2 * Copyright (c) 2011-2015 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@dragonflybsd.org>
6 * by Venkatesh Srinivas <vsrinivas@dragonflybsd.org>
7 * by Daniel Flores (GSOC 2013 - mentored by Matthew Dillon, compression)
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in
17 * the documentation and/or other materials provided with the
19 * 3. Neither the name of The DragonFly Project nor the names of its
20 * contributors may be used to endorse or promote products derived
21 * from this software without specific, prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
26 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
27 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
28 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
29 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
30 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
31 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
32 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
33 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * Kernel Filesystem interface
39 * NOTE! local ipdata pointers must be reloaded on any modifying operation
40 * to the inode as its underlying chain may have changed.
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/fcntl.h>
49 #include <sys/namei.h>
50 #include <sys/mount.h>
51 #include <sys/vnode.h>
52 #include <sys/mountctl.h>
53 #include <sys/dirent.h>
55 #include <sys/objcache.h>
56 #include <sys/event.h>
58 #include <vfs/fifofs/fifo.h>
62 static int hammer2_read_file(hammer2_inode_t *ip, struct uio *uio,
64 static int hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
65 int ioflag, int seqcount);
66 static void hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize);
67 static void hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize);
69 struct objcache *cache_xops;
73 hammer2_knote(struct vnode *vp, int flags)
76 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
80 * Last reference to a vnode is going away but it is still cached.
84 hammer2_vop_inactive(struct vop_inactive_args *ap)
103 * Check for deleted inodes and recycle immediately on the last
104 * release. Be sure to destroy any left-over buffer cache buffers
105 * so we do not waste time trying to flush them.
107 * Note that deleting the file block chains under the inode chain
108 * would just be a waste of energy, so don't do it.
110 * WARNING: nvtruncbuf() can only be safely called without the inode
111 * lock held due to the way our write thread works.
113 if (ip->flags & HAMMER2_INODE_ISUNLINKED) {
118 * Detect updates to the embedded data which may be
119 * synchronized by the strategy code. Simply mark the
120 * inode modified so it gets picked up by our normal flush.
122 nblksize = hammer2_calc_logical(ip, 0, &lbase, NULL);
123 nvtruncbuf(vp, 0, nblksize, 0, 0);
131 * Reclaim a vnode so that it can be reused; after the inode is
132 * disassociated, the filesystem must manage it alone.
136 hammer2_vop_reclaim(struct vop_reclaim_args *ap)
152 * The final close of a deleted file or directory marks it for
153 * destruction. The DELETED flag allows the flusher to shortcut
154 * any modified blocks still unflushed (that is, just ignore them).
156 * HAMMER2 usually does not try to optimize the freemap by returning
157 * deleted blocks to it as it does not usually know how many snapshots
158 * might be referencing portions of the file/dir.
164 * NOTE! We do not attempt to flush chains here, flushing is
165 * really fragile and could also deadlock.
170 * This occurs if the inode was unlinked while open. Reclamation of
171 * these inodes requires processing we cannot safely do here so add
172 * the inode to the sideq in that situation.
174 * A modified inode may require chain synchronization which will no
175 * longer be driven by a sync or fsync without the vnode, also use
176 * the sideq for that.
178 * A reclaim can occur at any time so we cannot safely start a
179 * transaction to handle reclamation of unlinked files. Instead,
180 * the ip is left with a reference and placed on a linked list and
184 if ((ip->flags & (HAMMER2_INODE_ISUNLINKED |
185 HAMMER2_INODE_MODIFIED |
186 HAMMER2_INODE_RESIZED)) &&
187 (ip->flags & HAMMER2_INODE_ISDELETED) == 0) {
188 hammer2_inode_sideq_t *ipul;
190 ipul = kmalloc(sizeof(*ipul), pmp->minode, M_WAITOK | M_ZERO);
193 hammer2_spin_ex(&pmp->list_spin);
194 if ((ip->flags & HAMMER2_INODE_ONSIDEQ) == 0) {
196 atomic_set_int(&ip->flags, HAMMER2_INODE_ONSIDEQ);
197 TAILQ_INSERT_TAIL(&pmp->sideq, ipul, entry);
198 hammer2_spin_unex(&pmp->list_spin);
200 hammer2_spin_unex(&pmp->list_spin);
201 kfree(ipul, pmp->minode);
202 hammer2_inode_drop(ip); /* vp ref */
204 /* retain ref from vp for ipul */
206 hammer2_inode_drop(ip); /* vp ref */
210 * XXX handle background sync when ip dirty, kernel will no longer
211 * notify us regarding this inode because there is no longer a
212 * vnode attached to it.
221 hammer2_vop_fsync(struct vop_fsync_args *ap)
231 /* XXX can't do this yet */
232 hammer2_trans_init(ip->pmp, HAMMER2_TRANS_ISFLUSH);
233 vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
235 hammer2_trans_init(ip->pmp, 0);
236 vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
239 * Calling chain_flush here creates a lot of duplicative
240 * COW operations due to non-optimal vnode ordering.
242 * Only do it for an actual fsync() syscall. The other forms
243 * which call this function will eventually call chain_flush
244 * on the volume root as a catch-all, which is far more optimal.
246 hammer2_inode_lock(ip, 0);
247 if (ip->flags & HAMMER2_INODE_MODIFIED)
248 hammer2_inode_chain_sync(ip);
249 hammer2_inode_unlock(ip);
250 hammer2_trans_done(ip->pmp);
258 hammer2_vop_access(struct vop_access_args *ap)
260 hammer2_inode_t *ip = VTOI(ap->a_vp);
266 hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
267 uid = hammer2_to_unix_xid(&ip->meta.uid);
268 gid = hammer2_to_unix_xid(&ip->meta.gid);
269 error = vop_helper_access(ap, uid, gid, ip->meta.mode, ip->meta.uflags);
270 hammer2_inode_unlock(ip);
278 hammer2_vop_getattr(struct vop_getattr_args *ap)
284 hammer2_chain_t *chain;
294 hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
296 vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0];
297 vap->va_fileid = ip->meta.inum;
298 vap->va_mode = ip->meta.mode;
299 vap->va_nlink = ip->meta.nlinks;
300 vap->va_uid = hammer2_to_unix_xid(&ip->meta.uid);
301 vap->va_gid = hammer2_to_unix_xid(&ip->meta.gid);
304 vap->va_size = ip->meta.size; /* protected by shared lock */
305 vap->va_blocksize = HAMMER2_PBUFSIZE;
306 vap->va_flags = ip->meta.uflags;
307 hammer2_time_to_timespec(ip->meta.ctime, &vap->va_ctime);
308 hammer2_time_to_timespec(ip->meta.mtime, &vap->va_mtime);
309 hammer2_time_to_timespec(ip->meta.mtime, &vap->va_atime);
312 if (ip->meta.type == HAMMER2_OBJTYPE_DIRECTORY) {
314 * Can't really calculate directory use sans the files under
315 * it, just assume one block for now.
