2 * Copyright (c) 2011-2014 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>
61 #include "hammer2_lz4.h"
63 #include "zlib/hammer2_zlib.h"
65 #define ZFOFFSET (-2LL)
67 static int hammer2_read_file(hammer2_inode_t *ip, struct uio *uio,
69 static int hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
70 int ioflag, int seqcount);
71 static void hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize);
72 static void hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize);
74 struct objcache *cache_buffer_read;
75 struct objcache *cache_buffer_write;
78 * Callback used in read path in case that a block is compressed with LZ4.
82 hammer2_decompress_LZ4_callback(const char *data, u_int bytes, struct bio *bio)
85 char *compressed_buffer;
92 if bio->bio_caller_info2.index &&
93 bio->bio_caller_info1.uvalue32 !=
94 crc32(bp->b_data, bp->b_bufsize) --- return error
97 KKASSERT(bp->b_bufsize <= HAMMER2_PBUFSIZE);
98 compressed_size = *(const int *)data;
99 KKASSERT(compressed_size <= bytes - sizeof(int));
101 compressed_buffer = objcache_get(cache_buffer_read, M_INTWAIT);
102 result = LZ4_decompress_safe(__DECONST(char *, &data[sizeof(int)]),
107 kprintf("READ PATH: Error during decompression."
109 (intmax_t)bio->bio_offset, bytes);
110 /* make sure it isn't random garbage */
111 bzero(compressed_buffer, bp->b_bufsize);
113 KKASSERT(result <= bp->b_bufsize);
114 bcopy(compressed_buffer, bp->b_data, bp->b_bufsize);
115 if (result < bp->b_bufsize)
116 bzero(bp->b_data + result, bp->b_bufsize - result);
117 objcache_put(cache_buffer_read, compressed_buffer);
119 bp->b_flags |= B_AGE;
123 * Callback used in read path in case that a block is compressed with ZLIB.
124 * It is almost identical to LZ4 callback, so in theory they can be unified,
125 * but we didn't want to make changes in bio structure for that.
129 hammer2_decompress_ZLIB_callback(const char *data, u_int bytes, struct bio *bio)
132 char *compressed_buffer;
133 z_stream strm_decompress;
139 KKASSERT(bp->b_bufsize <= HAMMER2_PBUFSIZE);
140 strm_decompress.avail_in = 0;
141 strm_decompress.next_in = Z_NULL;
143 ret = inflateInit(&strm_decompress);
146 kprintf("HAMMER2 ZLIB: Fatal error in inflateInit.\n");
148 compressed_buffer = objcache_get(cache_buffer_read, M_INTWAIT);
149 strm_decompress.next_in = __DECONST(char *, data);
151 /* XXX supply proper size, subset of device bp */
152 strm_decompress.avail_in = bytes;
153 strm_decompress.next_out = compressed_buffer;
154 strm_decompress.avail_out = bp->b_bufsize;
156 ret = inflate(&strm_decompress, Z_FINISH);
157 if (ret != Z_STREAM_END) {
158 kprintf("HAMMER2 ZLIB: Fatar error during decompression.\n");
159 bzero(compressed_buffer, bp->b_bufsize);
161 bcopy(compressed_buffer, bp->b_data, bp->b_bufsize);
162 result = bp->b_bufsize - strm_decompress.avail_out;
163 if (result < bp->b_bufsize)
164 bzero(bp->b_data + result, strm_decompress.avail_out);
165 objcache_put(cache_buffer_read, compressed_buffer);
166 ret = inflateEnd(&strm_decompress);
169 bp->b_flags |= B_AGE;
174 hammer2_knote(struct vnode *vp, int flags)
177 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
181 * Last reference to a vnode is going away but it is still cached.
185 hammer2_vop_inactive(struct vop_inactive_args *ap)
188 hammer2_cluster_t *cluster;
205 * Detect updates to the embedded data which may be synchronized by
206 * the strategy code. Simply mark the inode modified so it gets
207 * picked up by our normal flush.
209 cluster = hammer2_inode_lock(ip, HAMMER2_RESOLVE_NEVER |
210 HAMMER2_RESOLVE_RDONLY);
214 * Check for deleted inodes and recycle immediately.
216 * WARNING: nvtruncbuf() can only be safely called without the inode
217 * lock held due to the way our write thread works.
219 if (hammer2_cluster_isunlinked(cluster)) {
223 nblksize = hammer2_calc_logical(ip, 0, &lbase, NULL);
224 hammer2_inode_unlock(ip, cluster);
225 nvtruncbuf(vp, 0, nblksize, 0, 0);
228 hammer2_inode_unlock(ip, cluster);
235 * Reclaim a vnode so that it can be reused; after the inode is
236 * disassociated, the filesystem must manage it alone.
240 hammer2_vop_reclaim(struct vop_reclaim_args *ap)
242 hammer2_cluster_t *cluster;
256 * Inode must be locked for reclaim.
