ip->ino_data.parent_obj_id = tdip->obj_id;
1328 hammer_modify_inode(ip, HAMMER_INODE_DDIRTY);
1332 goto failed; /* XXX */
1335 * Locate the record in the originating directory and remove it.
1337 * Calculate the namekey and setup the key range for the scan. This
1338 * works kinda like a chained hash table where the lower 32 bits
1339 * of the namekey synthesize the chain.
1341 * The key range is inclusive of both key_beg and key_end.
1343 namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
1345 hammer_init_cursor(&trans, &cursor, &fdip->cache[0], fdip);
1346 cursor.key_beg.localization = HAMMER_LOCALIZE_MISC;
1347 cursor.key_beg.obj_id = fdip->obj_id;
1348 cursor.key_beg.key = namekey;
1349 cursor.key_beg.create_tid = 0;
1350 cursor.key_beg.delete_tid = 0;
1351 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1352 cursor.key_beg.obj_type = 0;
1354 cursor.key_end = cursor.key_beg;
1355 cursor.key_end.key |= 0xFFFFFFFFULL;
1356 cursor.asof = fdip->obj_asof;
1357 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1360 * Scan all matching records (the chain), locate the one matching
1361 * the requested path component.
1363 * The hammer_ip_*() functions merge in-memory records with on-disk
1364 * records for the purposes of the search.
1366 error = hammer_ip_first(&cursor);
1367 while (error == 0) {
1368 if (hammer_ip_resolve_data(&cursor) != 0)
1370 nlen = cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF;
1372 if (fncp->nc_nlen == nlen &&
1373 bcmp(fncp->nc_name, cursor.data->entry.name, nlen) == 0) {
1376 error = hammer_ip_next(&cursor);
1380 * If all is ok we have to get the inode so we can adjust nlinks.
1382 * WARNING: hammer_ip_del_directory() may have to terminate the
1383 * cursor to avoid a recursion. It's ok to call hammer_done_cursor()
1387 error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
1390 * XXX A deadlock here will break rename's atomicy for the purposes
1391 * of crash recovery.
1393 if (error == EDEADLK) {
1394 hammer_done_cursor(&cursor);
1399 * Cleanup and tell the kernel that the rename succeeded.
1401 hammer_done_cursor(&cursor);
1403 cache_rename(ap->a_fnch, ap->a_tnch);
1406 hammer_done_transaction(&trans);
1411 * hammer_vop_nrmdir { nch, dvp, cred }
1415 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1417 struct hammer_transaction trans;
1418 struct hammer_inode *dip;
1421 dip = VTOI(ap->a_dvp);
1423 if (hammer_nohistory(dip) == 0 &&
1424 (error = hammer_checkspace(dip->hmp)) != 0) {
1428 hammer_start_transaction(&trans, dip->hmp);
1429 error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1430 hammer_done_transaction(&trans);
1436 * hammer_vop_setattr { vp, vap, cred }
1440 hammer_vop_setattr(struct vop_setattr_args *ap)
1442 struct hammer_transaction trans;
1444 struct hammer_inode *ip;
1452 ip = ap->a_vp->v_data;
1455 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1457 if (ip->flags & HAMMER_INODE_RO)
1459 if (hammer_nohistory(ip) == 0 &&
1460 (error = hammer_checkspace(ip->hmp)) != 0) {
1464 hammer_start_transaction(&trans, ip->hmp);
1467 if (vap->va_flags != VNOVAL) {
1468 flags = ip->ino_data.uflags;
1469 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1470 hammer_to_unix_xid(&ip->ino_data.uid),
1473 if (ip->ino_data.uflags != flags) {
1474 ip->ino_data.uflags = flags;
1475 modflags |= HAMMER_INODE_DDIRTY;
1477 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1484 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1488 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
1489 mode_t cur_mode = ip->ino_data.mode;
1490 uid_t cur_uid = hammer_to_unix_xid(&ip->ino_data.uid);
1491 gid_t cur_gid = hammer_to_unix_xid(&ip->ino_data.gid);
1495 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
1497 &cur_uid, &cur_gid, &cur_mode);