317 vap->va_bytes += HAMMER2_INODE_BYTES;
319 for (i = 0; i < ip->cluster.nchains; ++i) {
320 if ((chain = ip->cluster.array[i].chain) != NULL) {
322 chain->bref.embed.stats.data_count) {
324 chain->bref.embed.stats.data_count;
329 vap->va_type = hammer2_get_vtype(ip->meta.type);
331 vap->va_uid_uuid = ip->meta.uid;
332 vap->va_gid_uuid = ip->meta.gid;
333 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
336 hammer2_inode_unlock(ip);
344 hammer2_vop_setattr(struct vop_setattr_args *ap)
356 hammer2_update_time(&ctime);
360 if (ip->pmp->ronly) {
365 hammer2_pfs_memory_wait(ip->pmp);
366 hammer2_trans_init(ip->pmp, 0);
367 hammer2_inode_lock(ip, 0);
370 if (vap->va_flags != VNOVAL) {
373 flags = ip->meta.uflags;
374 error = vop_helper_setattr_flags(&flags, vap->va_flags,
375 hammer2_to_unix_xid(&ip->meta.uid),
378 if (ip->meta.uflags != flags) {
379 hammer2_inode_modify(ip);
380 ip->meta.uflags = flags;
381 ip->meta.ctime = ctime;
382 kflags |= NOTE_ATTRIB;
384 if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
391 if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
395 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
396 mode_t cur_mode = ip->meta.mode;
397 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
398 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
402 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
404 &cur_uid, &cur_gid, &cur_mode);
406 hammer2_guid_to_uuid(&uuid_uid, cur_uid);
407 hammer2_guid_to_uuid(&uuid_gid, cur_gid);
408 if (bcmp(&uuid_uid, &ip->meta.uid, sizeof(uuid_uid)) ||
409 bcmp(&uuid_gid, &ip->meta.gid, sizeof(uuid_gid)) ||
410 ip->meta.mode != cur_mode
412 hammer2_inode_modify(ip);
413 ip->meta.uid = uuid_uid;
414 ip->meta.gid = uuid_gid;
415 ip->meta.mode = cur_mode;
416 ip->meta.ctime = ctime;
418 kflags |= NOTE_ATTRIB;
425 if (vap->va_size != VNOVAL && ip->meta.size != vap->va_size) {
428 if (vap->va_size == ip->meta.size)
430 if (vap->va_size < ip->meta.size) {
431 hammer2_mtx_ex(&ip->truncate_lock);
432 hammer2_truncate_file(ip, vap->va_size);
433 hammer2_mtx_unlock(&ip->truncate_lock);
435 hammer2_extend_file(ip, vap->va_size);
437 hammer2_inode_modify(ip);
438 ip->meta.mtime = ctime;
446 /* atime not supported */
447 if (vap->va_atime.tv_sec != VNOVAL) {
448 hammer2_inode_modify(ip);
449 ip->meta.atime = hammer2_timespec_to_time(&vap->va_atime);
450 kflags |= NOTE_ATTRIB;
453 if (vap->va_mode != (mode_t)VNOVAL) {
454 mode_t cur_mode = ip->meta.mode;
455 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
456 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
458 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
459 cur_uid, cur_gid, &cur_mode);
460 if (error == 0 && ip->meta.mode != cur_mode) {
461 hammer2_inode_modify(ip);
462 ip->meta.mode = cur_mode;
463 ip->meta.ctime = ctime;
464 kflags |= NOTE_ATTRIB;
468 if (vap->va_mtime.tv_sec != VNOVAL) {
469 hammer2_inode_modify(ip);
470 ip->meta.mtime = hammer2_timespec_to_time(&vap->va_mtime);
471 kflags |= NOTE_ATTRIB;
476 * If a truncation occurred we must call inode_fsync() now in order
477 * to trim the related data chains, otherwise a later expansion can
480 * If an extend occured that changed the DIRECTDATA state, we must
481 * call inode_fsync now in order to prepare the inode's indirect
484 if (ip->flags & HAMMER2_INODE_RESIZED)
485 hammer2_inode_chain_sync(ip);
490 hammer2_inode_unlock(ip);
491 hammer2_trans_done(ip->pmp);
492 hammer2_knote(ip->vp, kflags);
500 hammer2_vop_readdir(struct vop_readdir_args *ap)
502 hammer2_xop_readdir_t *xop;
503 hammer2_blockref_t bref;
519 saveoff = uio->uio_offset;
524 * Setup cookies directory entry cookies if requested
526 if (ap->a_ncookies) {
527 ncookies = uio->uio_resid / 16 + 1;
530 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
537 hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
540 * Handle artificial entries. To ensure that only positive 64 bit
541 * quantities are returned to userland we always strip off bit 63.
542 * The hash code is designed such that codes 0x0000-0x7FFF are not
543 * used, allowing us to use these codes for articial entries.
545 * Entry 0 is used for '.' and entry 1 is used for '..'. Do not
546 * allow '..' to cross the mount point into (e.g.) the super-root.
549 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
550 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
554 cookies[cookie_index] = saveoff;
557 if (cookie_index == ncookies)
563 * Be careful with lockorder when accessing ".."
565 * (ip is the current dir. xip is the parent dir).
567 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
568 if (ip != ip->pmp->iroot)
569 inum = ip->meta.iparent & HAMMER2_DIRHASH_USERMSK;
570 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
574 cookies[cookie_index] = saveoff;
577 if (cookie_index == ncookies)
581 lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
582 if (hammer2_debug & 0x0020)
583 kprintf("readdir: lkey %016jx\n", lkey);
588 * Use XOP for cluster scan.
590 * parent is the inode cluster, already locked for us. Don't
591 * double lock shared locks as this will screw up upgrades.
593 xop = hammer2_xop_alloc(ip, 0);
595 hammer2_xop_start(&xop->head, hammer2_xop_readdir);
598 const hammer2_inode_data_t *ripdata;
602 error = hammer2_xop_collect(&xop->head, 0);
605 if (cookie_index == ncookies)
607 if (hammer2_debug & 0x0020)
608 kprintf("cluster chain %p %p\n",
609 xop->head.cluster.focus,
610 (xop->head.cluster.focus ?