259 cluster = hammer2_inode_lock(ip, HAMMER2_RESOLVE_NEVER |
260 HAMMER2_RESOLVE_RDONLY);
263 * The final close of a deleted file or directory marks it for
264 * destruction. The DELETED flag allows the flusher to shortcut
265 * any modified blocks still unflushed (that is, just ignore them).
267 * HAMMER2 usually does not try to optimize the freemap by returning
268 * deleted blocks to it as it does not usually know how many snapshots
269 * might be referencing portions of the file/dir.
275 * NOTE! We do not attempt to flush chains here, flushing is
276 * really fragile and could also deadlock.
281 * A reclaim can occur at any time so we cannot safely start a
282 * transaction to handle reclamation of unlinked files. Instead,
283 * the ip is left with a reference and placed on a linked list and
286 if (hammer2_cluster_isunlinked(cluster)) {
287 hammer2_inode_unlink_t *ipul;
289 ipul = kmalloc(sizeof(*ipul), pmp->minode, M_WAITOK | M_ZERO);
292 hammer2_spin_ex(&pmp->list_spin);
293 TAILQ_INSERT_TAIL(&pmp->unlinkq, ipul, entry);
294 hammer2_spin_unex(&pmp->list_spin);
295 hammer2_inode_unlock(ip, cluster); /* unlock */
296 /* retain ref from vp for ipul */
298 hammer2_inode_unlock(ip, cluster); /* unlock */
299 hammer2_inode_drop(ip); /* vp ref */
301 /* cluster no longer referenced */
302 /* cluster = NULL; not needed */
305 * XXX handle background sync when ip dirty, kernel will no longer
306 * notify us regarding this inode because there is no longer a
307 * vnode attached to it.
316 hammer2_vop_fsync(struct vop_fsync_args *ap)
319 hammer2_trans_t trans;
320 hammer2_cluster_t *cluster;
328 /* XXX can't do this yet */
329 hammer2_trans_init(&trans, ip->pmp, HAMMER2_TRANS_ISFLUSH);
330 vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
332 hammer2_trans_init(&trans, ip->pmp, 0);
333 vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
336 * Calling chain_flush here creates a lot of duplicative
337 * COW operations due to non-optimal vnode ordering.
339 * Only do it for an actual fsync() syscall. The other forms
340 * which call this function will eventually call chain_flush
341 * on the volume root as a catch-all, which is far more optimal.
343 cluster = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS);
344 atomic_clear_int(&ip->flags, HAMMER2_INODE_MODIFIED);
346 if (ip->flags & (HAMMER2_INODE_RESIZED|HAMMER2_INODE_MTIME))
347 hammer2_inode_fsync(&trans, ip, cluster);
351 * XXX creates discontinuity w/modify_tid
353 if (ap->a_flags & VOP_FSYNC_SYSCALL) {
354 hammer2_flush(&trans, cluster);
357 hammer2_inode_unlock(ip, cluster);
358 hammer2_trans_done(&trans);
366 hammer2_vop_access(struct vop_access_args *ap)
368 hammer2_inode_t *ip = VTOI(ap->a_vp);
369 const hammer2_inode_data_t *ripdata;
370 hammer2_cluster_t *cluster;
376 cluster = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS |
377 HAMMER2_RESOLVE_SHARED);
378 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
379 uid = hammer2_to_unix_xid(&ripdata->uid);
380 gid = hammer2_to_unix_xid(&ripdata->gid);
381 error = vop_helper_access(ap, uid, gid, ripdata->mode, ripdata->uflags);
382 hammer2_inode_unlock(ip, cluster);
390 hammer2_vop_getattr(struct vop_getattr_args *ap)
392 const hammer2_inode_data_t *ripdata;
393 hammer2_cluster_t *cluster;
406 cluster = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS |
407 HAMMER2_RESOLVE_SHARED);
408 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
409 KKASSERT(hammer2_cluster_type(cluster) == HAMMER2_BREF_TYPE_INODE);
411 vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0];
412 vap->va_fileid = ripdata->inum;
413 vap->va_mode = ripdata->mode;
414 vap->va_nlink = ripdata->nlinks;
415 vap->va_uid = hammer2_to_unix_xid(&ripdata->uid);
416 vap->va_gid = hammer2_to_unix_xid(&ripdata->gid);
419 vap->va_size = ip->size; /* protected by shared lock */