1499 hammer_guid_to_uuid(&uuid_uid, cur_uid);
1500 hammer_guid_to_uuid(&uuid_gid, cur_gid);
1501 if (bcmp(&uuid_uid, &ip->ino_data.uid,
1502 sizeof(uuid_uid)) ||
1503 bcmp(&uuid_gid, &ip->ino_data.gid,
1504 sizeof(uuid_gid)) ||
1505 ip->ino_data.mode != cur_mode
1507 ip->ino_data.uid = uuid_uid;
1508 ip->ino_data.gid = uuid_gid;
1509 ip->ino_data.mode = cur_mode;
1511 modflags |= HAMMER_INODE_DDIRTY;
1514 while (vap->va_size != VNOVAL && ip->ino_data.size != vap->va_size) {
1515 switch(ap->a_vp->v_type) {
1517 if (vap->va_size == ip->ino_data.size)
1520 * XXX break atomicy, we can deadlock the backend
1521 * if we do not release the lock. Probably not a
1524 if (vap->va_size < ip->ino_data.size) {
1525 vtruncbuf(ap->a_vp, vap->va_size,
1529 vnode_pager_setsize(ap->a_vp, vap->va_size);
1532 ip->ino_data.size = vap->va_size;
1533 modflags |= HAMMER_INODE_DDIRTY;
1534 aligned_size = (vap->va_size + HAMMER_BUFMASK) &
1538 * on-media truncation is cached in the inode until
1539 * the inode is synchronized.
1542 hammer_ip_frontend_trunc(ip, vap->va_size);
1543 hammer_update_rsv_databufs(ip);
1544 #ifdef DEBUG_TRUNCATE
1545 if (HammerTruncIp == NULL)
1548 if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1549 ip->flags |= HAMMER_INODE_TRUNCATED;
1550 ip->trunc_off = vap->va_size;
1551 #ifdef DEBUG_TRUNCATE
1552 if (ip == HammerTruncIp)
1553 kprintf("truncate1 %016llx\n", ip->trunc_off);
1555 } else if (ip->trunc_off > vap->va_size) {
1556 ip->trunc_off = vap->va_size;
1557 #ifdef DEBUG_TRUNCATE
1558 if (ip == HammerTruncIp)
1559 kprintf("truncate2 %016llx\n", ip->trunc_off);
1562 #ifdef DEBUG_TRUNCATE
1563 if (ip == HammerTruncIp)
1564 kprintf("truncate3 %016llx (ignored)\n", vap->va_size);
1570 * If truncating we have to clean out a portion of
1571 * the last block on-disk. We do this in the
1572 * front-end buffer cache.
1574 if (truncating && vap->va_size < aligned_size) {
1578 aligned_size -= HAMMER_BUFSIZE;
1580 offset = vap->va_size & HAMMER_BUFMASK;
1581 error = bread(ap->a_vp, aligned_size,
1582 HAMMER_BUFSIZE, &bp);
1583 hammer_ip_frontend_trunc(ip, aligned_size);
1585 bzero(bp->b_data + offset,
1586 HAMMER_BUFSIZE - offset);
1589 kprintf("ERROR %d\n", error);
1595 if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1596 ip->flags |= HAMMER_INODE_TRUNCATED;
1597 ip->trunc_off = vap->va_size;
1598 } else if (ip->trunc_off > vap->va_size) {
1599 ip->trunc_off = vap->va_size;
1601 hammer_ip_frontend_trunc(ip, vap->va_size);
1602 ip->ino_data.size = vap->va_size;
1603 modflags |= HAMMER_INODE_DDIRTY;
1611 if (vap->va_atime.tv_sec != VNOVAL) {
1612 ip->ino_leaf.atime =
1613 hammer_timespec_to_transid(&vap->va_atime);
1614 modflags |= HAMMER_INODE_ITIMES;
1616 if (vap->va_mtime.tv_sec != VNOVAL) {
1617 ip->ino_data.mtime =
1618 hammer_timespec_to_transid(&vap->va_mtime);
1619 modflags |= HAMMER_INODE_ITIMES;
1620 modflags |= HAMMER_INODE_DDIRTY; /* XXX mtime */
1622 if (vap->va_mode != (mode_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);
1627 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
1628 cur_uid, cur_gid, &cur_mode);
1629 if (error == 0 && ip->ino_data.mode != cur_mode) {
1630 ip->ino_data.mode = cur_mode;
1631 modflags |= HAMMER_INODE_DDIRTY;
1636 hammer_modify_inode(ip, modflags);
1637 hammer_done_transaction(&trans);
1642 * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1646 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1648 struct hammer_transaction trans;
1649 struct hammer_inode *dip;
1650 struct hammer_inode *nip;
1651 struct nchandle *nch;
1652 hammer_record_t record;
1656 ap->a_vap->va_type = VLNK;
1659 dip = VTOI(ap->a_dvp);
1661 if (dip->flags & HAMMER_INODE_RO)
1663 if ((error = hammer_checkspace(dip->hmp)) != 0)
1667 * Create a transaction to cover the operations we perform.
