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)
101 * Check for deleted inodes and recycle immediately on the last
102 * release. Be sure to destroy any left-over buffer cache buffers
103 * so we do not waste time trying to flush them.
105 * Note that deleting the file block chains under the inode chain
106 * would just be a waste of energy, so don't do it.
108 * WARNING: nvtruncbuf() can only be safely called without the inode
109 * lock held due to the way our write thread works.
111 if (ip->flags & HAMMER2_INODE_ISUNLINKED) {
116 * Detect updates to the embedded data which may be
117 * synchronized by the strategy code. Simply mark the
118 * inode modified so it gets picked up by our normal flush.
120 nblksize = hammer2_calc_logical(ip, 0, &lbase, NULL);
121 nvtruncbuf(vp, 0, nblksize, 0, 0);
128 * Reclaim a vnode so that it can be reused; after the inode is
129 * disassociated, the filesystem must manage it alone.
133 hammer2_vop_reclaim(struct vop_reclaim_args *ap)
147 * The final close of a deleted file or directory marks it for
148 * destruction. The DELETED flag allows the flusher to shortcut
149 * any modified blocks still unflushed (that is, just ignore them).
151 * HAMMER2 usually does not try to optimize the freemap by returning
152 * deleted blocks to it as it does not usually know how many snapshots
153 * might be referencing portions of the file/dir.
159 * NOTE! We do not attempt to flush chains here, flushing is
160 * really fragile and could also deadlock.
165 * A modified inode may require chain synchronization. This
166 * synchronization is usually handled by VOP_SNYC / VOP_FSYNC
167 * when vfsync() is called. However, that requires a vnode.
169 * When the vnode is disassociated we must keep track of any modified
170 * inode via the sideq so that it is properly flushed. We cannot
171 * safely synchronize the inode from inside the reclaim due to
172 * potentially deep locks held as-of when the reclaim occurs.
173 * Interactions and potential deadlocks abound.
175 if ((ip->flags & (HAMMER2_INODE_ISUNLINKED |
176 HAMMER2_INODE_MODIFIED |
177 HAMMER2_INODE_RESIZED)) &&
178 (ip->flags & HAMMER2_INODE_ISDELETED) == 0) {
179 hammer2_inode_sideq_t *ipul;
181 ipul = kmalloc(sizeof(*ipul), pmp->minode, M_WAITOK | M_ZERO);
184 hammer2_spin_ex(&pmp->list_spin);
185 if ((ip->flags & HAMMER2_INODE_ONSIDEQ) == 0) {
187 atomic_set_int(&ip->flags, HAMMER2_INODE_ONSIDEQ);
188 TAILQ_INSERT_TAIL(&pmp->sideq, ipul, entry);
189 hammer2_spin_unex(&pmp->list_spin);
191 hammer2_spin_unex(&pmp->list_spin);
192 kfree(ipul, pmp->minode);
193 hammer2_inode_drop(ip); /* vp ref */
195 /* retain ref from vp for ipul */
197 hammer2_inode_drop(ip); /* vp ref */
201 * XXX handle background sync when ip dirty, kernel will no longer
202 * notify us regarding this inode because there is no longer a
203 * vnode attached to it.
211 hammer2_vop_fsync(struct vop_fsync_args *ap)
220 /* XXX can't do this yet */
221 hammer2_trans_init(ip->pmp, HAMMER2_TRANS_ISFLUSH);
222 vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
224 hammer2_trans_init(ip->pmp, 0);
225 vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
228 * Calling chain_flush here creates a lot of duplicative
229 * COW operations due to non-optimal vnode ordering.
231 * Only do it for an actual fsync() syscall. The other forms
232 * which call this function will eventually call chain_flush
233 * on the volume root as a catch-all, which is far more optimal.
235 hammer2_inode_lock(ip, 0);
236 if (ip->flags & HAMMER2_INODE_MODIFIED)
237 hammer2_inode_chain_sync(ip);
238 hammer2_inode_unlock(ip);
239 hammer2_trans_done(ip->pmp);
246 hammer2_vop_access(struct vop_access_args *ap)
248 hammer2_inode_t *ip = VTOI(ap->a_vp);
253 hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
254 uid = hammer2_to_unix_xid(&ip->meta.uid);
255 gid = hammer2_to_unix_xid(&ip->meta.gid);
256 error = vop_helper_access(ap, uid, gid, ip->meta.mode, ip->meta.uflags);
257 hammer2_inode_unlock(ip);
264 hammer2_vop_getattr(struct vop_getattr_args *ap)
270 hammer2_chain_t *chain;
279 hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
281 vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0];
282 vap->va_fileid = ip->meta.inum;
283 vap->va_mode = ip->meta.mode;
284 vap->va_nlink = ip->meta.nlinks;
285 vap->va_uid = hammer2_to_unix_xid(&ip->meta.uid);
286 vap->va_gid = hammer2_to_unix_xid(&ip->meta.gid);
289 vap->va_size = ip->meta.size; /* protected by shared lock */
290 vap->va_blocksize = HAMMER2_PBUFSIZE;
291 vap->va_flags = ip->meta.uflags;
292 hammer2_time_to_timespec(ip->meta.ctime, &vap->va_ctime);
293 hammer2_time_to_timespec(ip->meta.mtime, &vap->va_mtime);
294 hammer2_time_to_timespec(ip->meta.mtime, &vap->va_atime);
297 if (ip->meta.type == HAMMER2_OBJTYPE_DIRECTORY) {
299 * Can't really calculate directory use sans the files under
300 * it, just assume one block for now.