611 xop->head.cluster.focus->data : (void *)-1));
612 hammer2_cluster_bref(&xop->head.cluster, &bref);
614 if (bref.type == HAMMER2_BREF_TYPE_INODE) {
616 &hammer2_cluster_rdata(&xop->head.cluster)->ipdata;
617 dtype = hammer2_get_dtype(ripdata->meta.type);
618 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
619 r = vop_write_dirent(&error, uio,
621 HAMMER2_DIRHASH_USERMSK,
623 ripdata->meta.name_len,
628 cookies[cookie_index] = saveoff;
630 } else if (bref.type == HAMMER2_BREF_TYPE_DIRENT) {
631 dtype = hammer2_get_dtype(bref.embed.dirent.type);
632 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
633 if (bref.embed.dirent.namlen <=
634 sizeof(bref.check.buf)) {
635 dname = bref.check.buf;
638 hammer2_cluster_rdata(&xop->head.cluster)->buf;
640 r = vop_write_dirent(&error, uio,
641 bref.embed.dirent.inum,
643 bref.embed.dirent.namlen,
648 cookies[cookie_index] = saveoff;
651 /* XXX chain error */
652 kprintf("bad chain type readdir %d\n", bref.type);
655 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
656 if (error == ENOENT) {
659 saveoff = (hammer2_key_t)-1;
661 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
664 hammer2_inode_unlock(ip);
666 *ap->a_eofflag = eofflag;
667 if (hammer2_debug & 0x0020)
668 kprintf("readdir: done at %016jx\n", saveoff);
669 uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
670 if (error && cookie_index == 0) {
672 kfree(cookies, M_TEMP);
674 *ap->a_cookies = NULL;
678 *ap->a_ncookies = cookie_index;
679 *ap->a_cookies = cookies;
687 * hammer2_vop_readlink { vp, uio, cred }
691 hammer2_vop_readlink(struct vop_readlink_args *ap)
698 if (vp->v_type != VLNK)
702 error = hammer2_read_file(ip, ap->a_uio, 0);
708 hammer2_vop_read(struct vop_read_args *ap)
718 * Read operations supported on this vnode?
721 if (vp->v_type != VREG)
731 seqcount = ap->a_ioflag >> 16;
732 bigread = (uio->uio_resid > 100 * 1024 * 1024);
734 error = hammer2_read_file(ip, uio, seqcount);
740 hammer2_vop_write(struct vop_write_args *ap)
750 * Read operations supported on this vnode?
753 if (vp->v_type != VREG)
762 if (ip->pmp->ronly) {
766 seqcount = ap->a_ioflag >> 16;
769 * Check resource limit
771 if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc &&
772 uio->uio_offset + uio->uio_resid >
773 td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
774 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
779 * The transaction interlocks against flush initiations
780 * (note: but will run concurrently with the actual flush).
782 * To avoid deadlocking against the VM system, we must flag any
783 * transaction related to the buffer cache or other direct
784 * VM page manipulation.
786 if (uio->uio_segflg == UIO_NOCOPY)
787 hammer2_trans_init(ip->pmp, HAMMER2_TRANS_BUFCACHE);
789 hammer2_trans_init(ip->pmp, 0);
790 error = hammer2_write_file(ip, uio, ap->a_ioflag, seqcount);
791 hammer2_trans_done(ip->pmp);
797 * Perform read operations on a file or symlink given an UNLOCKED
800 * The passed ip is not locked.
804 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
815 * WARNING! Assumes that the kernel interlocks size changes at the
818 hammer2_mtx_sh(&ip->lock);
819 hammer2_mtx_sh(&ip->truncate_lock);
820 size = ip->meta.size;
821 hammer2_mtx_unlock(&ip->lock);
823 while (uio->uio_resid > 0 && uio->uio_offset < size) {
830 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
834 error = cluster_read(ip->vp, leof, lbase, lblksize,
835 uio->uio_resid, seqcount * MAXBSIZE,
838 if (uio->uio_segflg == UIO_NOCOPY) {
839 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
840 if (bp->b_flags & B_CACHE) {
843 if (bp->b_xio.xio_npages != 16)
844 kprintf("NPAGES BAD\n");
845 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
847 m = bp->b_xio.xio_pages[i];
848 if (m == NULL || m->valid == 0) {
849 kprintf("bp %016jx %016jx pg %d inv",
852 kprintf("m->object %p/%p", m->object, ip->vp->v_object);
858 kprintf("b_flags %08x, b_error %d\n", bp->b_flags, bp->b_error);
862 error = bread(ip->vp, lbase, lblksize, &bp);
868 loff = (int)(uio->uio_offset - lbase);
870 if (n > uio->uio_resid)
872 if (n > size - uio->uio_offset)
873 n = (int)(size - uio->uio_offset);
874 bp->b_flags |= B_AGE;
875 uiomovebp(bp, (char *)bp->b_data + loff, n, uio);
878 hammer2_mtx_unlock(&ip->truncate_lock);
884 * Write to the file represented by the inode via the logical buffer cache.
885 * The inode may represent a regular file or a symlink.
887 * The inode must not be locked.
891 hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
892 int ioflag, int seqcount)
894 hammer2_key_t old_eof;
895 hammer2_key_t new_eof;
904 * WARNING! Assumes that the kernel interlocks size changes at the
907 hammer2_mtx_ex(&ip->lock);
908 hammer2_mtx_sh(&ip->truncate_lock);
909 if (ioflag & IO_APPEND)
910 uio->uio_offset = ip->meta.size;
911 old_eof = ip->meta.size;
914 * Extend the file if necessary. If the write fails at some point
915 * we will truncate it back down to cover as much as we were able
918 * Doing this now makes it easier to calculate buffer sizes in
925 if (uio->uio_offset + uio->uio_resid > old_eof) {
926 new_eof = uio->uio_offset + uio->uio_resid;
928 hammer2_extend_file(ip, new_eof);
929 kflags |= NOTE_EXTEND;
933 hammer2_mtx_unlock(&ip->lock);
938 while (uio->uio_resid > 0) {
947 * Don't allow the buffer build to blow out the buffer
950 if ((ioflag & IO_RECURSE) == 0)
951 bwillwrite(HAMMER2_PBUFSIZE);
954 * This nominally tells us how much we can cluster and
955 * what the logical buffer size needs to be. Currently
956 * we don't try to cluster the write and just handle one
959 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
961 loff = (int)(uio->uio_offset - lbase);
963 KKASSERT(lblksize <= 65536);
966 * Calculate bytes to copy this transfer and whether the
967 * copy completely covers the buffer or not.