420 vap->va_blocksize = HAMMER2_PBUFSIZE;
421 vap->va_flags = ripdata->uflags;
422 hammer2_time_to_timespec(ripdata->ctime, &vap->va_ctime);
423 hammer2_time_to_timespec(ripdata->mtime, &vap->va_mtime);
424 hammer2_time_to_timespec(ripdata->mtime, &vap->va_atime);
426 vap->va_bytes = vap->va_size; /* XXX */
427 vap->va_type = hammer2_get_vtype(ripdata);
429 vap->va_uid_uuid = ripdata->uid;
430 vap->va_gid_uuid = ripdata->gid;
431 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
434 hammer2_inode_unlock(ip, cluster);
442 hammer2_vop_setattr(struct vop_setattr_args *ap)
444 const hammer2_inode_data_t *ripdata;
445 hammer2_inode_data_t *wipdata;
447 hammer2_cluster_t *cluster;
448 hammer2_trans_t trans;
460 hammer2_update_time(&ctime);
464 if (ip->pmp->ronly) {
469 hammer2_pfs_memory_wait(ip->pmp);
470 hammer2_trans_init(&trans, ip->pmp, 0);
471 cluster = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS);
472 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
475 if (vap->va_flags != VNOVAL) {
478 flags = ripdata->uflags;
479 error = vop_helper_setattr_flags(&flags, vap->va_flags,
480 hammer2_to_unix_xid(&ripdata->uid),
483 if (ripdata->uflags != flags) {
484 wipdata = hammer2_cluster_modify_ip(&trans, ip,
486 wipdata->uflags = flags;
487 wipdata->ctime = ctime;
488 kflags |= NOTE_ATTRIB;
492 if (ripdata->uflags & (IMMUTABLE | APPEND)) {
499 if (ripdata->uflags & (IMMUTABLE | APPEND)) {
503 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
504 mode_t cur_mode = ripdata->mode;
505 uid_t cur_uid = hammer2_to_unix_xid(&ripdata->uid);
506 gid_t cur_gid = hammer2_to_unix_xid(&ripdata->gid);
510 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
512 &cur_uid, &cur_gid, &cur_mode);
514 hammer2_guid_to_uuid(&uuid_uid, cur_uid);
515 hammer2_guid_to_uuid(&uuid_gid, cur_gid);
516 if (bcmp(&uuid_uid, &ripdata->uid, sizeof(uuid_uid)) ||
517 bcmp(&uuid_gid, &ripdata->gid, sizeof(uuid_gid)) ||
518 ripdata->mode != cur_mode
520 wipdata = hammer2_cluster_modify_ip(&trans, ip,
522 wipdata->uid = uuid_uid;
523 wipdata->gid = uuid_gid;
524 wipdata->mode = cur_mode;
525 wipdata->ctime = ctime;
529 kflags |= NOTE_ATTRIB;
536 if (vap->va_size != VNOVAL && ip->size != vap->va_size) {
539 if (vap->va_size == ip->size)
541 hammer2_inode_unlock(ip, cluster);
542 if (vap->va_size < ip->size) {
543 hammer2_truncate_file(ip, vap->va_size);
545 hammer2_extend_file(ip, vap->va_size);
547 cluster = hammer2_inode_lock(ip,
548 HAMMER2_RESOLVE_ALWAYS);
550 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
559 /* atime not supported */
560 if (vap->va_atime.tv_sec != VNOVAL) {
561 wipdata = hammer2_cluster_modify_ip(&trans, ip, cluster, 0);
562 wipdata->atime = hammer2_timespec_to_time(&vap->va_atime);
563 kflags |= NOTE_ATTRIB;
568 if (vap->va_mtime.tv_sec != VNOVAL) {
569 wipdata = hammer2_cluster_modify_ip(&trans, ip, cluster, 0);
570 wipdata->mtime = hammer2_timespec_to_time(&vap->va_mtime);
571 kflags |= NOTE_ATTRIB;
576 if (vap->va_mode != (mode_t)VNOVAL) {
577 mode_t cur_mode = ripdata->mode;
578 uid_t cur_uid = hammer2_to_unix_xid(&ripdata->uid);
579 gid_t cur_gid = hammer2_to_unix_xid(&ripdata->gid);
581 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
582 cur_uid, cur_gid, &cur_mode);
583 if (error == 0 && ripdata->mode != cur_mode) {
584 wipdata = hammer2_cluster_modify_ip(&trans, ip,
586 wipdata->mode = cur_mode;
587 wipdata->ctime = ctime;
588 kflags |= NOTE_ATTRIB;
595 * If a truncation occurred we must call inode_fsync() now in order
596 * to trim the related data chains, otherwise a later expansion can
600 hammer2_cluster_modsync(cluster);
603 hammer2_inode_fsync(&trans, ip, cluster);
606 * Cleanup. If domtime is set an additional inode modification
607 * must be flagged. All other modifications will have already
608 * set INODE_MODIFIED and called vsetisdirty().