302 vap->va_bytes += HAMMER2_INODE_BYTES;
304 for (i = 0; i < ip->cluster.nchains; ++i) {
305 if ((chain = ip->cluster.array[i].chain) != NULL) {
307 chain->bref.embed.stats.data_count) {
309 chain->bref.embed.stats.data_count;
314 vap->va_type = hammer2_get_vtype(ip->meta.type);
316 vap->va_uid_uuid = ip->meta.uid;
317 vap->va_gid_uuid = ip->meta.gid;
318 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
321 hammer2_inode_unlock(ip);
328 hammer2_vop_setattr(struct vop_setattr_args *ap)
339 hammer2_update_time(&ctime);
345 if (hammer2_vfs_enospace(ip, 0, ap->a_cred) > 1)
348 hammer2_pfs_memory_wait(ip->pmp);
349 hammer2_trans_init(ip->pmp, 0);
350 hammer2_inode_lock(ip, 0);
353 if (vap->va_flags != VNOVAL) {
356 flags = ip->meta.uflags;
357 error = vop_helper_setattr_flags(&flags, vap->va_flags,
358 hammer2_to_unix_xid(&ip->meta.uid),
361 if (ip->meta.uflags != flags) {
362 hammer2_inode_modify(ip);
363 ip->meta.uflags = flags;
364 ip->meta.ctime = ctime;
365 kflags |= NOTE_ATTRIB;
367 if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
374 if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
378 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
379 mode_t cur_mode = ip->meta.mode;
380 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
381 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
385 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
387 &cur_uid, &cur_gid, &cur_mode);
389 hammer2_guid_to_uuid(&uuid_uid, cur_uid);
390 hammer2_guid_to_uuid(&uuid_gid, cur_gid);
391 if (bcmp(&uuid_uid, &ip->meta.uid, sizeof(uuid_uid)) ||
392 bcmp(&uuid_gid, &ip->meta.gid, sizeof(uuid_gid)) ||
393 ip->meta.mode != cur_mode
395 hammer2_inode_modify(ip);
396 ip->meta.uid = uuid_uid;
397 ip->meta.gid = uuid_gid;
398 ip->meta.mode = cur_mode;
399 ip->meta.ctime = ctime;
401 kflags |= NOTE_ATTRIB;
408 if (vap->va_size != VNOVAL && ip->meta.size != vap->va_size) {
411 if (vap->va_size == ip->meta.size)
413 if (vap->va_size < ip->meta.size) {
414 hammer2_mtx_ex(&ip->truncate_lock);
415 hammer2_truncate_file(ip, vap->va_size);
416 hammer2_mtx_unlock(&ip->truncate_lock);
417 kflags |= NOTE_WRITE;
419 hammer2_extend_file(ip, vap->va_size);
420 kflags |= NOTE_WRITE | NOTE_EXTEND;
422 hammer2_inode_modify(ip);
423 ip->meta.mtime = ctime;
431 /* atime not supported */
432 if (vap->va_atime.tv_sec != VNOVAL) {
433 hammer2_inode_modify(ip);
434 ip->meta.atime = hammer2_timespec_to_time(&vap->va_atime);
435 kflags |= NOTE_ATTRIB;
438 if (vap->va_mode != (mode_t)VNOVAL) {
439 mode_t cur_mode = ip->meta.mode;
440 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
441 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
443 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
444 cur_uid, cur_gid, &cur_mode);
445 if (error == 0 && ip->meta.mode != cur_mode) {
446 hammer2_inode_modify(ip);
447 ip->meta.mode = cur_mode;
448 ip->meta.ctime = ctime;
449 kflags |= NOTE_ATTRIB;
453 if (vap->va_mtime.tv_sec != VNOVAL) {
454 hammer2_inode_modify(ip);
455 ip->meta.mtime = hammer2_timespec_to_time(&vap->va_mtime);
456 kflags |= NOTE_ATTRIB;
461 * If a truncation occurred we must call inode_fsync() now in order
462 * to trim the related data chains, otherwise a later expansion can
465 * If an extend occured that changed the DIRECTDATA state, we must
466 * call inode_fsync now in order to prepare the inode's indirect
469 if (ip->flags & HAMMER2_INODE_RESIZED)
470 hammer2_inode_chain_sync(ip);
475 hammer2_inode_unlock(ip);
476 hammer2_trans_done(ip->pmp);
477 hammer2_knote(ip->vp, kflags);
484 hammer2_vop_readdir(struct vop_readdir_args *ap)
486 hammer2_xop_readdir_t *xop;
487 hammer2_blockref_t bref;
502 saveoff = uio->uio_offset;
507 * Setup cookies directory entry cookies if requested
509 if (ap->a_ncookies) {
510 ncookies = uio->uio_resid / 16 + 1;
513 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
520 hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
523 * Handle artificial entries. To ensure that only positive 64 bit
524 * quantities are returned to userland we always strip off bit 63.
525 * The hash code is designed such that codes 0x0000-0x7FFF are not
526 * used, allowing us to use these codes for articial entries.
528 * Entry 0 is used for '.' and entry 1 is used for '..'. Do not
529 * allow '..' to cross the mount point into (e.g.) the super-root.
532 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
533 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
537 cookies[cookie_index] = saveoff;
540 if (cookie_index == ncookies)
546 * Be careful with lockorder when accessing ".."
548 * (ip is the current dir. xip is the parent dir).
550 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
551 if (ip != ip->pmp->iroot)
552 inum = ip->meta.iparent & HAMMER2_DIRHASH_USERMSK;
553 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
557 cookies[cookie_index] = saveoff;
560 if (cookie_index == ncookies)
564 lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
565 if (hammer2_debug & 0x0020)
566 kprintf("readdir: lkey %016jx\n", lkey);
571 * Use XOP for cluster scan.
573 * parent is the inode cluster, already locked for us. Don't
574 * double lock shared locks as this will screw up upgrades.
576 xop = hammer2_xop_alloc(ip, 0);
578 hammer2_xop_start(&xop->head, hammer2_xop_readdir);
581 const hammer2_inode_data_t *ripdata;
585 error = hammer2_xop_collect(&xop->head, 0);
586 error = hammer2_error_to_errno(error);
590 if (cookie_index == ncookies)
592 if (hammer2_debug & 0x0020)
593 kprintf("cluster chain %p %p\n",
594 xop->head.cluster.focus,
595 (xop->head.cluster.focus ?
596 xop->head.cluster.focus->data : (void *)-1));
597 hammer2_cluster_bref(&xop->head.cluster, &bref);
599 if (bref.type == HAMMER2_BREF_TYPE_INODE) {
601 &hammer2_cluster_rdata(&xop->head.cluster)->ipdata;
602 dtype = hammer2_get_dtype(ripdata->meta.type);
603 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
604 r = vop_write_dirent(&error, uio,
606 HAMMER2_DIRHASH_USERMSK,
608 ripdata->meta.name_len,
613 cookies[cookie_index] = saveoff;
615 } else if (bref.type == HAMMER2_BREF_TYPE_DIRENT) {
616 dtype = hammer2_get_dtype(bref.embed.dirent.type);
617 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
618 if (bref.embed.dirent.namlen <=
619 sizeof(bref.check.buf)) {
620 dname = bref.check.buf;
623 hammer2_cluster_rdata(&xop->head.cluster)->buf;
625 r = vop_write_dirent(&error, uio,
626 bref.embed.dirent.inum,
628 bref.embed.dirent.namlen,
633 cookies[cookie_index] = saveoff;
636 /* XXX chain error */
637 kprintf("bad chain type readdir %d\n", bref.type);
640 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
641 if (error == ENOENT) {
644 saveoff = (hammer2_key_t)-1;
646 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
649 hammer2_inode_unlock(ip);
651 *ap->a_eofflag = eofflag;
652 if (hammer2_debug & 0x0020)
653 kprintf("readdir: done at %016jx\n", saveoff);
654 uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
655 if (error && cookie_index == 0) {
657 kfree(cookies, M_TEMP);
659 *ap->a_cookies = NULL;
663 *ap->a_ncookies = cookie_index;
664 *ap->a_cookies = cookies;
671 * hammer2_vop_readlink { vp, uio, cred }
675 hammer2_vop_readlink(struct vop_readlink_args *ap)
682 if (vp->v_type != VLNK)
686 error = hammer2_read_file(ip, ap->a_uio, 0);
692 hammer2_vop_read(struct vop_read_args *ap)
702 * Read operations supported on this vnode?