971 if (n > uio->uio_resid) {
973 if (loff == lbase && uio->uio_offset + n == new_eof)
981 if (lbase >= new_eof)
987 if (uio->uio_segflg == UIO_NOCOPY) {
989 * Issuing a write with the same data backing the
990 * buffer. Instantiate the buffer to collect the
991 * backing vm pages, then read-in any missing bits.
993 * This case is used by vop_stdputpages().
995 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
996 if ((bp->b_flags & B_CACHE) == 0) {
998 error = bread(ip->vp, lbase, lblksize, &bp);
1000 } else if (trivial) {
1002 * Even though we are entirely overwriting the buffer
1003 * we may still have to zero it out to avoid a
1004 * mmap/write visibility issue.
1006 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
1007 if ((bp->b_flags & B_CACHE) == 0)
1011 * Partial overwrite, read in any missing bits then
1012 * replace the portion being written.
1014 * (The strategy code will detect zero-fill physical
1015 * blocks for this case).
1017 error = bread(ip->vp, lbase, lblksize, &bp);
1028 * Ok, copy the data in
1030 error = uiomovebp(bp, bp->b_data + loff, n, uio);
1031 kflags |= NOTE_WRITE;
1039 * WARNING: Pageout daemon will issue UIO_NOCOPY writes
1040 * with IO_SYNC or IO_ASYNC set. These writes
1041 * must be handled as the pageout daemon expects.
1043 * NOTE! H2 relies on cluster_write() here because it
1044 * cannot preallocate disk blocks at the logical
1045 * level due to not knowing what the compression
1046 * size will be at this time.
1048 * We must use cluster_write() here and we depend
1049 * on the write-behind feature to flush buffers
1050 * appropriately. If we let the buffer daemons do
1051 * it the block allocations will be all over the
1054 if (ioflag & IO_SYNC) {
1056 } else if ((ioflag & IO_DIRECT) && endofblk) {
1058 } else if (ioflag & IO_ASYNC) {
1060 } else if (ip->vp->v_mount->mnt_flag & MNT_NOCLUSTERW) {
1064 bp->b_flags |= B_CLUSTEROK;
1065 cluster_write(bp, new_eof, lblksize, seqcount);
1067 bp->b_flags |= B_CLUSTEROK;
1074 * Cleanup. If we extended the file EOF but failed to write through
1075 * the entire write is a failure and we have to back-up.
1077 if (error && new_eof != old_eof) {
1078 hammer2_mtx_unlock(&ip->truncate_lock);
1079 hammer2_mtx_ex(&ip->lock);
1080 hammer2_mtx_ex(&ip->truncate_lock);
1081 hammer2_truncate_file(ip, old_eof);
1082 if (ip->flags & HAMMER2_INODE_MODIFIED)
1083 hammer2_inode_chain_sync(ip);
1084 hammer2_mtx_unlock(&ip->lock);
1085 } else if (modified) {
1086 hammer2_mtx_ex(&ip->lock);
1087 hammer2_inode_modify(ip);
1088 hammer2_update_time(&ip->meta.mtime);
1089 if (ip->flags & HAMMER2_INODE_MODIFIED)
1090 hammer2_inode_chain_sync(ip);
1091 hammer2_mtx_unlock(&ip->lock);
1092 hammer2_knote(ip->vp, kflags);
1094 hammer2_trans_assert_strategy(ip->pmp);
1095 hammer2_mtx_unlock(&ip->truncate_lock);
1101 * Truncate the size of a file. The inode must not be locked.
1103 * We must unconditionally set HAMMER2_INODE_RESIZED to properly
1104 * ensure that any on-media data beyond the new file EOF has been destroyed.
1106 * WARNING: nvtruncbuf() can only be safely called without the inode lock
1107 * held due to the way our write thread works. If the truncation
1108 * occurs in the middle of a buffer, nvtruncbuf() is responsible
1109 * for dirtying that buffer and zeroing out trailing bytes.
1111 * WARNING! Assumes that the kernel interlocks size changes at the
1114 * WARNING! Caller assumes responsibility for removing dead blocks
1115 * if INODE_RESIZED is set.
1119 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1121 hammer2_key_t lbase;
1125 hammer2_mtx_unlock(&ip->lock);
1127 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1128 nvtruncbuf(ip->vp, nsize,
1129 nblksize, (int)nsize & (nblksize - 1),
1132 hammer2_mtx_ex(&ip->lock);
1133 KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1134 ip->osize = ip->meta.size;
1135 ip->meta.size = nsize;
1136 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1137 hammer2_inode_modify(ip);
1142 * Extend the size of a file. The inode must not be locked.
1144 * Even though the file size is changing, we do not have to set the
1145 * INODE_RESIZED bit unless the file size crosses the EMBEDDED_BYTES
1146 * boundary. When this occurs a hammer2_inode_chain_sync() is required
1147 * to prepare the inode cluster's indirect block table, otherwise
1148 * async execution of the strategy code will implode on us.
1150 * WARNING! Assumes that the kernel interlocks size changes at the
1153 * WARNING! Caller assumes responsibility for transitioning out
1154 * of the inode DIRECTDATA mode if INODE_RESIZED is set.
1158 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1160 hammer2_key_t lbase;
1161 hammer2_key_t osize;
1167 KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1168 hammer2_inode_modify(ip);
1169 osize = ip->meta.size;
1171 ip->meta.size = nsize;
1173 if (osize <= HAMMER2_EMBEDDED_BYTES && nsize > HAMMER2_EMBEDDED_BYTES) {
1174 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1175 hammer2_inode_chain_sync(ip);
1178 hammer2_mtx_unlock(&ip->lock);
1180 oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL);
1181 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1187 hammer2_mtx_ex(&ip->lock);
1194 hammer2_vop_nresolve(struct vop_nresolve_args *ap)
1196 hammer2_xop_nresolve_t *xop;
1197 hammer2_inode_t *ip;
1198 hammer2_inode_t *dip;
1199 struct namecache *ncp;
1204 dip = VTOI(ap->a_dvp);
1205 xop = hammer2_xop_alloc(dip, 0);
1207 ncp = ap->a_nch->ncp;
1208 hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1211 * Note: In DragonFly the kernel handles '.' and '..'.
1213 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1214 hammer2_xop_start(&xop->head, hammer2_xop_nresolve);
1216 error = hammer2_xop_collect(&xop->head, 0);
1220 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1222 hammer2_inode_unlock(dip);
1225 * Acquire the related vnode
1227 * NOTE: For error processing, only ENOENT resolves the namecache
1228 * entry to NULL, otherwise we just return the error and
1229 * leave the namecache unresolved.
1231 * NOTE: multiple hammer2_inode structures can be aliased to the
1232 * same chain element, for example for hardlinks. This
1233 * use case does not 'reattach' inode associations that
1234 * might already exist, but always allocates a new one.