1669 hammer_start_transaction(&trans, dip->hmp);
1672 * Create a new filesystem object of the requested type. The
1673 * returned inode will be referenced but not locked.
1676 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
1678 hammer_done_transaction(&trans);
1684 * Add a record representing the symlink. symlink stores the link
1685 * as pure data, not a string, and is no \0 terminated.
1688 bytes = strlen(ap->a_target);
1690 if (bytes <= HAMMER_INODE_BASESYMLEN) {
1691 bcopy(ap->a_target, nip->ino_data.ext.symlink, bytes);
1693 record = hammer_alloc_mem_record(nip, bytes);
1694 record->type = HAMMER_MEM_RECORD_GENERAL;
1696 record->leaf.base.localization = HAMMER_LOCALIZE_MISC;
1697 record->leaf.base.key = HAMMER_FIXKEY_SYMLINK;
1698 record->leaf.base.rec_type = HAMMER_RECTYPE_FIX;
1699 record->leaf.data_len = bytes;
1700 KKASSERT(HAMMER_SYMLINK_NAME_OFF == 0);
1701 bcopy(ap->a_target, record->data->symlink.name, bytes);
1702 error = hammer_ip_add_record(&trans, record);
1706 * Set the file size to the length of the link.
1709 nip->ino_data.size = bytes;
1710 hammer_modify_inode(nip, HAMMER_INODE_DDIRTY);
1714 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
1720 hammer_rel_inode(nip, 0);
1723 error = hammer_get_vnode(nip, ap->a_vpp);
1724 hammer_rel_inode(nip, 0);
1726 cache_setunresolved(ap->a_nch);
1727 cache_setvp(ap->a_nch, *ap->a_vpp);
1730 hammer_done_transaction(&trans);
1735 * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1739 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1741 struct hammer_transaction trans;
1742 struct hammer_inode *dip;
1745 dip = VTOI(ap->a_dvp);
1747 if (hammer_nohistory(dip) == 0 &&
1748 (error = hammer_checkspace(dip->hmp)) != 0) {
1752 hammer_start_transaction(&trans, dip->hmp);
1753 error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp,
1754 ap->a_cred, ap->a_flags);
1755 hammer_done_transaction(&trans);
1761 * hammer_vop_ioctl { vp, command, data, fflag, cred }
1765 hammer_vop_ioctl(struct vop_ioctl_args *ap)
1767 struct hammer_inode *ip = ap->a_vp->v_data;
1769 return(hammer_ioctl(ip, ap->a_command, ap->a_data,
1770 ap->a_fflag, ap->a_cred));
1775 hammer_vop_mountctl(struct vop_mountctl_args *ap)
1780 mp = ap->a_head.a_ops->head.vv_mount;
1783 case MOUNTCTL_SET_EXPORT:
1784 if (ap->a_ctllen != sizeof(struct export_args))
1786 error = hammer_vfs_export(mp, ap->a_op,
1787 (const struct export_args *)ap->a_ctl);
1790 error = journal_mountctl(ap);
1797 * hammer_vop_strategy { vp, bio }
1799 * Strategy call, used for regular file read & write only. Note that the
1800 * bp may represent a cluster.