705 if (vp->v_type != VREG)
715 seqcount = ap->a_ioflag >> 16;
716 bigread = (uio->uio_resid > 100 * 1024 * 1024);
718 error = hammer2_read_file(ip, uio, seqcount);
724 hammer2_vop_write(struct vop_write_args *ap)
735 * Read operations supported on this vnode?
738 if (vp->v_type != VREG)
745 ioflag = ap->a_ioflag;
750 switch (hammer2_vfs_enospace(ip, uio->uio_resid, ap->a_cred)) {
754 ioflag |= IO_DIRECT; /* semi-synchronous */
760 seqcount = ioflag >> 16;
763 * Check resource limit
765 if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc &&
766 uio->uio_offset + uio->uio_resid >
767 td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
768 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
773 * The transaction interlocks against flush initiations
774 * (note: but will run concurrently with the actual flush).
776 * To avoid deadlocking against the VM system, we must flag any
777 * transaction related to the buffer cache or other direct
778 * VM page manipulation.
780 if (uio->uio_segflg == UIO_NOCOPY)
781 hammer2_trans_init(ip->pmp, HAMMER2_TRANS_BUFCACHE);
783 hammer2_trans_init(ip->pmp, 0);
784 error = hammer2_write_file(ip, uio, ioflag, seqcount);
785 hammer2_trans_done(ip->pmp);
791 * Perform read operations on a file or symlink given an UNLOCKED
794 * The passed ip is not locked.
798 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
809 * WARNING! Assumes that the kernel interlocks size changes at the
812 hammer2_mtx_sh(&ip->lock);
813 hammer2_mtx_sh(&ip->truncate_lock);
814 size = ip->meta.size;
815 hammer2_mtx_unlock(&ip->lock);
817 while (uio->uio_resid > 0 && uio->uio_offset < size) {
824 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
828 error = cluster_read(ip->vp, leof, lbase, lblksize,
829 uio->uio_resid, seqcount * MAXBSIZE,
832 if (uio->uio_segflg == UIO_NOCOPY) {
833 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
834 if (bp->b_flags & B_CACHE) {
837 if (bp->b_xio.xio_npages != 16)
838 kprintf("NPAGES BAD\n");
839 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
841 m = bp->b_xio.xio_pages[i];
842 if (m == NULL || m->valid == 0) {
843 kprintf("bp %016jx %016jx pg %d inv",
846 kprintf("m->object %p/%p", m->object, ip->vp->v_object);
852 kprintf("b_flags %08x, b_error %d\n", bp->b_flags, bp->b_error);
856 error = bread(ip->vp, lbase, lblksize, &bp);
862 loff = (int)(uio->uio_offset - lbase);
864 if (n > uio->uio_resid)
866 if (n > size - uio->uio_offset)
867 n = (int)(size - uio->uio_offset);
868 bp->b_flags |= B_AGE;
869 uiomovebp(bp, (char *)bp->b_data + loff, n, uio);
872 hammer2_mtx_unlock(&ip->truncate_lock);
878 * Write to the file represented by the inode via the logical buffer cache.
879 * The inode may represent a regular file or a symlink.
881 * The inode must not be locked.
885 hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
886 int ioflag, int seqcount)
888 hammer2_key_t old_eof;
889 hammer2_key_t new_eof;
898 * WARNING! Assumes that the kernel interlocks size changes at the
901 hammer2_mtx_ex(&ip->lock);
902 hammer2_mtx_sh(&ip->truncate_lock);
903 if (ioflag & IO_APPEND)
904 uio->uio_offset = ip->meta.size;
905 old_eof = ip->meta.size;
908 * Extend the file if necessary. If the write fails at some point
909 * we will truncate it back down to cover as much as we were able
912 * Doing this now makes it easier to calculate buffer sizes in
919 if (uio->uio_offset + uio->uio_resid > old_eof) {
920 new_eof = uio->uio_offset + uio->uio_resid;
922 hammer2_extend_file(ip, new_eof);
923 kflags |= NOTE_EXTEND;
927 hammer2_mtx_unlock(&ip->lock);
932 while (uio->uio_resid > 0) {
941 * Don't allow the buffer build to blow out the buffer
944 if ((ioflag & IO_RECURSE) == 0)
945 bwillwrite(HAMMER2_PBUFSIZE);
948 * This nominally tells us how much we can cluster and
949 * what the logical buffer size needs to be. Currently
950 * we don't try to cluster the write and just handle one
953 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
955 loff = (int)(uio->uio_offset - lbase);
957 KKASSERT(lblksize <= 65536);
960 * Calculate bytes to copy this transfer and whether the
961 * copy completely covers the buffer or not.
965 if (n > uio->uio_resid) {
967 if (loff == lbase && uio->uio_offset + n == new_eof)
975 if (lbase >= new_eof)
981 if (uio->uio_segflg == UIO_NOCOPY) {
983 * Issuing a write with the same data backing the
984 * buffer. Instantiate the buffer to collect the
985 * backing vm pages, then read-in any missing bits.
987 * This case is used by vop_stdputpages().
989 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
990 if ((bp->b_flags & B_CACHE) == 0) {
992 error = bread(ip->vp, lbase, lblksize, &bp);
994 } else if (trivial) {
996 * Even though we are entirely overwriting the buffer
997 * we may still have to zero it out to avoid a
998 * mmap/write visibility issue.
1000 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
1001 if ((bp->b_flags & B_CACHE) == 0)
1005 * Partial overwrite, read in any missing bits then
1006 * replace the portion being written.
1008 * (The strategy code will detect zero-fill physical
1009 * blocks for this case).
1011 error = bread(ip->vp, lbase, lblksize, &bp);
1022 * Ok, copy the data in
1024 error = uiomovebp(bp, bp->b_data + loff, n, uio);
1025 kflags |= NOTE_WRITE;
1033 * WARNING: Pageout daemon will issue UIO_NOCOPY writes
1034 * with IO_SYNC or IO_ASYNC set. These writes
1035 * must be handled as the pageout daemon expects.
1037 * NOTE! H2 relies on cluster_write() here because it
1038 * cannot preallocate disk blocks at the logical
1039 * level due to not knowing what the compression
1040 * size will be at this time.
1042 * We must use cluster_write() here and we depend
1043 * on the write-behind feature to flush buffers
1044 * appropriately. If we let the buffer daemons do
1045 * it the block allocations will be all over the
1048 if (ioflag & IO_SYNC) {
1050 } else if ((ioflag & IO_DIRECT) && endofblk) {
1052 } else if (ioflag & IO_ASYNC) {
1054 } else if (ip->vp->v_mount->mnt_flag & MNT_NOCLUSTERW) {
1058 bp->b_flags |= B_CLUSTEROK;
1059 cluster_write(bp, new_eof, lblksize, seqcount);
1061 bp->b_flags |= B_CLUSTEROK;
1068 * Cleanup. If we extended the file EOF but failed to write through
1069 * the entire write is a failure and we have to back-up.