1236 * WARNING: inode structure is locked exclusively via inode_get
1237 * but chain was locked shared. inode_unlock()
1238 * will handle it properly.
1241 vp = hammer2_igetv(ip, &error);
1244 cache_setvp(ap->a_nch, vp);
1245 } else if (error == ENOENT) {
1246 cache_setvp(ap->a_nch, NULL);
1248 hammer2_inode_unlock(ip);
1251 * The vp should not be released until after we've disposed
1252 * of our locks, because it might cause vop_inactive() to
1259 cache_setvp(ap->a_nch, NULL);
1261 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1262 KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
1263 ("resolve error %d/%p ap %p\n",
1264 error, ap->a_nch->ncp->nc_vp, ap));
1272 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
1274 hammer2_inode_t *dip;
1279 dip = VTOI(ap->a_dvp);
1280 inum = dip->meta.iparent;
1284 error = hammer2_vfs_vget(ap->a_dvp->v_mount, NULL,
1295 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
1297 hammer2_inode_t *dip;
1298 hammer2_inode_t *nip;
1299 struct namecache *ncp;
1300 const uint8_t *name;
1306 dip = VTOI(ap->a_dvp);
1307 if (dip->pmp->ronly) {
1312 ncp = ap->a_nch->ncp;
1313 name = ncp->nc_name;
1314 name_len = ncp->nc_nlen;
1316 hammer2_pfs_memory_wait(dip->pmp);
1317 hammer2_trans_init(dip->pmp, 0);
1319 inum = hammer2_trans_newinum(dip->pmp);
1322 * Create the actual inode as a hidden file in the iroot, then
1323 * create the directory entry. The creation of the actual inode
1324 * sets its nlinks to 1 which is the value we desire.
1326 nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1331 error = hammer2_dirent_create(dip, name, name_len,
1332 nip->meta.inum, nip->meta.type);
1336 KKASSERT(nip == NULL);
1339 *ap->a_vpp = hammer2_igetv(nip, &error);
1340 hammer2_inode_unlock(nip);
1344 * Update dip's mtime
1349 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1350 hammer2_update_time(&mtime);
1351 hammer2_inode_modify(dip);
1352 dip->meta.mtime = mtime;
1353 hammer2_inode_unlock(dip);
1356 hammer2_trans_done(dip->pmp);
1359 cache_setunresolved(ap->a_nch);
1360 cache_setvp(ap->a_nch, *ap->a_vpp);
1368 hammer2_vop_open(struct vop_open_args *ap)
1370 return vop_stdopen(ap);
1374 * hammer2_vop_advlock { vp, id, op, fl, flags }
1378 hammer2_vop_advlock(struct vop_advlock_args *ap)
1380 hammer2_inode_t *ip = VTOI(ap->a_vp);
1383 size = ip->meta.size;
1384 return (lf_advlock(ap, &ip->advlock, size));
1389 hammer2_vop_close(struct vop_close_args *ap)
1391 return vop_stdclose(ap);
1395 * hammer2_vop_nlink { nch, dvp, vp, cred }
1397 * Create a hardlink from (vp) to {dvp, nch}.
1401 hammer2_vop_nlink(struct vop_nlink_args *ap)
1403 hammer2_inode_t *tdip; /* target directory to create link in */
1404 hammer2_inode_t *ip; /* inode we are hardlinking to */
1405 struct namecache *ncp;
1406 const uint8_t *name;
1410 if (ap->a_dvp->v_mount != ap->a_vp->v_mount)
1414 tdip = VTOI(ap->a_dvp);
1415 if (tdip->pmp->ronly) {
1420 ncp = ap->a_nch->ncp;
1421 name = ncp->nc_name;
1422 name_len = ncp->nc_nlen;
1425 * ip represents the file being hardlinked. The file could be a
1426 * normal file or a hardlink target if it has already been hardlinked.
1427 * (with the new semantics, it will almost always be a hardlink
1430 * Bump nlinks and potentially also create or move the hardlink
1431 * target in the parent directory common to (ip) and (tdip). The
1432 * consolidation code can modify ip->cluster. The returned cluster
1435 ip = VTOI(ap->a_vp);
1436 KASSERT(ip->pmp, ("ip->pmp is NULL %p %p", ip, ip->pmp));
1437 hammer2_pfs_memory_wait(ip->pmp);
1438 hammer2_trans_init(ip->pmp, 0);
1441 * Target should be an indexed inode or there's no way we will ever
1442 * be able to find it!
1444 KKASSERT((ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0);
1449 * Can return NULL and error == EXDEV if the common parent
1450 * crosses a directory with the xlink flag set.
1452 hammer2_inode_lock(tdip, 0);
1453 hammer2_inode_lock(ip, 0);
1456 * Create the directory entry and bump nlinks.
1459 error = hammer2_dirent_create(tdip, name, name_len,
1460 ip->meta.inum, ip->meta.type);
1461 hammer2_inode_modify(ip);
1466 * Update dip's mtime
1470 hammer2_update_time(&mtime);
1471 hammer2_inode_modify(tdip);
1472 tdip->meta.mtime = mtime;
1474 cache_setunresolved(ap->a_nch);
1475 cache_setvp(ap->a_nch, ap->a_vp);
1477 hammer2_inode_unlock(ip);
1478 hammer2_inode_unlock(tdip);
1480 hammer2_trans_done(ip->pmp);
1487 * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1489 * The operating system has already ensured that the directory entry
1490 * does not exist and done all appropriate namespace locking.
1494 hammer2_vop_ncreate(struct vop_ncreate_args *ap)
1496 hammer2_inode_t *dip;
1497 hammer2_inode_t *nip;
1498 struct namecache *ncp;
1499 const uint8_t *name;
1505 dip = VTOI(ap->a_dvp);
1506 if (dip->pmp->ronly) {
1511 ncp = ap->a_nch->ncp;
1512 name = ncp->nc_name;
1513 name_len = ncp->nc_nlen;
1514 hammer2_pfs_memory_wait(dip->pmp);
1515 hammer2_trans_init(dip->pmp, 0);
1517 inum = hammer2_trans_newinum(dip->pmp);
1520 * Create the actual inode as a hidden file in the iroot, then
1521 * create the directory entry. The creation of the actual inode
1522 * sets its nlinks to 1 which is the value we desire.