612 atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED |
613 HAMMER2_INODE_MTIME);
617 hammer2_cluster_modsync(cluster);
618 hammer2_inode_unlock(ip, cluster);
619 hammer2_trans_done(&trans);
620 hammer2_knote(ip->vp, kflags);
628 hammer2_vop_readdir(struct vop_readdir_args *ap)
630 const hammer2_inode_data_t *ripdata;
632 hammer2_inode_t *xip;
633 hammer2_cluster_t *cparent;
634 hammer2_cluster_t *cluster;
635 hammer2_cluster_t *xcluster;
636 hammer2_blockref_t bref;
638 hammer2_key_t key_next;
652 saveoff = uio->uio_offset;
655 * Setup cookies directory entry cookies if requested
657 if (ap->a_ncookies) {
658 ncookies = uio->uio_resid / 16 + 1;
661 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
668 cparent = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS |
669 HAMMER2_RESOLVE_SHARED);
671 ripdata = &hammer2_cluster_rdata(cparent)->ipdata;
674 * Handle artificial entries. To ensure that only positive 64 bit
675 * quantities are returned to userland we always strip off bit 63.
676 * The hash code is designed such that codes 0x0000-0x7FFF are not
677 * used, allowing us to use these codes for articial entries.
679 * Entry 0 is used for '.' and entry 1 is used for '..'. Do not
680 * allow '..' to cross the mount point into (e.g.) the super-root.
683 cluster = (void *)(intptr_t)-1; /* non-NULL for early goto done case */
686 inum = ripdata->inum & HAMMER2_DIRHASH_USERMSK;
687 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
691 cookies[cookie_index] = saveoff;
694 if (cookie_index == ncookies)
700 * Be careful with lockorder when accessing ".."
702 * (ip is the current dir. xip is the parent dir).
704 inum = ripdata->inum & HAMMER2_DIRHASH_USERMSK;
705 while (ip->pip != NULL && ip != ip->pmp->iroot) {
707 hammer2_inode_ref(xip);
708 hammer2_inode_unlock(ip, cparent);
709 xcluster = hammer2_inode_lock(xip,
710 HAMMER2_RESOLVE_ALWAYS |
711 HAMMER2_RESOLVE_SHARED);
713 cparent = hammer2_inode_lock(ip,
714 HAMMER2_RESOLVE_ALWAYS |
715 HAMMER2_RESOLVE_SHARED);
716 hammer2_inode_drop(xip);
717 ripdata = &hammer2_cluster_rdata(cparent)->ipdata;
718 if (xip == ip->pip) {
719 inum = hammer2_cluster_rdata(xcluster)->
720 ipdata.inum & HAMMER2_DIRHASH_USERMSK;
721 hammer2_inode_unlock(xip, xcluster);
724 hammer2_inode_unlock(xip, xcluster);
726 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
730 cookies[cookie_index] = saveoff;
733 if (cookie_index == ncookies)
737 lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
738 if (hammer2_debug & 0x0020)
739 kprintf("readdir: lkey %016jx\n", lkey);
742 * parent is the inode cluster, already locked for us. Don't
743 * double lock shared locks as this will screw up upgrades.
748 cluster = hammer2_cluster_lookup(cparent, &key_next, lkey, lkey,
749 HAMMER2_LOOKUP_SHARED);
750 if (cluster == NULL) {
751 cluster = hammer2_cluster_lookup(cparent, &key_next,
752 lkey, (hammer2_key_t)-1,
753 HAMMER2_LOOKUP_SHARED);
756 hammer2_cluster_bref(cluster, &bref);
758 if (hammer2_debug & 0x0020)
759 kprintf("readdir: p=%p chain=%p %016jx (next %016jx)\n",
760 cparent->focus, cluster->focus,
763 if (bref.type == HAMMER2_BREF_TYPE_INODE) {
764 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
765 dtype = hammer2_get_dtype(ripdata);
766 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
767 r = vop_write_dirent(&error, uio,
769 HAMMER2_DIRHASH_USERMSK,
776 cookies[cookie_index] = saveoff;
779 /* XXX chain error */
780 kprintf("bad chain type readdir %d\n", bref.type);
784 * Keys may not be returned in order so once we have a
785 * placemarker (cluster) the scan must allow the full range
786 * or some entries will be missed.
788 cluster = hammer2_cluster_next(cparent, cluster, &key_next,
789 key_next, (hammer2_key_t)-1,
790 HAMMER2_LOOKUP_SHARED);
792 hammer2_cluster_bref(cluster, &bref);
793 saveoff = (bref.key & HAMMER2_DIRHASH_USERMSK) + 1;
795 saveoff = (hammer2_key_t)-1;
797 if (cookie_index == ncookies)
801 hammer2_cluster_unlock(cluster);
803 hammer2_inode_unlock(ip, cparent);
805 *ap->a_eofflag = (cluster == NULL);
806 if (hammer2_debug & 0x0020)
807 kprintf("readdir: done at %016jx\n", saveoff);
808 uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
809 if (error && cookie_index == 0) {
811 kfree(cookies, M_TEMP);
813 *ap->a_cookies = NULL;
817 *ap->a_ncookies = cookie_index;
818 *ap->a_cookies = cookies;
826 * hammer2_vop_readlink { vp, uio, cred }
830 hammer2_vop_readlink(struct vop_readlink_args *ap)
837 if (vp->v_type != VLNK)
841 error = hammer2_read_file(ip, ap->a_uio, 0);
847 hammer2_vop_read(struct vop_read_args *ap)
857 * Read operations supported on this vnode?