1071 if (error && new_eof != old_eof) {
1072 hammer2_mtx_unlock(&ip->truncate_lock);
1073 hammer2_mtx_ex(&ip->lock);
1074 hammer2_mtx_ex(&ip->truncate_lock);
1075 hammer2_truncate_file(ip, old_eof);
1076 if (ip->flags & HAMMER2_INODE_MODIFIED)
1077 hammer2_inode_chain_sync(ip);
1078 hammer2_mtx_unlock(&ip->lock);
1079 } else if (modified) {
1080 hammer2_mtx_ex(&ip->lock);
1081 hammer2_inode_modify(ip);
1082 hammer2_update_time(&ip->meta.mtime);
1083 if (ip->flags & HAMMER2_INODE_MODIFIED)
1084 hammer2_inode_chain_sync(ip);
1085 hammer2_mtx_unlock(&ip->lock);
1086 hammer2_knote(ip->vp, kflags);
1088 hammer2_trans_assert_strategy(ip->pmp);
1089 hammer2_mtx_unlock(&ip->truncate_lock);
1095 * Truncate the size of a file. The inode must not be locked.
1097 * We must unconditionally set HAMMER2_INODE_RESIZED to properly
1098 * ensure that any on-media data beyond the new file EOF has been destroyed.
1100 * WARNING: nvtruncbuf() can only be safely called without the inode lock
1101 * held due to the way our write thread works. If the truncation
1102 * occurs in the middle of a buffer, nvtruncbuf() is responsible
1103 * for dirtying that buffer and zeroing out trailing bytes.
1105 * WARNING! Assumes that the kernel interlocks size changes at the
1108 * WARNING! Caller assumes responsibility for removing dead blocks
1109 * if INODE_RESIZED is set.
1113 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1115 hammer2_key_t lbase;
1118 hammer2_mtx_unlock(&ip->lock);
1120 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1121 nvtruncbuf(ip->vp, nsize,
1122 nblksize, (int)nsize & (nblksize - 1),
1125 hammer2_mtx_ex(&ip->lock);
1126 KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1127 ip->osize = ip->meta.size;
1128 ip->meta.size = nsize;
1129 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1130 hammer2_inode_modify(ip);
1134 * Extend the size of a file. The inode must not be locked.
1136 * Even though the file size is changing, we do not have to set the
1137 * INODE_RESIZED bit unless the file size crosses the EMBEDDED_BYTES
1138 * boundary. When this occurs a hammer2_inode_chain_sync() is required
1139 * to prepare the inode cluster's indirect block table, otherwise
1140 * async execution of the strategy code will implode on us.
1142 * WARNING! Assumes that the kernel interlocks size changes at the
1145 * WARNING! Caller assumes responsibility for transitioning out
1146 * of the inode DIRECTDATA mode if INODE_RESIZED is set.
1150 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1152 hammer2_key_t lbase;
1153 hammer2_key_t osize;
1157 KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1158 hammer2_inode_modify(ip);
1159 osize = ip->meta.size;
1161 ip->meta.size = nsize;
1163 if (osize <= HAMMER2_EMBEDDED_BYTES && nsize > HAMMER2_EMBEDDED_BYTES) {
1164 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1165 hammer2_inode_chain_sync(ip);
1168 hammer2_mtx_unlock(&ip->lock);
1170 oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL);
1171 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1177 hammer2_mtx_ex(&ip->lock);
1182 hammer2_vop_nresolve(struct vop_nresolve_args *ap)
1184 hammer2_xop_nresolve_t *xop;
1185 hammer2_inode_t *ip;
1186 hammer2_inode_t *dip;
1187 struct namecache *ncp;
1191 dip = VTOI(ap->a_dvp);
1192 xop = hammer2_xop_alloc(dip, 0);
1194 ncp = ap->a_nch->ncp;
1195 hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1198 * Note: In DragonFly the kernel handles '.' and '..'.
1200 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1201 hammer2_xop_start(&xop->head, hammer2_xop_nresolve);
1203 error = hammer2_xop_collect(&xop->head, 0);
1204 error = hammer2_error_to_errno(error);
1208 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1210 hammer2_inode_unlock(dip);
1213 * Acquire the related vnode
1215 * NOTE: For error processing, only ENOENT resolves the namecache
1216 * entry to NULL, otherwise we just return the error and
1217 * leave the namecache unresolved.
1219 * NOTE: multiple hammer2_inode structures can be aliased to the
1220 * same chain element, for example for hardlinks. This
1221 * use case does not 'reattach' inode associations that
1222 * might already exist, but always allocates a new one.
1224 * WARNING: inode structure is locked exclusively via inode_get
1225 * but chain was locked shared. inode_unlock()
1226 * will handle it properly.
1229 vp = hammer2_igetv(ip, &error); /* error set to UNIX error */
1232 cache_setvp(ap->a_nch, vp);
1233 } else if (error == ENOENT) {
1234 cache_setvp(ap->a_nch, NULL);
1236 hammer2_inode_unlock(ip);
1239 * The vp should not be released until after we've disposed
1240 * of our locks, because it might cause vop_inactive() to
1247 cache_setvp(ap->a_nch, NULL);
1249 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1250 KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
1251 ("resolve error %d/%p ap %p\n",
1252 error, ap->a_nch->ncp->nc_vp, ap));
1259 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
1261 hammer2_inode_t *dip;
1265 dip = VTOI(ap->a_dvp);
1266 inum = dip->meta.iparent;
1270 error = hammer2_vfs_vget(ap->a_dvp->v_mount, NULL,
1280 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
1282 hammer2_inode_t *dip;
1283 hammer2_inode_t *nip;
1284 struct namecache *ncp;
1285 const uint8_t *name;
1290 dip = VTOI(ap->a_dvp);
1291 if (dip->pmp->ronly)
1293 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1296 ncp = ap->a_nch->ncp;
1297 name = ncp->nc_name;
1298 name_len = ncp->nc_nlen;
1300 hammer2_pfs_memory_wait(dip->pmp);
1301 hammer2_trans_init(dip->pmp, 0);
1303 inum = hammer2_trans_newinum(dip->pmp);
1306 * Create the actual inode as a hidden file in the iroot, then
1307 * create the directory entry. The creation of the actual inode
1308 * sets its nlinks to 1 which is the value we desire.