1802 * To simplify operation and allow better optimizations in the future,
1803 * this code does not make any assumptions with regards to buffer alignment
1808 hammer_vop_strategy(struct vop_strategy_args *ap)
1813 bp = ap->a_bio->bio_buf;
1817 error = hammer_vop_strategy_read(ap);
1820 error = hammer_vop_strategy_write(ap);
1823 bp->b_error = error = EINVAL;
1824 bp->b_flags |= B_ERROR;
1832 * Read from a regular file. Iterate the related records and fill in the
1833 * BIO/BUF. Gaps are zero-filled.
1835 * The support code in hammer_object.c should be used to deal with mixed
1836 * in-memory and on-disk records.
1842 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1844 struct hammer_transaction trans;
1845 struct hammer_inode *ip;
1846 struct hammer_cursor cursor;
1847 hammer_base_elm_t base;
1861 ip = ap->a_vp->v_data;
1864 * The zone-2 disk offset may have been set by the cluster code via
1865 * a BMAP operation. Take care not to confuse it with the bio_offset
1866 * set by hammer_io_direct_write(), which is a device-relative offset.
1868 * Checking the high bits should suffice.
1870 nbio = push_bio(bio);
1871 if ((nbio->bio_offset & HAMMER_OFF_ZONE_MASK) == HAMMER_ZONE_RAW_BUFFER) {
1872 error = hammer_io_direct_read(ip->hmp, nbio->bio_offset, bio);
1879 hammer_simple_transaction(&trans, ip->hmp);
1880 hammer_init_cursor(&trans, &cursor, &ip->cache[1], ip);
1883 * Key range (begin and end inclusive) to scan. Note that the key's
1884 * stored in the actual records represent BASE+LEN, not BASE. The
1885 * first record containing bio_offset will have a key > bio_offset.
1887 cursor.key_beg.localization = HAMMER_LOCALIZE_MISC;
1888 cursor.key_beg.obj_id = ip->obj_id;
1889 cursor.key_beg.create_tid = 0;
1890 cursor.key_beg.delete_tid = 0;
1891 cursor.key_beg.obj_type = 0;
1892 cursor.key_beg.key = bio->bio_offset + 1;
1893 cursor.asof = ip->obj_asof;
1894 cursor.flags |= HAMMER_CURSOR_ASOF | HAMMER_CURSOR_DATAEXTOK;
1896 cursor.key_end = cursor.key_beg;
1897 KKASSERT(ip->ino_data.obj_type == HAMMER_OBJTYPE_REGFILE);
1899 if (ip->ino_data.obj_type == HAMMER_OBJTYPE_DBFILE) {
1900 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1901 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1902 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1906 ran_end = bio->bio_offset + bp->b_bufsize;
1907 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1908 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1909 tmp64 = ran_end + MAXPHYS + 1; /* work-around GCC-4 bug */
1910 if (tmp64 < ran_end)
1911 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1913 cursor.key_end.key = ran_end + MAXPHYS + 1;
1915 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1917 error = hammer_ip_first(&cursor);
1920 while (error == 0) {
1922 * Get the base file offset of the record. The key for
1923 * data records is (base + bytes) rather then (base).
1925 base = &cursor.leaf->base;
1926 rec_offset = base->key - cursor.leaf->data_len;
1929 * Calculate the gap, if any, and zero-fill it.
1931 * n is the offset of the start of the record verses our
1932 * current seek offset in the bio.
1934 n = (int)(rec_offset - (bio->bio_offset + boff));
1936 if (n > bp->b_bufsize - boff)
1937 n = bp->b_bufsize - boff;
1938 bzero((char *)bp->b_data + boff, n);
1944 * Calculate the data offset in the record and the number
1945 * of bytes we can copy.
1947 * There are two degenerate cases. First, boff may already
1948 * be at bp->b_bufsize. Secondly, the data offset within
1949 * the record may exceed the record's size.
1953 n = cursor.leaf->data_len - roff;
1955 kprintf("strategy_read: bad n=%d roff=%d\n", n, roff);
1957 } else if (n > bp->b_bufsize - boff) {
1958 n = bp->b_bufsize - boff;
1962 * Deal with cached truncations. This cool bit of code
1963 * allows truncate()/ftruncate() to avoid having to sync
1966 * If the frontend is truncated then all backend records are
1967 * subject to the frontend's truncation.