860 if (vp->v_type != VREG)
870 seqcount = ap->a_ioflag >> 16;
871 bigread = (uio->uio_resid > 100 * 1024 * 1024);
873 error = hammer2_read_file(ip, uio, seqcount);
879 hammer2_vop_write(struct vop_write_args *ap)
882 hammer2_trans_t trans;
891 * Read operations supported on this vnode?
894 if (vp->v_type != VREG)
903 if (ip->pmp->ronly) {
907 seqcount = ap->a_ioflag >> 16;
908 bigwrite = (uio->uio_resid > 100 * 1024 * 1024);
911 * Check resource limit
913 if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc &&
914 uio->uio_offset + uio->uio_resid >
915 td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
916 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
920 bigwrite = (uio->uio_resid > 100 * 1024 * 1024);
923 * The transaction interlocks against flushes initiations
924 * (note: but will run concurrently with the actual flush).
926 hammer2_trans_init(&trans, ip->pmp, 0);
927 error = hammer2_write_file(ip, uio, ap->a_ioflag, seqcount);
928 hammer2_trans_done(&trans);
934 * Perform read operations on a file or symlink given an UNLOCKED
937 * The passed ip is not locked.
941 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
952 * WARNING! Assumes that the kernel interlocks size changes at the
955 hammer2_mtx_sh(&ip->lock);
957 hammer2_mtx_unlock(&ip->lock);
959 while (uio->uio_resid > 0 && uio->uio_offset < size) {
966 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
969 error = cluster_read(ip->vp, leof, lbase, lblksize,
970 uio->uio_resid, seqcount * BKVASIZE,
975 loff = (int)(uio->uio_offset - lbase);
977 if (n > uio->uio_resid)
979 if (n > size - uio->uio_offset)
980 n = (int)(size - uio->uio_offset);
981 bp->b_flags |= B_AGE;
982 uiomove((char *)bp->b_data + loff, n, uio);
989 * Write to the file represented by the inode via the logical buffer cache.
990 * The inode may represent a regular file or a symlink.
992 * The inode must not be locked.
996 hammer2_write_file(hammer2_inode_t *ip,
997 struct uio *uio, int ioflag, int seqcount)
999 hammer2_key_t old_eof;
1000 hammer2_key_t new_eof;
1009 * WARNING! Assumes that the kernel interlocks size changes at the
1012 hammer2_mtx_ex(&ip->lock);
1013 if (ioflag & IO_APPEND)
1014 uio->uio_offset = ip->size;
1016 hammer2_mtx_unlock(&ip->lock);
1019 * Extend the file if necessary. If the write fails at some point
1020 * we will truncate it back down to cover as much as we were able
1023 * Doing this now makes it easier to calculate buffer sizes in
1030 if (uio->uio_offset + uio->uio_resid > old_eof) {
1031 new_eof = uio->uio_offset + uio->uio_resid;
1033 hammer2_extend_file(ip, new_eof);
1034 kflags |= NOTE_EXTEND;
1042 while (uio->uio_resid > 0) {
1043 hammer2_key_t lbase;
1051 * Don't allow the buffer build to blow out the buffer
1054 if ((ioflag & IO_RECURSE) == 0)
1055 bwillwrite(HAMMER2_PBUFSIZE);
1058 * This nominally tells us how much we can cluster and
1059 * what the logical buffer size needs to be. Currently
1060 * we don't try to cluster the write and just handle one
1063 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
1065 loff = (int)(uio->uio_offset - lbase);
1067 KKASSERT(lblksize <= 65536);
1070 * Calculate bytes to copy this transfer and whether the
1071 * copy completely covers the buffer or not.
1074 n = lblksize - loff;
1075 if (n > uio->uio_resid) {
1077 if (loff == lbase && uio->uio_offset + n == new_eof)
1089 if (uio->uio_segflg == UIO_NOCOPY) {
1091 * Issuing a write with the same data backing the
1092 * buffer. Instantiate the buffer to collect the
1093 * backing vm pages, then read-in any missing bits.
1095 * This case is used by vop_stdputpages().
1097 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
1098 if ((bp->b_flags & B_CACHE) == 0) {
1100 error = bread(ip->vp, lbase, lblksize, &bp);
1102 } else if (trivial) {
1104 * Even though we are entirely overwriting the buffer
1105 * we may still have to zero it out to avoid a
1106 * mmap/write visibility issue.
1108 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
1109 if ((bp->b_flags & B_CACHE) == 0)
1113 * Partial overwrite, read in any missing bits then
1114 * replace the portion being written.
1116 * (The strategy code will detect zero-fill physical
1117 * blocks for this case).