1310 nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1315 error = hammer2_error_to_errno(error);
1317 error = hammer2_dirent_create(dip, name, name_len,
1318 nip->meta.inum, nip->meta.type);
1319 /* returns UNIX error code */
1323 hammer2_inode_unlink_finisher(nip, 0);
1324 hammer2_inode_unlock(nip);
1329 *ap->a_vpp = hammer2_igetv(nip, &error);
1330 hammer2_inode_unlock(nip);
1334 * Update dip's mtime
1339 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1340 hammer2_update_time(&mtime);
1341 hammer2_inode_modify(dip);
1342 dip->meta.mtime = mtime;
1343 hammer2_inode_unlock(dip);
1346 hammer2_trans_done(dip->pmp);
1349 cache_setunresolved(ap->a_nch);
1350 cache_setvp(ap->a_nch, *ap->a_vpp);
1351 hammer2_knote(ap->a_dvp, NOTE_WRITE | NOTE_LINK);
1358 hammer2_vop_open(struct vop_open_args *ap)
1360 return vop_stdopen(ap);
1364 * hammer2_vop_advlock { vp, id, op, fl, flags }
1368 hammer2_vop_advlock(struct vop_advlock_args *ap)
1370 hammer2_inode_t *ip = VTOI(ap->a_vp);
1373 size = ip->meta.size;
1374 return (lf_advlock(ap, &ip->advlock, size));
1379 hammer2_vop_close(struct vop_close_args *ap)
1381 return vop_stdclose(ap);
1385 * hammer2_vop_nlink { nch, dvp, vp, cred }
1387 * Create a hardlink from (vp) to {dvp, nch}.
1391 hammer2_vop_nlink(struct vop_nlink_args *ap)
1393 hammer2_inode_t *tdip; /* target directory to create link in */
1394 hammer2_inode_t *ip; /* inode we are hardlinking to */
1395 struct namecache *ncp;
1396 const uint8_t *name;
1400 if (ap->a_dvp->v_mount != ap->a_vp->v_mount)
1403 tdip = VTOI(ap->a_dvp);
1404 if (tdip->pmp->ronly)
1406 if (hammer2_vfs_enospace(tdip, 0, ap->a_cred) > 1)
1409 ncp = ap->a_nch->ncp;
1410 name = ncp->nc_name;
1411 name_len = ncp->nc_nlen;
1414 * ip represents the file being hardlinked. The file could be a
1415 * normal file or a hardlink target if it has already been hardlinked.
1416 * (with the new semantics, it will almost always be a hardlink
1419 * Bump nlinks and potentially also create or move the hardlink
1420 * target in the parent directory common to (ip) and (tdip). The
1421 * consolidation code can modify ip->cluster. The returned cluster
1424 ip = VTOI(ap->a_vp);
1425 KASSERT(ip->pmp, ("ip->pmp is NULL %p %p", ip, ip->pmp));
1426 hammer2_pfs_memory_wait(ip->pmp);
1427 hammer2_trans_init(ip->pmp, 0);
1430 * Target should be an indexed inode or there's no way we will ever
1431 * be able to find it!
1433 KKASSERT((ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0);
1438 * Can return NULL and error == EXDEV if the common parent
1439 * crosses a directory with the xlink flag set.
1441 hammer2_inode_lock(tdip, 0);
1442 hammer2_inode_lock(ip, 0);
1445 * Create the directory entry and bump nlinks.
1448 error = hammer2_dirent_create(tdip, name, name_len,
1449 ip->meta.inum, ip->meta.type);
1450 hammer2_inode_modify(ip);
1455 * Update dip's mtime
1459 hammer2_update_time(&mtime);
1460 hammer2_inode_modify(tdip);
1461 tdip->meta.mtime = mtime;
1463 cache_setunresolved(ap->a_nch);
1464 cache_setvp(ap->a_nch, ap->a_vp);
1466 hammer2_inode_unlock(ip);
1467 hammer2_inode_unlock(tdip);
1469 hammer2_trans_done(ip->pmp);
1470 hammer2_knote(ap->a_vp, NOTE_LINK);
1471 hammer2_knote(ap->a_dvp, NOTE_WRITE);
1477 * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1479 * The operating system has already ensured that the directory entry
1480 * does not exist and done all appropriate namespace locking.
1484 hammer2_vop_ncreate(struct vop_ncreate_args *ap)
1486 hammer2_inode_t *dip;
1487 hammer2_inode_t *nip;
1488 struct namecache *ncp;
1489 const uint8_t *name;
1494 dip = VTOI(ap->a_dvp);
1495 if (dip->pmp->ronly)
1497 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1500 ncp = ap->a_nch->ncp;
1501 name = ncp->nc_name;
1502 name_len = ncp->nc_nlen;
1503 hammer2_pfs_memory_wait(dip->pmp);
1504 hammer2_trans_init(dip->pmp, 0);
1506 inum = hammer2_trans_newinum(dip->pmp);
1509 * Create the actual inode as a hidden file in the iroot, then
1510 * create the directory entry. The creation of the actual inode
1511 * sets its nlinks to 1 which is the value we desire.
1513 nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1519 error = hammer2_dirent_create(dip, name, name_len,
1520 nip->meta.inum, nip->meta.type);
1524 hammer2_inode_unlink_finisher(nip, 0);
1525 hammer2_inode_unlock(nip);
1530 *ap->a_vpp = hammer2_igetv(nip, &error);
1531 hammer2_inode_unlock(nip);
1535 * Update dip's mtime
1540 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1541 hammer2_update_time(&mtime);
1542 hammer2_inode_modify(dip);
1543 dip->meta.mtime = mtime;
1544 hammer2_inode_unlock(dip);
1547 hammer2_trans_done(dip->pmp);
1550 cache_setunresolved(ap->a_nch);
1551 cache_setvp(ap->a_nch, *ap->a_vpp);
1552 hammer2_knote(ap->a_dvp, NOTE_WRITE);
1558 * Make a device node (typically a fifo)
1562 hammer2_vop_nmknod(struct vop_nmknod_args *ap)
1564 hammer2_inode_t *dip;
1565 hammer2_inode_t *nip;
1566 struct namecache *ncp;
1567 const uint8_t *name;
1572 dip = VTOI(ap->a_dvp);
1573 if (dip->pmp->ronly)
1575 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1578 ncp = ap->a_nch->ncp;
1579 name = ncp->nc_name;
1580 name_len = ncp->nc_nlen;
1581 hammer2_pfs_memory_wait(dip->pmp);
1582 hammer2_trans_init(dip->pmp, 0);
1585 * Create the device inode and then create the directory entry.