1969 * If the backend is truncated then backend records on-disk
1970 * (but not in-memory) are subject to the backend's
1971 * truncation. In-memory records owned by the backend
1972 * represent data written after the truncation point on the
1973 * backend and must not be truncated.
1975 * Truncate operations deal with frontend buffer cache
1976 * buffers and frontend-owned in-memory records synchronously.
1978 if (ip->flags & HAMMER_INODE_TRUNCATED) {
1979 if (hammer_cursor_ondisk(&cursor) ||
1980 cursor.iprec->flush_state == HAMMER_FST_FLUSH) {
1981 if (ip->trunc_off <= rec_offset)
1983 else if (ip->trunc_off < rec_offset + n)
1984 n = (int)(ip->trunc_off - rec_offset);
1987 if (ip->sync_flags & HAMMER_INODE_TRUNCATED) {
1988 if (hammer_cursor_ondisk(&cursor)) {
1989 if (ip->sync_trunc_off <= rec_offset)
1991 else if (ip->sync_trunc_off < rec_offset + n)
1992 n = (int)(ip->sync_trunc_off - rec_offset);
1997 * Try to issue a direct read into our bio if possible,
1998 * otherwise resolve the element data into a hammer_buffer
2002 ((cursor.leaf->data_offset + roff) & HAMMER_BUFMASK) == 0) {
2003 error = hammer_io_direct_read(
2005 cursor.leaf->data_offset + roff,
2009 error = hammer_ip_resolve_data(&cursor);
2011 bcopy((char *)cursor.data + roff,
2012 (char *)bp->b_data + boff, n);
2019 * Iterate until we have filled the request.
2022 if (boff == bp->b_bufsize)
2024 error = hammer_ip_next(&cursor);
2028 * There may have been a gap after the last record
2030 if (error == ENOENT)
2032 if (error == 0 && boff != bp->b_bufsize) {
2033 KKASSERT(boff < bp->b_bufsize);
2034 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
2035 /* boff = bp->b_bufsize; */
2038 bp->b_error = error;
2040 bp->b_flags |= B_ERROR;
2045 hammer_cache_node(cursor.node, &ip->cache[1]);
2046 hammer_done_cursor(&cursor);
2047 hammer_done_transaction(&trans);
2052 * BMAP operation - used to support cluster_read() only.
2054 * (struct vnode *vp, off_t loffset, off_t *doffsetp, int *runp, int *runb)
2056 * This routine may return EOPNOTSUPP if the opration is not supported for
2057 * the specified offset. The contents of the pointer arguments do not
2058 * need to be initialized in that case.
2060 * If a disk address is available and properly aligned return 0 with
2061 * *doffsetp set to the zone-2 address, and *runp / *runb set appropriately
2062 * to the run-length relative to that offset. Callers may assume that
2063 * *doffsetp is valid if 0 is returned, even if *runp is not sufficiently
2064 * large, so return EOPNOTSUPP if it is not sufficiently large.
2068 hammer_vop_bmap(struct vop_bmap_args *ap)
2070 struct hammer_transaction trans;
2071 struct hammer_inode *ip;
2072 struct hammer_cursor cursor;
2073 hammer_base_elm_t base;
2077 int64_t base_offset;
2078 int64_t base_disk_offset;
2079 int64_t last_offset;
2080 hammer_off_t last_disk_offset;
2081 hammer_off_t disk_offset;
2085 ip = ap->a_vp->v_data;
2088 * We can only BMAP regular files. We can't BMAP database files,
2091 if (ip->ino_data.obj_type != HAMMER_OBJTYPE_REGFILE)
2095 * bmap is typically called with runp/runb both NULL when used
2096 * for writing. We do not support BMAP for writing atm.
2098 if (ap->a_runp == NULL && ap->a_runb == NULL)
2102 * Scan the B-Tree to acquire blockmap addresses, then translate
2105 hammer_simple_transaction(&trans, ip->hmp);
2106 hammer_init_cursor(&trans, &cursor, &ip->cache[1], ip);
2109 * Key range (begin and end inclusive) to scan. Note that the key's
2110 * stored in the actual records represent BASE+LEN, not BASE. The
2111 * first record containing bio_offset will have a key > bio_offset.