1119 error = bread(ip->vp, lbase, lblksize, &bp);
1130 * Ok, copy the data in
1132 error = uiomove(bp->b_data + loff, n, uio);
1133 kflags |= NOTE_WRITE;
1141 * WARNING: Pageout daemon will issue UIO_NOCOPY writes
1142 * with IO_SYNC or IO_ASYNC set. These writes
1143 * must be handled as the pageout daemon expects.
1145 if (ioflag & IO_SYNC) {
1147 } else if ((ioflag & IO_DIRECT) && endofblk) {
1149 } else if (ioflag & IO_ASYNC) {
1157 * Cleanup. If we extended the file EOF but failed to write through
1158 * the entire write is a failure and we have to back-up.
1160 if (error && new_eof != old_eof) {
1161 hammer2_truncate_file(ip, old_eof);
1162 } else if (modified) {
1163 hammer2_mtx_ex(&ip->lock);
1164 hammer2_update_time(&ip->mtime);
1165 atomic_set_int(&ip->flags, HAMMER2_INODE_MTIME);
1166 hammer2_mtx_unlock(&ip->lock);
1168 atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED);
1169 hammer2_knote(ip->vp, kflags);
1170 vsetisdirty(ip->vp);
1176 * Truncate the size of a file. The inode must not be locked.
1178 * NOTE: Caller handles setting HAMMER2_INODE_MODIFIED
1180 * WARNING: nvtruncbuf() can only be safely called without the inode lock
1181 * held due to the way our write thread works.
1183 * WARNING! Assumes that the kernel interlocks size changes at the
1188 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1190 hammer2_key_t lbase;
1195 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1196 nvtruncbuf(ip->vp, nsize,
1197 nblksize, (int)nsize & (nblksize - 1),
1200 hammer2_mtx_ex(&ip->lock);
1202 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1203 hammer2_mtx_unlock(&ip->lock);
1208 * Extend the size of a file. The inode must not be locked.
1210 * WARNING! Assumes that the kernel interlocks size changes at the
1213 * NOTE: Caller handles setting HAMMER2_INODE_MODIFIED
1217 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1219 hammer2_key_t lbase;
1220 hammer2_key_t osize;
1225 hammer2_mtx_ex(&ip->lock);
1228 hammer2_mtx_unlock(&ip->lock);
1231 oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL);
1232 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1238 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1244 hammer2_vop_nresolve(struct vop_nresolve_args *ap)
1246 hammer2_inode_t *ip;
1247 hammer2_inode_t *dip;
1248 hammer2_cluster_t *cparent;
1249 hammer2_cluster_t *cluster;
1250 const hammer2_inode_data_t *ripdata;
1251 hammer2_key_t key_next;
1253 struct namecache *ncp;
1254 const uint8_t *name;
1260 dip = VTOI(ap->a_dvp);
1261 ncp = ap->a_nch->ncp;
1262 name = ncp->nc_name;
1263 name_len = ncp->nc_nlen;
1264 lhc = hammer2_dirhash(name, name_len);
1267 * Note: In DragonFly the kernel handles '.' and '..'.
1269 cparent = hammer2_inode_lock(dip, HAMMER2_RESOLVE_ALWAYS |
1270 HAMMER2_RESOLVE_SHARED);
1272 cluster = hammer2_cluster_lookup(cparent, &key_next,
1273 lhc, lhc + HAMMER2_DIRHASH_LOMASK,
1274 HAMMER2_LOOKUP_SHARED);
1276 if (hammer2_cluster_type(cluster) == HAMMER2_BREF_TYPE_INODE) {
1277 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
1278 if (ripdata->name_len == name_len &&
1279 bcmp(ripdata->filename, name, name_len) == 0) {
1283 cluster = hammer2_cluster_next(cparent, cluster, &key_next,
1285 lhc + HAMMER2_DIRHASH_LOMASK,
1286 HAMMER2_LOOKUP_SHARED);
1288 hammer2_inode_unlock(dip, cparent);
1291 * Resolve hardlink entries before acquiring the inode.
1294 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
1295 if (ripdata->type == HAMMER2_OBJTYPE_HARDLINK) {
1296 hammer2_tid_t inum = ripdata->inum;
1297 error = hammer2_hardlink_find(dip, NULL, &cluster);
1299 kprintf("hammer2: unable to find hardlink "
1300 "0x%016jx\n", inum);
1309 * nresolve needs to resolve hardlinks, the original cluster is not
1313 ip = hammer2_inode_get(dip->pmp, dip, cluster);
1314 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
1315 if (ripdata->type == HAMMER2_OBJTYPE_HARDLINK) {
1316 kprintf("nresolve: fixup hardlink\n");
1317 hammer2_inode_ref(ip);
1318 hammer2_inode_unlock(ip, NULL);
1319 hammer2_cluster_unlock(cluster);
1320 cluster = hammer2_inode_lock(ip,
1321 HAMMER2_RESOLVE_ALWAYS);
1322 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
1323 hammer2_inode_drop(ip);
1324 kprintf("nresolve: fixup to type %02x\n",
1333 * Deconsolidate any hardlink whos nlinks == 1. Ignore errors.