1587 inum = hammer2_trans_newinum(dip->pmp);
1588 nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1593 error = hammer2_dirent_create(dip, name, name_len,
1594 nip->meta.inum, nip->meta.type);
1598 hammer2_inode_unlink_finisher(nip, 0);
1599 hammer2_inode_unlock(nip);
1604 *ap->a_vpp = hammer2_igetv(nip, &error);
1605 hammer2_inode_unlock(nip);
1609 * Update dip's mtime
1614 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1615 hammer2_update_time(&mtime);
1616 hammer2_inode_modify(dip);
1617 dip->meta.mtime = mtime;
1618 hammer2_inode_unlock(dip);
1621 hammer2_trans_done(dip->pmp);
1624 cache_setunresolved(ap->a_nch);
1625 cache_setvp(ap->a_nch, *ap->a_vpp);
1626 hammer2_knote(ap->a_dvp, NOTE_WRITE);
1632 * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1636 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
1638 hammer2_inode_t *dip;
1639 hammer2_inode_t *nip;
1640 struct namecache *ncp;
1641 const uint8_t *name;
1646 dip = VTOI(ap->a_dvp);
1647 if (dip->pmp->ronly)
1649 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1652 ncp = ap->a_nch->ncp;
1653 name = ncp->nc_name;
1654 name_len = ncp->nc_nlen;
1655 hammer2_pfs_memory_wait(dip->pmp);
1656 hammer2_trans_init(dip->pmp, 0);
1658 ap->a_vap->va_type = VLNK; /* enforce type */
1661 * Create the softlink as an inode and then create the directory
1664 inum = hammer2_trans_newinum(dip->pmp);
1666 nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1671 error = hammer2_dirent_create(dip, name, name_len,
1672 nip->meta.inum, nip->meta.type);
1676 hammer2_inode_unlink_finisher(nip, 0);
1677 hammer2_inode_unlock(nip);
1681 hammer2_trans_done(dip->pmp);
1684 *ap->a_vpp = hammer2_igetv(nip, &error);
1687 * Build the softlink (~like file data) and finalize the namecache.
1694 bytes = strlen(ap->a_target);
1696 hammer2_inode_unlock(nip);
1697 bzero(&auio, sizeof(auio));
1698 bzero(&aiov, sizeof(aiov));
1699 auio.uio_iov = &aiov;
1700 auio.uio_segflg = UIO_SYSSPACE;
1701 auio.uio_rw = UIO_WRITE;
1702 auio.uio_resid = bytes;
1703 auio.uio_iovcnt = 1;
1704 auio.uio_td = curthread;
1705 aiov.iov_base = ap->a_target;
1706 aiov.iov_len = bytes;
1707 error = hammer2_write_file(nip, &auio, IO_APPEND, 0);
1708 /* XXX handle error */
1711 hammer2_inode_unlock(nip);
1715 * Update dip's mtime
1720 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1721 hammer2_update_time(&mtime);
1722 hammer2_inode_modify(dip);
1723 dip->meta.mtime = mtime;
1724 hammer2_inode_unlock(dip);
1727 hammer2_trans_done(dip->pmp);
1730 * Finalize namecache
1733 cache_setunresolved(ap->a_nch);
1734 cache_setvp(ap->a_nch, *ap->a_vpp);
1735 hammer2_knote(ap->a_dvp, NOTE_WRITE);
1741 * hammer2_vop_nremove { nch, dvp, cred }
1745 hammer2_vop_nremove(struct vop_nremove_args *ap)
1747 hammer2_xop_unlink_t *xop;
1748 hammer2_inode_t *dip;
1749 hammer2_inode_t *ip;
1750 struct namecache *ncp;
1754 dip = VTOI(ap->a_dvp);
1755 if (dip->pmp->ronly)
1758 /* allow removals, except user to also bulkfree */
1759 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
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);
1778 * The namecache entry is locked so nobody can use this namespace.
1779 * Calculate isopen to determine if this namespace has an open vp
1780 * associated with it and resolve the vp only if it does.
1782 * We try to avoid resolving the vnode if nobody has it open, but
1783 * note that the test is via this namespace only.
1785 isopen = cache_isopen(ap->a_nch);
1787 xop->dopermanent = 0;
1788 hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1791 * Collect the real inode and adjust nlinks, destroy the real
1792 * inode if nlinks transitions to 0 and it was the real inode
1793 * (else it has already been removed).
1795 error = hammer2_xop_collect(&xop->head, 0);
1796 error = hammer2_error_to_errno(error);
1797 hammer2_inode_unlock(dip);
1800 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1801 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1803 hammer2_inode_unlink_finisher(ip, isopen);
1804 hammer2_inode_unlock(ip);
1807 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1811 * Update dip's mtime
1816 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1817 hammer2_update_time(&mtime);
1818 hammer2_inode_modify(dip);
1819 dip->meta.mtime = mtime;
1820 hammer2_inode_unlock(dip);
1823 hammer2_inode_run_sideq(dip->pmp);
1824 hammer2_trans_done(dip->pmp);
1826 cache_unlink(ap->a_nch);
1827 hammer2_knote(ap->a_dvp, NOTE_WRITE);
1833 * hammer2_vop_nrmdir { nch, dvp, cred }
1837 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
1839 hammer2_xop_unlink_t *xop;
1840 hammer2_inode_t *dip;
1841 hammer2_inode_t *ip;
1842 struct namecache *ncp;
1846 dip = VTOI(ap->a_dvp);
1847 if (dip->pmp->ronly)
1850 /* allow removals, except user to also bulkfree */
1851 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1855 hammer2_pfs_memory_wait(dip->pmp);
1856 hammer2_trans_init(dip->pmp, 0);
1857 hammer2_inode_lock(dip, 0);
1859 xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1861 ncp = ap->a_nch->ncp;
1862 hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1863 isopen = cache_isopen(ap->a_nch);
1865 xop->dopermanent = 0;
1866 hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1869 * Collect the real inode and adjust nlinks, destroy the real
1870 * inode if nlinks transitions to 0 and it was the real inode
1871 * (else it has already been removed).
1873 error = hammer2_xop_collect(&xop->head, 0);
1874 error = hammer2_error_to_errno(error);
1875 hammer2_inode_unlock(dip);
1878 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1879 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1881 hammer2_inode_unlink_finisher(ip, isopen);
1882 hammer2_inode_unlock(ip);
1885 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1889 * Update dip's mtime
1894 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1895 hammer2_update_time(&mtime);
1896 hammer2_inode_modify(dip);
1897 dip->meta.mtime = mtime;
1898 hammer2_inode_unlock(dip);
1901 hammer2_inode_run_sideq(dip->pmp);
1902 hammer2_trans_done(dip->pmp);
1904 cache_unlink(ap->a_nch);
1905 hammer2_knote(ap->a_dvp, NOTE_WRITE | NOTE_LINK);
1911 * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1915 hammer2_vop_nrename(struct vop_nrename_args *ap)
1917 struct namecache *fncp;
1918 struct namecache *tncp;
1919 hammer2_inode_t *fdip; /* source directory */
1920 hammer2_inode_t *tdip; /* target directory */
1921 hammer2_inode_t *ip; /* file being renamed */
1922 hammer2_inode_t *tip; /* replaced target during rename or NULL */
1923 const uint8_t *fname;
1925 const uint8_t *tname;
1932 if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
1934 if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
1937 fdip = VTOI(ap->a_fdvp); /* source directory */
1938 tdip = VTOI(ap->a_tdvp); /* target directory */
1940 if (fdip->pmp->ronly)
1942 if (hammer2_vfs_enospace(fdip, 0, ap->a_cred) > 1)
1945 fncp = ap->a_fnch->ncp; /* entry name in source */
1946 fname = fncp->nc_name;
1947 fname_len = fncp->nc_nlen;
1949 tncp = ap->a_tnch->ncp; /* entry name in target */
1950 tname = tncp->nc_name;
1951 tname_len = tncp->nc_nlen;
1953 hammer2_pfs_memory_wait(tdip->pmp);
1954 hammer2_trans_init(tdip->pmp, 0);
1959 ip = VTOI(fncp->nc_vp);
1960 hammer2_inode_ref(ip); /* extra ref */
1963 * Lookup the target name to determine if a directory entry
1964 * is being overwritten. We only hold related inode locks
1965 * temporarily, the operating system is expected to protect
1966 * against rename races.