2113 cursor.key_beg.localization = HAMMER_LOCALIZE_MISC;
2114 cursor.key_beg.obj_id = ip->obj_id;
2115 cursor.key_beg.create_tid = 0;
2116 cursor.key_beg.delete_tid = 0;
2117 cursor.key_beg.obj_type = 0;
2119 cursor.key_beg.key = ap->a_loffset - MAXPHYS + 1;
2121 cursor.key_beg.key = ap->a_loffset + 1;
2122 if (cursor.key_beg.key < 0)
2123 cursor.key_beg.key = 0;
2124 cursor.asof = ip->obj_asof;
2125 cursor.flags |= HAMMER_CURSOR_ASOF | HAMMER_CURSOR_DATAEXTOK;
2127 cursor.key_end = cursor.key_beg;
2128 KKASSERT(ip->ino_data.obj_type == HAMMER_OBJTYPE_REGFILE);
2130 ran_end = ap->a_loffset + MAXPHYS;
2131 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
2132 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
2133 tmp64 = ran_end + MAXPHYS + 1; /* work-around GCC-4 bug */
2134 if (tmp64 < ran_end)
2135 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
2137 cursor.key_end.key = ran_end + MAXPHYS + 1;
2139 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
2141 error = hammer_ip_first(&cursor);
2142 base_offset = last_offset = 0;
2143 base_disk_offset = last_disk_offset = 0;
2145 while (error == 0) {
2147 * Get the base file offset of the record. The key for
2148 * data records is (base + bytes) rather then (base).
2150 base = &cursor.leaf->base;
2151 rec_offset = base->key - cursor.leaf->data_len;
2152 rec_len = cursor.leaf->data_len;
2155 * Incorporate any cached truncation
2157 if (ip->flags & HAMMER_INODE_TRUNCATED) {
2158 if (hammer_cursor_ondisk(&cursor) ||
2159 cursor.iprec->flush_state == HAMMER_FST_FLUSH) {
2160 if (ip->trunc_off <= rec_offset)
2162 else if (ip->trunc_off < rec_offset + rec_len)
2163 rec_len = (int)(ip->trunc_off - rec_offset);
2166 if (ip->sync_flags & HAMMER_INODE_TRUNCATED) {
2167 if (hammer_cursor_ondisk(&cursor)) {
2168 if (ip->sync_trunc_off <= rec_offset)
2170 else if (ip->sync_trunc_off < rec_offset + rec_len)
2171 rec_len = (int)(ip->sync_trunc_off - rec_offset);
2176 * Accumulate information. If we have hit a discontiguous
2177 * block reset base_offset unless we are already beyond the
2178 * requested offset. If we are, that's it, we stop.
2180 disk_offset = hammer_blockmap_lookup(trans.hmp,
2181 cursor.leaf->data_offset,
2185 if (rec_offset != last_offset ||
2186 disk_offset != last_disk_offset) {
2187 if (rec_offset > ap->a_loffset)
2189 base_offset = rec_offset;
2190 base_disk_offset = disk_offset;
2192 last_offset = rec_offset + rec_len;
2193 last_disk_offset = disk_offset + rec_len;
2195 error = hammer_ip_next(&cursor);
2199 kprintf("BMAP %016llx: %016llx - %016llx\n",
2200 ap->a_loffset, base_offset, last_offset);
2201 kprintf("BMAP %16s: %016llx - %016llx\n",
2202 "", base_disk_offset, last_disk_offset);
2206 hammer_cache_node(cursor.node, &ip->cache[1]);
2207 hammer_done_cursor(&cursor);
2208 hammer_done_transaction(&trans);
2210 if (base_offset == 0 || base_offset > ap->a_loffset ||
2211 last_offset < ap->a_loffset) {
2214 disk_offset = base_disk_offset + (ap->a_loffset - base_offset);
2217 * If doffsetp is not aligned or the forward run size does
2218 * not cover a whole buffer, disallow the direct I/O.