1334 * If an error occurs chain and ip are left alone.
1336 * XXX upgrade shared lock?
1338 if (ochain && chain &&
1339 chain->data->ipdata.nlinks == 1 && !dip->pmp->ronly) {
1340 kprintf("hammer2: need to unconsolidate hardlink for %s\n",
1341 chain->data->ipdata.filename);
1342 /* XXX retain shared lock on dip? (currently not held) */
1343 hammer2_trans_init(&trans, dip->pmp, 0);
1344 hammer2_hardlink_deconsolidate(&trans, dip, &chain, &ochain);
1345 hammer2_trans_done(&trans);
1350 * Acquire the related vnode
1352 * NOTE: For error processing, only ENOENT resolves the namecache
1353 * entry to NULL, otherwise we just return the error and
1354 * leave the namecache unresolved.
1356 * NOTE: multiple hammer2_inode structures can be aliased to the
1357 * same chain element, for example for hardlinks. This
1358 * use case does not 'reattach' inode associations that
1359 * might already exist, but always allocates a new one.
1361 * WARNING: inode structure is locked exclusively via inode_get
1362 * but chain was locked shared. inode_unlock()
1363 * will handle it properly.
1366 vp = hammer2_igetv(ip, cluster, &error);
1369 cache_setvp(ap->a_nch, vp);
1370 } else if (error == ENOENT) {
1371 cache_setvp(ap->a_nch, NULL);
1373 hammer2_inode_unlock(ip, cluster);
1376 * The vp should not be released until after we've disposed
1377 * of our locks, because it might cause vop_inactive() to
1384 cache_setvp(ap->a_nch, NULL);
1386 KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
1387 ("resolve error %d/%p ap %p\n",
1388 error, ap->a_nch->ncp->nc_vp, ap));
1395 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
1397 hammer2_inode_t *dip;
1398 hammer2_inode_t *ip;
1399 hammer2_cluster_t *cparent;
1403 dip = VTOI(ap->a_dvp);
1405 if ((ip = dip->pip) == NULL) {
1410 cparent = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS);
1411 *ap->a_vpp = hammer2_igetv(ip, cparent, &error);
1412 hammer2_inode_unlock(ip, cparent);
1420 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
1422 hammer2_inode_t *dip;
1423 hammer2_inode_t *nip;
1424 hammer2_trans_t trans;
1425 hammer2_cluster_t *cluster;
1426 struct namecache *ncp;
1427 const uint8_t *name;
1432 dip = VTOI(ap->a_dvp);
1433 if (dip->pmp->ronly) {
1438 ncp = ap->a_nch->ncp;
1439 name = ncp->nc_name;
1440 name_len = ncp->nc_nlen;
1443 hammer2_pfs_memory_wait(dip->pmp);
1444 hammer2_trans_init(&trans, dip->pmp, HAMMER2_TRANS_NEWINODE);
1445 nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1447 &cluster, 0, &error);
1449 KKASSERT(nip == NULL);
1452 *ap->a_vpp = hammer2_igetv(nip, cluster, &error);
1453 hammer2_inode_unlock(nip, cluster);
1455 hammer2_trans_done(&trans);
1458 cache_setunresolved(ap->a_nch);
1459 cache_setvp(ap->a_nch, *ap->a_vpp);
1466 * Return the largest contiguous physical disk range for the logical
1467 * request, in bytes.
1469 * (struct vnode *vp, off_t loffset, off_t *doffsetp, int *runp, int *runb)
1471 * Basically disabled, the logical buffer write thread has to deal with
1472 * buffers one-at-a-time.
1476 hammer2_vop_bmap(struct vop_bmap_args *ap)
1478 *ap->a_doffsetp = NOOFFSET;
1483 return (EOPNOTSUPP);
1488 hammer2_vop_open(struct vop_open_args *ap)
1490 return vop_stdopen(ap);
1494 * hammer2_vop_advlock { vp, id, op, fl, flags }
1498 hammer2_vop_advlock(struct vop_advlock_args *ap)
1500 hammer2_inode_t *ip = VTOI(ap->a_vp);
1501 const hammer2_inode_data_t *ripdata;
1502 hammer2_cluster_t *cparent;
1505 cparent = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS |
1506 HAMMER2_RESOLVE_SHARED);
1507 ripdata = &hammer2_cluster_rdata(cparent)->ipdata;
1508 size = ripdata->size;
1509 hammer2_inode_unlock(ip, cparent);
1510 return (lf_advlock(ap, &ip->advlock, size));
1516 hammer2_vop_close(struct vop_close_args *ap)
1518 return vop_stdclose(ap);
1522 * hammer2_vop_nlink { nch, dvp, vp, cred }
1524 * Create a hardlink from (vp) to {dvp, nch}.