1968 tip = tncp->nc_vp ? VTOI(tncp->nc_vp) : NULL;
1970 hammer2_inode_ref(tip); /* extra ref */
1973 * Can return NULL and error == EXDEV if the common parent
1974 * crosses a directory with the xlink flag set.
1976 * For now try to avoid deadlocks with a simple pointer address
1977 * test. (tip) can be NULL.
1981 hammer2_inode_lock(fdip, 0);
1982 hammer2_inode_lock(tdip, 0);
1984 hammer2_inode_lock(tdip, 0);
1985 hammer2_inode_lock(fdip, 0);
1989 hammer2_inode_lock(ip, 0);
1990 hammer2_inode_lock(tip, 0);
1992 hammer2_inode_lock(tip, 0);
1993 hammer2_inode_lock(ip, 0);
1996 hammer2_inode_lock(ip, 0);
2001 * Delete the target namespace.
2003 * REMOVED - NOW FOLDED INTO XOP_NRENAME OPERATION
2006 hammer2_xop_unlink_t *xop2;
2007 hammer2_inode_t *tip;
2011 * The unlink XOP unlinks the path from the directory and
2012 * locates and returns the cluster associated with the real
2013 * inode. We have to handle nlinks here on the frontend.
2015 xop2 = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
2016 hammer2_xop_setname(&xop2->head, tname, tname_len);
2017 isopen = cache_isopen(ap->a_tnch);
2019 xop2->dopermanent = 0;
2020 hammer2_xop_start(&xop2->head, hammer2_xop_unlink);
2023 * Collect the real inode and adjust nlinks, destroy the real
2024 * inode if nlinks transitions to 0 and it was the real inode
2025 * (else it has already been removed).
2027 tnch_error = hammer2_xop_collect(&xop2->head, 0);
2028 tnch_error = hammer2_error_to_errno(tnch_error);
2029 /* hammer2_inode_unlock(tdip); */
2031 if (tnch_error == 0) {
2032 tip = hammer2_inode_get(tdip->pmp, NULL,
2033 &xop2->head.cluster, -1);
2034 hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
2036 hammer2_inode_unlink_finisher(tip, isopen);
2037 hammer2_inode_unlock(tip);
2040 hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
2042 /* hammer2_inode_lock(tdip, 0); */
2044 if (tnch_error && tnch_error != ENOENT) {
2053 * Resolve the collision space for (tdip, tname, tname_len)
2055 * tdip must be held exclusively locked to prevent races since
2056 * multiple filenames can end up in the same collision space.
2059 hammer2_xop_scanlhc_t *sxop;
2060 hammer2_tid_t lhcbase;
2062 tlhc = hammer2_dirhash(tname, tname_len);
2064 sxop = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
2066 hammer2_xop_start(&sxop->head, hammer2_xop_scanlhc);
2067 while ((error = hammer2_xop_collect(&sxop->head, 0)) == 0) {
2068 if (tlhc != sxop->head.cluster.focus->bref.key)
2072 error = hammer2_error_to_errno(error);
2073 hammer2_xop_retire(&sxop->head, HAMMER2_XOPMASK_VOP);
2076 if (error != ENOENT)
2081 if ((lhcbase ^ tlhc) & ~HAMMER2_DIRHASH_LOMASK) {
2088 * Ready to go, issue the rename to the backend. Note that meta-data
2089 * updates to the related inodes occur separately from the rename
2092 * NOTE: While it is not necessary to update ip->meta.name*, doing
2093 * so aids catastrophic recovery and debugging.
2096 hammer2_xop_nrename_t *xop4;
2098 xop4 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
2100 xop4->ip_key = ip->meta.name_key;
2101 hammer2_xop_setip2(&xop4->head, ip);
2102 hammer2_xop_setip3(&xop4->head, tdip);
2103 hammer2_xop_setname(&xop4->head, fname, fname_len);
2104 hammer2_xop_setname2(&xop4->head, tname, tname_len);
2105 hammer2_xop_start(&xop4->head, hammer2_xop_nrename);
2107 error = hammer2_xop_collect(&xop4->head, 0);
2108 error = hammer2_error_to_errno(error);
2109 hammer2_xop_retire(&xop4->head, HAMMER2_XOPMASK_VOP);
2111 if (error == ENOENT)
2115 * Update inode meta-data.
2117 * WARNING! The in-memory inode (ip) structure does not
2118 * maintain a copy of the inode's filename buffer.
2121 (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
2122 hammer2_inode_modify(ip);
2123 ip->meta.name_len = tname_len;
2124 ip->meta.name_key = tlhc;
2127 hammer2_inode_modify(ip);
2128 ip->meta.iparent = tdip->meta.inum;
2136 * If no error, the backend has replaced the target directory entry.
2137 * We must adjust nlinks on the original replace target if it exists.
2139 if (error == 0 && tip) {
2142 isopen = cache_isopen(ap->a_tnch);
2143 hammer2_inode_unlink_finisher(tip, isopen);
2147 * Update directory mtimes to represent the something changed.
2149 if (update_fdip || update_tdip) {
2152 hammer2_update_time(&mtime);
2154 hammer2_inode_modify(fdip);
2155 fdip->meta.mtime = mtime;
2158 hammer2_inode_modify(tdip);
2159 tdip->meta.mtime = mtime;
2163 hammer2_inode_unlock(tip);
2164 hammer2_inode_drop(tip);
2166 hammer2_inode_unlock(ip);
2167 hammer2_inode_unlock(tdip);
2168 hammer2_inode_unlock(fdip);
2169 hammer2_inode_drop(ip);
2170 hammer2_inode_run_sideq(fdip->pmp);
2172 hammer2_trans_done(tdip->pmp);
2175 * Issue the namecache update after unlocking all the internal
2176 * hammer structures, otherwise we might deadlock.