2220 if ((disk_offset & HAMMER_BUFMASK) ||
2221 (last_offset - ap->a_loffset) < HAMMER_BUFSIZE) {
2224 *ap->a_doffsetp = disk_offset;
2226 *ap->a_runb = ap->a_loffset - base_offset;
2228 *ap->a_runp = last_offset - ap->a_loffset;
2236 * Write to a regular file. Because this is a strategy call the OS is
2237 * trying to actually sync data to the media. HAMMER can only flush
2238 * the entire inode (so the TID remains properly synchronized).
2240 * Basically all we do here is place the bio on the inode's flush queue
2241 * and activate the flusher.
2245 hammer_vop_strategy_write(struct vop_strategy_args *ap)
2247 hammer_record_t record;
2257 ip = ap->a_vp->v_data;
2260 if (ip->flags & HAMMER_INODE_RO) {
2261 bp->b_error = EROFS;
2262 bp->b_flags |= B_ERROR;
2264 hammer_cleanup_write_io(ip);
2269 * Interlock with inode destruction (no in-kernel or directory
2270 * topology visibility). If we queue new IO while trying to
2271 * destroy the inode we can deadlock the vtrunc call in
2272 * hammer_inode_unloadable_check().
2274 if (ip->flags & (HAMMER_INODE_DELETING|HAMMER_INODE_DELETED)) {
2277 hammer_cleanup_write_io(ip);
2282 * Reserve space and issue a direct-write from the front-end.
2283 * NOTE: The direct_io code will hammer_bread/bcopy smaller
2286 * An in-memory record will be installed to reference the storage
2287 * until the flusher can get to it.
2289 * Since we own the high level bio the front-end will not try to
2290 * do a direct-read until the write completes.
2292 * NOTE: The only time we do not reserve a full-sized buffers
2293 * worth of data is if the file is small. We do not try to
2294 * allocate a fragment (from the small-data zone) at the end of
2295 * an otherwise large file as this can lead to wildly separated
2298 KKASSERT((bio->bio_offset & HAMMER_BUFMASK) == 0);
2299 KKASSERT(bio->bio_offset < ip->ino_data.size);
2300 if (bio->bio_offset || ip->ino_data.size > HAMMER_BUFSIZE / 2)
2301 bytes = (bp->b_bufsize + HAMMER_BUFMASK) & ~HAMMER_BUFMASK;
2303 bytes = ((int)ip->ino_data.size + 15) & ~15;
2305 record = hammer_ip_add_bulk(ip, bio->bio_offset, bp->b_data,
2308 hammer_io_direct_write(hmp, &record->leaf, bio);
2309 hammer_rel_mem_record(record);
2310 if (hmp->rsv_recs > hammer_limit_recs &&
2311 ip->rsv_recs > hammer_limit_irecs / 10) {
2312 hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
2313 } else if (ip->rsv_recs > hammer_limit_irecs / 2) {
2314 hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
2317 bp->b_bio2.bio_offset = NOOFFSET;
2318 bp->b_error = error;
2319 bp->b_flags |= B_ERROR;
2322 hammer_cleanup_write_io(ip);
2327 * Clean-up after disposing of a dirty frontend buffer's data.
2328 * This is somewhat heuristical so try to be robust.
2331 hammer_cleanup_write_io(hammer_inode_t ip)
2333 if (ip->rsv_databufs) {
2335 --ip->hmp->rsv_databufs;
2340 * We can lose track of dirty buffer cache buffers if we truncate, this
2341 * routine will resynchronize the count.
2345 hammer_update_rsv_databufs(hammer_inode_t ip)
2353 RB_FOREACH(bp, buf_rb_tree, &ip->vp->v_rbdirty_tree) {
2359 delta = n - ip->rsv_databufs;
2360 ip->rsv_databufs += delta;
2361 ip->hmp->rsv_databufs += delta;
2365 * dounlink - disconnect a directory entry
2367 * XXX whiteout support not really in yet
2370 hammer_dounlink(hammer_transaction_t trans, struct nchandle *nch,
2371 struct vnode *dvp, struct ucred *cred, int flags)
2373 struct namecache *ncp;
2376 struct hammer_cursor cursor;
2381 * Calculate the namekey and setup the key range for the scan. This
2382 * works kinda like a chained hash table where the lower 32 bits
2383 * of the namekey synthesize the chain.