1524 nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1530 error = hammer2_dirent_create(dip, name, name_len,
1531 nip->meta.inum, nip->meta.type);
1534 KKASSERT(nip == NULL);
1537 *ap->a_vpp = hammer2_igetv(nip, &error);
1538 hammer2_inode_unlock(nip);
1542 * Update dip's mtime
1547 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1548 hammer2_update_time(&mtime);
1549 hammer2_inode_modify(dip);
1550 dip->meta.mtime = mtime;
1551 hammer2_inode_unlock(dip);
1554 hammer2_trans_done(dip->pmp);
1557 cache_setunresolved(ap->a_nch);
1558 cache_setvp(ap->a_nch, *ap->a_vpp);
1565 * Make a device node (typically a fifo)
1569 hammer2_vop_nmknod(struct vop_nmknod_args *ap)
1571 hammer2_inode_t *dip;
1572 hammer2_inode_t *nip;
1573 struct namecache *ncp;
1574 const uint8_t *name;
1580 dip = VTOI(ap->a_dvp);
1581 if (dip->pmp->ronly) {
1586 ncp = ap->a_nch->ncp;
1587 name = ncp->nc_name;
1588 name_len = ncp->nc_nlen;
1589 hammer2_pfs_memory_wait(dip->pmp);
1590 hammer2_trans_init(dip->pmp, 0);
1593 * Create the device inode and then create the directory entry.
1595 inum = hammer2_trans_newinum(dip->pmp);
1596 nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1601 error = hammer2_dirent_create(dip, name, name_len,
1602 nip->meta.inum, nip->meta.type);
1607 KKASSERT(nip == NULL);
1610 *ap->a_vpp = hammer2_igetv(nip, &error);
1611 hammer2_inode_unlock(nip);
1615 * Update dip's mtime
1620 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1621 hammer2_update_time(&mtime);
1622 hammer2_inode_modify(dip);
1623 dip->meta.mtime = mtime;
1624 hammer2_inode_unlock(dip);
1627 hammer2_trans_done(dip->pmp);
1630 cache_setunresolved(ap->a_nch);
1631 cache_setvp(ap->a_nch, *ap->a_vpp);
1638 * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1642 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
1644 hammer2_inode_t *dip;
1645 hammer2_inode_t *nip;
1646 struct namecache *ncp;
1647 const uint8_t *name;
1652 dip = VTOI(ap->a_dvp);
1653 if (dip->pmp->ronly)
1656 ncp = ap->a_nch->ncp;
1657 name = ncp->nc_name;
1658 name_len = ncp->nc_nlen;
1659 hammer2_pfs_memory_wait(dip->pmp);
1660 hammer2_trans_init(dip->pmp, 0);
1662 ap->a_vap->va_type = VLNK; /* enforce type */
1665 * Create the softlink as an inode and then create the directory
1668 inum = hammer2_trans_newinum(dip->pmp);
1670 nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1675 error = hammer2_dirent_create(dip, name, name_len,
1676 nip->meta.inum, nip->meta.type);
1681 KKASSERT(nip == NULL);
1683 hammer2_trans_done(dip->pmp);
1686 *ap->a_vpp = hammer2_igetv(nip, &error);
1689 * Build the softlink (~like file data) and finalize the namecache.
1696 bytes = strlen(ap->a_target);
1698 hammer2_inode_unlock(nip);
1699 bzero(&auio, sizeof(auio));
1700 bzero(&aiov, sizeof(aiov));
1701 auio.uio_iov = &aiov;
1702 auio.uio_segflg = UIO_SYSSPACE;
1703 auio.uio_rw = UIO_WRITE;
1704 auio.uio_resid = bytes;
1705 auio.uio_iovcnt = 1;
1706 auio.uio_td = curthread;
1707 aiov.iov_base = ap->a_target;
1708 aiov.iov_len = bytes;
1709 error = hammer2_write_file(nip, &auio, IO_APPEND, 0);
1710 /* XXX handle error */
1713 hammer2_inode_unlock(nip);
1717 * Update dip's mtime
1722 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1723 hammer2_update_time(&mtime);
1724 hammer2_inode_modify(dip);
1725 dip->meta.mtime = mtime;
1726 hammer2_inode_unlock(dip);
1729 hammer2_trans_done(dip->pmp);
1732 * Finalize namecache
1735 cache_setunresolved(ap->a_nch);
1736 cache_setvp(ap->a_nch, *ap->a_vpp);
1737 /* hammer2_knote(ap->a_dvp, NOTE_WRITE); */
1743 * hammer2_vop_nremove { nch, dvp, cred }
1747 hammer2_vop_nremove(struct vop_nremove_args *ap)
1749 hammer2_xop_unlink_t *xop;
1750 hammer2_inode_t *dip;
1751 hammer2_inode_t *ip;
1752 struct namecache *ncp;
1757 dip = VTOI(ap->a_dvp);
1758 if (dip->pmp->ronly) {
1763 ncp = ap->a_nch->ncp;
1765 hammer2_pfs_memory_wait(dip->pmp);
1766 hammer2_trans_init(dip->pmp, 0);
1767 hammer2_inode_lock(dip, 0);
1770 * The unlink XOP unlinks the path from the directory and
1771 * locates and returns the cluster associated with the real inode.
1772 * We have to handle nlinks here on the frontend.
1774 xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1775 hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1776 isopen = cache_isopen(ap->a_nch);
1778 xop->dopermanent = 0;
1779 hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1782 * Collect the real inode and adjust nlinks, destroy the real
1783 * inode if nlinks transitions to 0 and it was the real inode
1784 * (else it has already been removed).
1786 error = hammer2_xop_collect(&xop->head, 0);
1787 hammer2_inode_unlock(dip);
1790 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1791 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1793 hammer2_inode_unlink_finisher(ip, isopen);
1794 hammer2_inode_unlock(ip);
1797 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1801 * Update dip's mtime
1806 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1807 hammer2_update_time(&mtime);
1808 hammer2_inode_modify(dip);
1809 dip->meta.mtime = mtime;
1810 hammer2_inode_unlock(dip);
1813 hammer2_inode_run_sideq(dip->pmp);
1814 hammer2_trans_done(dip->pmp);
1816 cache_unlink(ap->a_nch);
1822 * hammer2_vop_nrmdir { nch, dvp, cred }
1826 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
1828 hammer2_xop_unlink_t *xop;
1829 hammer2_inode_t *dip;
1830 hammer2_inode_t *ip;
1831 struct namecache *ncp;
1836 dip = VTOI(ap->a_dvp);
1837 if (dip->pmp->ronly) {
1842 hammer2_pfs_memory_wait(dip->pmp);
1843 hammer2_trans_init(dip->pmp, 0);
1844 hammer2_inode_lock(dip, 0);
1846 xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1848 ncp = ap->a_nch->ncp;
1849 hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1850 isopen = cache_isopen(ap->a_nch);
1852 xop->dopermanent = 0;
1853 hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1856 * Collect the real inode and adjust nlinks, destroy the real
1857 * inode if nlinks transitions to 0 and it was the real inode
1858 * (else it has already been removed).