2178 if (error == 0 && tip) {
2179 cache_unlink(ap->a_tnch);
2180 cache_setunresolved(ap->a_tnch);
2183 cache_rename(ap->a_fnch, ap->a_tnch);
2184 hammer2_knote(ap->a_fdvp, NOTE_WRITE);
2185 hammer2_knote(ap->a_tdvp, NOTE_WRITE);
2186 hammer2_knote(fncp->nc_vp, NOTE_RENAME);
2193 * hammer2_vop_ioctl { vp, command, data, fflag, cred }
2197 hammer2_vop_ioctl(struct vop_ioctl_args *ap)
2199 hammer2_inode_t *ip;
2202 ip = VTOI(ap->a_vp);
2204 error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
2205 ap->a_fflag, ap->a_cred);
2211 hammer2_vop_mountctl(struct vop_mountctl_args *ap)
2218 case (MOUNTCTL_SET_EXPORT):
2219 mp = ap->a_head.a_ops->head.vv_mount;
2222 if (ap->a_ctllen != sizeof(struct export_args))
2225 rc = vfs_export(mp, &pmp->export,
2226 (const struct export_args *)ap->a_ctl);
2229 rc = vop_stdmountctl(ap);
2238 static void filt_hammer2detach(struct knote *kn);
2239 static int filt_hammer2read(struct knote *kn, long hint);
2240 static int filt_hammer2write(struct knote *kn, long hint);
2241 static int filt_hammer2vnode(struct knote *kn, long hint);
2243 static struct filterops hammer2read_filtops =
2244 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2245 NULL, filt_hammer2detach, filt_hammer2read };
2246 static struct filterops hammer2write_filtops =
2247 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2248 NULL, filt_hammer2detach, filt_hammer2write };
2249 static struct filterops hammer2vnode_filtops =
2250 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2251 NULL, filt_hammer2detach, filt_hammer2vnode };
2255 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap)
2257 struct vnode *vp = ap->a_vp;
2258 struct knote *kn = ap->a_kn;
2260 switch (kn->kn_filter) {
2262 kn->kn_fop = &hammer2read_filtops;
2265 kn->kn_fop = &hammer2write_filtops;
2268 kn->kn_fop = &hammer2vnode_filtops;
2271 return (EOPNOTSUPP);
2274 kn->kn_hook = (caddr_t)vp;
2276 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2282 filt_hammer2detach(struct knote *kn)
2284 struct vnode *vp = (void *)kn->kn_hook;
2286 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2290 filt_hammer2read(struct knote *kn, long hint)
2292 struct vnode *vp = (void *)kn->kn_hook;
2293 hammer2_inode_t *ip = VTOI(vp);
2296 if (hint == NOTE_REVOKE) {
2297 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2300 off = ip->meta.size - kn->kn_fp->f_offset;
2301 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2302 if (kn->kn_sfflags & NOTE_OLDAPI)
2304 return (kn->kn_data != 0);
2309 filt_hammer2write(struct knote *kn, long hint)
2311 if (hint == NOTE_REVOKE)
2312 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2318 filt_hammer2vnode(struct knote *kn, long hint)
2320 if (kn->kn_sfflags & hint)
2321 kn->kn_fflags |= hint;
2322 if (hint == NOTE_REVOKE) {
2323 kn->kn_flags |= (EV_EOF | EV_NODATA);
2326 return (kn->kn_fflags != 0);
2334 hammer2_vop_markatime(struct vop_markatime_args *ap)
2336 hammer2_inode_t *ip;
2349 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap)
2353 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2355 error = hammer2_vop_kqfilter(ap);
2362 struct vop_ops hammer2_vnode_vops = {
2363 .vop_default = vop_defaultop,
2364 .vop_fsync = hammer2_vop_fsync,
2365 .vop_getpages = vop_stdgetpages,
2366 .vop_putpages = vop_stdputpages,
2367 .vop_access = hammer2_vop_access,
2368 .vop_advlock = hammer2_vop_advlock,
2369 .vop_close = hammer2_vop_close,
2370 .vop_nlink = hammer2_vop_nlink,
2371 .vop_ncreate = hammer2_vop_ncreate,
2372 .vop_nsymlink = hammer2_vop_nsymlink,
2373 .vop_nremove = hammer2_vop_nremove,
2374 .vop_nrmdir = hammer2_vop_nrmdir,
2375 .vop_nrename = hammer2_vop_nrename,
2376 .vop_getattr = hammer2_vop_getattr,
2377 .vop_setattr = hammer2_vop_setattr,
2378 .vop_readdir = hammer2_vop_readdir,
2379 .vop_readlink = hammer2_vop_readlink,
2380 .vop_getpages = vop_stdgetpages,
2381 .vop_putpages = vop_stdputpages,
2382 .vop_read = hammer2_vop_read,
2383 .vop_write = hammer2_vop_write,
2384 .vop_open = hammer2_vop_open,
2385 .vop_inactive = hammer2_vop_inactive,
2386 .vop_reclaim = hammer2_vop_reclaim,
2387 .vop_nresolve = hammer2_vop_nresolve,
2388 .vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
2389 .vop_nmkdir = hammer2_vop_nmkdir,
2390 .vop_nmknod = hammer2_vop_nmknod,
2391 .vop_ioctl = hammer2_vop_ioctl,
2392 .vop_mountctl = hammer2_vop_mountctl,
2393 .vop_bmap = hammer2_vop_bmap,
2394 .vop_strategy = hammer2_vop_strategy,
2395 .vop_kqfilter = hammer2_vop_kqfilter
2398 struct vop_ops hammer2_spec_vops = {
2399 .vop_default = vop_defaultop,
2400 .vop_fsync = hammer2_vop_fsync,
2401 .vop_read = vop_stdnoread,
2402 .vop_write = vop_stdnowrite,
2403 .vop_access = hammer2_vop_access,
2404 .vop_close = hammer2_vop_close,
2405 .vop_markatime = hammer2_vop_markatime,
2406 .vop_getattr = hammer2_vop_getattr,
2407 .vop_inactive = hammer2_vop_inactive,
2408 .vop_reclaim = hammer2_vop_reclaim,
2409 .vop_setattr = hammer2_vop_setattr
2412 struct vop_ops hammer2_fifo_vops = {
2413 .vop_default = fifo_vnoperate,
2414 .vop_fsync = hammer2_vop_fsync,
2416 .vop_read = hammer2_vop_fiforead,
2417 .vop_write = hammer2_vop_fifowrite,
2419 .vop_access = hammer2_vop_access,
2421 .vop_close = hammer2_vop_fifoclose,
2423 .vop_markatime = hammer2_vop_markatime,
2424 .vop_getattr = hammer2_vop_getattr,
2425 .vop_inactive = hammer2_vop_inactive,
2426 .vop_reclaim = hammer2_vop_reclaim,
2427 .vop_setattr = hammer2_vop_setattr,
2428 .vop_kqfilter = hammer2_vop_fifokqfilter