2385 * The key range is inclusive of both key_beg and key_end.
2390 if (dip->flags & HAMMER_INODE_RO)
2393 namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
2395 hammer_init_cursor(trans, &cursor, &dip->cache[0], dip);
2396 cursor.key_beg.localization = HAMMER_LOCALIZE_MISC;
2397 cursor.key_beg.obj_id = dip->obj_id;
2398 cursor.key_beg.key = namekey;
2399 cursor.key_beg.create_tid = 0;
2400 cursor.key_beg.delete_tid = 0;
2401 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
2402 cursor.key_beg.obj_type = 0;
2404 cursor.key_end = cursor.key_beg;
2405 cursor.key_end.key |= 0xFFFFFFFFULL;
2406 cursor.asof = dip->obj_asof;
2407 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
2410 * Scan all matching records (the chain), locate the one matching
2411 * the requested path component. info->last_error contains the
2412 * error code on search termination and could be 0, ENOENT, or
2415 * The hammer_ip_*() functions merge in-memory records with on-disk
2416 * records for the purposes of the search.
2418 error = hammer_ip_first(&cursor);
2420 while (error == 0) {
2421 error = hammer_ip_resolve_data(&cursor);
2424 nlen = cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF;
2426 if (ncp->nc_nlen == nlen &&
2427 bcmp(ncp->nc_name, cursor.data->entry.name, nlen) == 0) {
2430 error = hammer_ip_next(&cursor);
2434 * If all is ok we have to get the inode so we can adjust nlinks.
2435 * To avoid a deadlock with the flusher we must release the inode
2436 * lock on the directory when acquiring the inode for the entry.
2438 * If the target is a directory, it must be empty.
2441 hammer_unlock(&cursor.ip->lock);
2442 ip = hammer_get_inode(trans, &dip->cache[1],
2443 cursor.data->entry.obj_id,
2444 dip->hmp->asof, 0, &error);
2445 hammer_lock_sh(&cursor.ip->lock);
2446 if (error == ENOENT) {
2447 kprintf("obj_id %016llx\n", cursor.data->entry.obj_id);
2448 Debugger("ENOENT unlinking object that should exist");
2452 * If we are trying to remove a directory the directory must
2455 * WARNING: hammer_ip_check_directory_empty() may have to
2456 * terminate the cursor to avoid a deadlock. It is ok to
2457 * call hammer_done_cursor() twice.
2459 if (error == 0 && ip->ino_data.obj_type ==
2460 HAMMER_OBJTYPE_DIRECTORY) {
2461 error = hammer_ip_check_directory_empty(trans, ip);
2465 * Delete the directory entry.
2467 * WARNING: hammer_ip_del_directory() may have to terminate
2468 * the cursor to avoid a deadlock. It is ok to call
2469 * hammer_done_cursor() twice.
2472 error = hammer_ip_del_directory(trans, &cursor,
2475 hammer_done_cursor(&cursor);
2477 cache_setunresolved(nch);
2478 cache_setvp(nch, NULL);
2481 cache_inval_vp(ip->vp, CINV_DESTROY);
2484 hammer_rel_inode(ip, 0);
2486 hammer_done_cursor(&cursor);
2488 if (error == EDEADLK)
2494 /************************************************************************
2495 * FIFO AND SPECFS OPS *
2496 ************************************************************************
2501 hammer_vop_fifoclose (struct vop_close_args *ap)
2503 /* XXX update itimes */
2504 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
2508 hammer_vop_fiforead (struct vop_read_args *ap)
2512 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2513 /* XXX update access time */
2518 hammer_vop_fifowrite (struct vop_write_args *ap)
2522 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2523 /* XXX update access time */
2528 hammer_vop_specclose (struct vop_close_args *ap)
2530 /* XXX update itimes */
2531 return (VOCALL(&spec_vnode_vops, &ap->a_head));
2535 hammer_vop_specread (struct vop_read_args *ap)
2537 /* XXX update access time */
2538 return (VOCALL(&spec_vnode_vops, &ap->a_head));
2542 hammer_vop_specwrite (struct vop_write_args *ap)
2544 /* XXX update last change time */
2545 return (VOCALL(&spec_vnode_vops, &ap->a_head));