1860 error = hammer2_xop_collect(&xop->head, 0);
1861 hammer2_inode_unlock(dip);
1864 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1865 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1867 hammer2_inode_unlink_finisher(ip, isopen);
1868 hammer2_inode_unlock(ip);
1871 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1875 * Update dip's mtime
1880 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1881 hammer2_update_time(&mtime);
1882 hammer2_inode_modify(dip);
1883 dip->meta.mtime = mtime;
1884 hammer2_inode_unlock(dip);
1887 hammer2_inode_run_sideq(dip->pmp);
1888 hammer2_trans_done(dip->pmp);
1890 cache_unlink(ap->a_nch);
1896 * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1900 hammer2_vop_nrename(struct vop_nrename_args *ap)
1902 struct namecache *fncp;
1903 struct namecache *tncp;
1904 hammer2_inode_t *fdip;
1905 hammer2_inode_t *tdip;
1906 hammer2_inode_t *ip;
1907 const uint8_t *fname;
1909 const uint8_t *tname;
1917 if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
1919 if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
1922 fdip = VTOI(ap->a_fdvp); /* source directory */
1923 tdip = VTOI(ap->a_tdvp); /* target directory */
1925 if (fdip->pmp->ronly)
1929 fncp = ap->a_fnch->ncp; /* entry name in source */
1930 fname = fncp->nc_name;
1931 fname_len = fncp->nc_nlen;
1933 tncp = ap->a_tnch->ncp; /* entry name in target */
1934 tname = tncp->nc_name;
1935 tname_len = tncp->nc_nlen;
1937 hammer2_pfs_memory_wait(tdip->pmp);
1938 hammer2_trans_init(tdip->pmp, 0);
1944 * ip is the inode being renamed. If this is a hardlink then
1945 * ip represents the actual file and not the hardlink marker.
1947 ip = VTOI(fncp->nc_vp);
1949 KKASSERT((ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0);
1952 * Can return NULL and error == EXDEV if the common parent
1953 * crosses a directory with the xlink flag set.
1956 hammer2_inode_lock(fdip, 0);
1957 hammer2_inode_lock(tdip, 0);
1958 hammer2_inode_ref(ip); /* extra ref */
1960 hammer2_inode_lock(ip, 0);
1963 * Delete the target namespace.
1966 hammer2_xop_unlink_t *xop2;
1967 hammer2_inode_t *tip;
1971 * The unlink XOP unlinks the path from the directory and
1972 * locates and returns the cluster associated with the real
1973 * inode. We have to handle nlinks here on the frontend.
1975 xop2 = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
1976 hammer2_xop_setname(&xop2->head, tname, tname_len);
1977 isopen = cache_isopen(ap->a_tnch);
1979 xop2->dopermanent = 0;
1980 hammer2_xop_start(&xop2->head, hammer2_xop_unlink);
1983 * Collect the real inode and adjust nlinks, destroy the real
1984 * inode if nlinks transitions to 0 and it was the real inode
1985 * (else it has already been removed).
1987 tnch_error = hammer2_xop_collect(&xop2->head, 0);
1988 /* hammer2_inode_unlock(tdip); */
1990 if (tnch_error == 0) {
1991 tip = hammer2_inode_get(tdip->pmp, NULL,
1992 &xop2->head.cluster, -1);
1993 hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
1995 hammer2_inode_unlink_finisher(tip, isopen);
1996 hammer2_inode_unlock(tip);
1999 hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
2001 /* hammer2_inode_lock(tdip, 0); */
2003 if (tnch_error && tnch_error != ENOENT) {
2011 * Resolve the collision space for (tdip, tname, tname_len)
2013 * tdip must be held exclusively locked to prevent races.
2016 hammer2_xop_scanlhc_t *sxop;
2017 hammer2_tid_t lhcbase;
2019 tlhc = hammer2_dirhash(tname, tname_len);
2021 sxop = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
2023 hammer2_xop_start(&sxop->head, hammer2_xop_scanlhc);
2024 while ((error = hammer2_xop_collect(&sxop->head, 0)) == 0) {
2025 if (tlhc != sxop->head.cluster.focus->bref.key)
2029 hammer2_xop_retire(&sxop->head, HAMMER2_XOPMASK_VOP);
2032 if (error != ENOENT)
2037 if ((lhcbase ^ tlhc) & ~HAMMER2_DIRHASH_LOMASK) {
2044 * Everything is setup, do the rename.
2046 * We have to synchronize ip->meta to the underlying operation.
2048 * NOTE: To avoid deadlocks we cannot lock (ip) while we are
2049 * unlinking elements from their directories. Locking
2050 * the nlinks field does not lock the whole inode.
2052 /* hammer2_inode_lock(ip, 0); */
2054 hammer2_xop_nrename_t *xop4;
2056 xop4 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
2058 xop4->ip_key = ip->meta.name_key;
2059 hammer2_xop_setip2(&xop4->head, ip);
2060 hammer2_xop_setip3(&xop4->head, tdip);
2061 hammer2_xop_setname(&xop4->head, fname, fname_len);
2062 hammer2_xop_setname2(&xop4->head, tname, tname_len);
2063 hammer2_xop_start(&xop4->head, hammer2_xop_nrename);
2065 error = hammer2_xop_collect(&xop4->head, 0);
2066 hammer2_xop_retire(&xop4->head, HAMMER2_XOPMASK_VOP);
2068 if (error == ENOENT)
2071 (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
2072 hammer2_inode_modify(ip);
2073 ip->meta.name_len = tname_len;
2074 ip->meta.name_key = tlhc;
2083 * Update directory mtimes to represent the something changed.
2085 if (update_fdip || update_tdip) {
2088 hammer2_update_time(&mtime);
2090 hammer2_inode_modify(fdip);
2091 fdip->meta.mtime = mtime;
2094 hammer2_inode_modify(tdip);
2095 tdip->meta.mtime = mtime;
2098 hammer2_inode_unlock(ip);
2099 hammer2_inode_unlock(tdip);
2100 hammer2_inode_unlock(fdip);
2101 hammer2_inode_drop(ip);
2102 hammer2_inode_run_sideq(fdip->pmp);
2104 hammer2_trans_done(tdip->pmp);
2107 * Issue the namecache update after unlocking all the internal
2108 * hammer structures, otherwise we might deadlock.
2110 if (tnch_error == 0) {
2111 cache_unlink(ap->a_tnch);
2112 cache_setunresolved(ap->a_tnch);
2115 cache_rename(ap->a_fnch, ap->a_tnch);
2122 * hammer2_vop_ioctl { vp, command, data, fflag, cred }
2126 hammer2_vop_ioctl(struct vop_ioctl_args *ap)
2128 hammer2_inode_t *ip;
2132 ip = VTOI(ap->a_vp);
2134 error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
2135 ap->a_fflag, ap->a_cred);
2142 hammer2_vop_mountctl(struct vop_mountctl_args *ap)
2150 case (MOUNTCTL_SET_EXPORT):
2151 mp = ap->a_head.a_ops->head.vv_mount;
2154 if (ap->a_ctllen != sizeof(struct export_args))
2157 rc = vfs_export(mp, &pmp->export,
2158 (const struct export_args *)ap->a_ctl);
2161 rc = vop_stdmountctl(ap);
2171 static void filt_hammer2detach(struct knote *kn);
2172 static int filt_hammer2read(struct knote *kn, long hint);
2173 static int filt_hammer2write(struct knote *kn, long hint);
2174 static int filt_hammer2vnode(struct knote *kn, long hint);
2176 static struct filterops hammer2read_filtops =
2177 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2178 NULL, filt_hammer2detach, filt_hammer2read };
2179 static struct filterops hammer2write_filtops =
2180 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2181 NULL, filt_hammer2detach, filt_hammer2write };
2182 static struct filterops hammer2vnode_filtops =
2183 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2184 NULL, filt_hammer2detach, filt_hammer2vnode };
2188 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap)
2190 struct vnode *vp = ap->a_vp;
2191 struct knote *kn = ap->a_kn;
2193 switch (kn->kn_filter) {
2195 kn->kn_fop = &hammer2read_filtops;
2198 kn->kn_fop = &hammer2write_filtops;
2201 kn->kn_fop = &hammer2vnode_filtops;
2204 return (EOPNOTSUPP);
2207 kn->kn_hook = (caddr_t)vp;
2209 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2215 filt_hammer2detach(struct knote *kn)
2217 struct vnode *vp = (void *)kn->kn_hook;
2219 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2223 filt_hammer2read(struct knote *kn, long hint)
2225 struct vnode *vp = (void *)kn->kn_hook;
2226 hammer2_inode_t *ip = VTOI(vp);
2229 if (hint == NOTE_REVOKE) {
2230 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2233 off = ip->meta.size - kn->kn_fp->f_offset;
2234 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2235 if (kn->kn_sfflags & NOTE_OLDAPI)
2237 return (kn->kn_data != 0);
2242 filt_hammer2write(struct knote *kn, long hint)
2244 if (hint == NOTE_REVOKE)
2245 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2251 filt_hammer2vnode(struct knote *kn, long hint)
2253 if (kn->kn_sfflags & hint)
2254 kn->kn_fflags |= hint;
2255 if (hint == NOTE_REVOKE) {
2256 kn->kn_flags |= (EV_EOF | EV_NODATA);
2259 return (kn->kn_fflags != 0);
2267 hammer2_vop_markatime(struct vop_markatime_args *ap)
2269 hammer2_inode_t *ip;
2282 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap)
2286 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2288 error = hammer2_vop_kqfilter(ap);
2295 struct vop_ops hammer2_vnode_vops = {
2296 .vop_default = vop_defaultop,
2297 .vop_fsync = hammer2_vop_fsync,
2298 .vop_getpages = vop_stdgetpages,
2299 .vop_putpages = vop_stdputpages,
2300 .vop_access = hammer2_vop_access,
2301 .vop_advlock = hammer2_vop_advlock,
2302 .vop_close = hammer2_vop_close,
2303 .vop_nlink = hammer2_vop_nlink,
2304 .vop_ncreate = hammer2_vop_ncreate,
2305 .vop_nsymlink = hammer2_vop_nsymlink,
2306 .vop_nremove = hammer2_vop_nremove,
2307 .vop_nrmdir = hammer2_vop_nrmdir,
2308 .vop_nrename = hammer2_vop_nrename,
2309 .vop_getattr = hammer2_vop_getattr,
2310 .vop_setattr = hammer2_vop_setattr,
2311 .vop_readdir = hammer2_vop_readdir,
2312 .vop_readlink = hammer2_vop_readlink,
2313 .vop_getpages = vop_stdgetpages,
2314 .vop_putpages = vop_stdputpages,
2315 .vop_read = hammer2_vop_read,
2316 .vop_write = hammer2_vop_write,
2317 .vop_open = hammer2_vop_open,
2318 .vop_inactive = hammer2_vop_inactive,
2319 .vop_reclaim = hammer2_vop_reclaim,
2320 .vop_nresolve = hammer2_vop_nresolve,
2321 .vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
2322 .vop_nmkdir = hammer2_vop_nmkdir,
2323 .vop_nmknod = hammer2_vop_nmknod,
2324 .vop_ioctl = hammer2_vop_ioctl,
2325 .vop_mountctl = hammer2_vop_mountctl,
2326 .vop_bmap = hammer2_vop_bmap,
2327 .vop_strategy = hammer2_vop_strategy,
2328 .vop_kqfilter = hammer2_vop_kqfilter
2331 struct vop_ops hammer2_spec_vops = {
2332 .vop_default = vop_defaultop,
2333 .vop_fsync = hammer2_vop_fsync,
2334 .vop_read = vop_stdnoread,
2335 .vop_write = vop_stdnowrite,
2336 .vop_access = hammer2_vop_access,
2337 .vop_close = hammer2_vop_close,
2338 .vop_markatime = hammer2_vop_markatime,
2339 .vop_getattr = hammer2_vop_getattr,
2340 .vop_inactive = hammer2_vop_inactive,
2341 .vop_reclaim = hammer2_vop_reclaim,
2342 .vop_setattr = hammer2_vop_setattr
2345 struct vop_ops hammer2_fifo_vops = {
2346 .vop_default = fifo_vnoperate,
2347 .vop_fsync = hammer2_vop_fsync,
2349 .vop_read = hammer2_vop_fiforead,
2350 .vop_write = hammer2_vop_fifowrite,
2352 .vop_access = hammer2_vop_access,
2354 .vop_close = hammer2_vop_fifoclose,
2356 .vop_markatime = hammer2_vop_markatime,
2357 .vop_getattr = hammer2_vop_getattr,
2358 .vop_inactive = hammer2_vop_inactive,
2359 .vop_reclaim = hammer2_vop_reclaim,
2360 .vop_setattr = hammer2_vop_setattr,
2361 .vop_kqfilter = hammer2_vop_fifokqfilter