1528 hammer2_vop_nlink(struct vop_nlink_args *ap)
1530 hammer2_inode_t *fdip; /* target directory to create link in */
1531 hammer2_inode_t *tdip; /* target directory to create link in */
1532 hammer2_inode_t *cdip; /* common parent directory */
1533 hammer2_inode_t *ip; /* inode we are hardlinking to */
1534 hammer2_cluster_t *cluster;
1535 hammer2_cluster_t *fdcluster;
1536 hammer2_cluster_t *tdcluster;
1537 hammer2_cluster_t *cdcluster;
1538 hammer2_trans_t trans;
1539 struct namecache *ncp;
1540 const uint8_t *name;
1545 tdip = VTOI(ap->a_dvp);
1546 if (tdip->pmp->ronly) {
1551 ncp = ap->a_nch->ncp;
1552 name = ncp->nc_name;
1553 name_len = ncp->nc_nlen;
1556 * ip represents the file being hardlinked. The file could be a
1557 * normal file or a hardlink target if it has already been hardlinked.
1558 * If ip is a hardlinked target then ip->pip represents the location
1559 * of the hardlinked target, NOT the location of the hardlink pointer.
1561 * Bump nlinks and potentially also create or move the hardlink
1562 * target in the parent directory common to (ip) and (tdip). The
1563 * consolidation code can modify ip->cluster and ip->pip. The
1564 * returned cluster is locked.
1566 ip = VTOI(ap->a_vp);
1567 hammer2_pfs_memory_wait(ip->pmp);
1568 hammer2_trans_init(&trans, ip->pmp, HAMMER2_TRANS_NEWINODE);
1571 * The common parent directory must be locked first to avoid deadlocks.
1572 * Also note that fdip and/or tdip might match cdip.
1575 cdip = hammer2_inode_common_parent(fdip, tdip);
1576 cdcluster = hammer2_inode_lock(cdip, HAMMER2_RESOLVE_ALWAYS);
1577 fdcluster = hammer2_inode_lock(fdip, HAMMER2_RESOLVE_ALWAYS);
1578 tdcluster = hammer2_inode_lock(tdip, HAMMER2_RESOLVE_ALWAYS);
1579 cluster = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS);
1580 error = hammer2_hardlink_consolidate(&trans, ip, &cluster,
1581 cdip, cdcluster, 1);
1586 * Create a directory entry connected to the specified cluster.
1588 * WARNING! chain can get moved by the connect (indirectly due to
1589 * potential indirect block creation).
1591 error = hammer2_inode_connect(&trans, &cluster, 1,
1595 cache_setunresolved(ap->a_nch);
1596 cache_setvp(ap->a_nch, ap->a_vp);
1599 hammer2_inode_unlock(ip, cluster);
1600 hammer2_inode_unlock(tdip, tdcluster);
1601 hammer2_inode_unlock(fdip, fdcluster);
1602 hammer2_inode_unlock(cdip, cdcluster);
1603 hammer2_inode_drop(cdip);
1604 hammer2_trans_done(&trans);
1611 * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1613 * The operating system has already ensured that the directory entry
1614 * does not exist and done all appropriate namespace locking.
1618 hammer2_vop_ncreate(struct vop_ncreate_args *ap)
1620 hammer2_inode_t *dip;
1621 hammer2_inode_t *nip;
1622 hammer2_trans_t trans;
1623 hammer2_cluster_t *ncluster;
1624 struct namecache *ncp;
1625 const uint8_t *name;
1630 dip = VTOI(ap->a_dvp);
1631 if (dip->pmp->ronly) {
1636 ncp = ap->a_nch->ncp;
1637 name = ncp->nc_name;
1638 name_len = ncp->nc_nlen;
1639 hammer2_pfs_memory_wait(dip->pmp);
1640 hammer2_trans_init(&trans, dip->pmp, HAMMER2_TRANS_NEWINODE);
1643 nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1645 &ncluster, 0, &error);
1647 KKASSERT(nip == NULL);
1650 *ap->a_vpp = hammer2_igetv(nip, ncluster, &error);
1651 hammer2_inode_unlock(nip, ncluster);
1653 hammer2_trans_done(&trans);
1656 cache_setunresolved(ap->a_nch);
1657 cache_setvp(ap->a_nch, *ap->a_vpp);
1664 * Make a device node (typically a fifo)
1668 hammer2_vop_nmknod(struct vop_nmknod_args *ap)
1670 hammer2_inode_t *dip;
1671 hammer2_inode_t *nip;
1672 hammer2_trans_t trans;
1673 hammer2_cluster_t *ncluster;
1674 struct namecache *ncp;
1675 const uint8_t *name;
1680 dip = VTOI(ap->a_dvp);
1681 if (dip->pmp->ronly) {
1686 ncp = ap->a_nch->ncp;
1687 name = ncp->nc_name;
1688 name_len = ncp->nc_nlen;
1689 hammer2_pfs_memory_wait(dip->pmp);
1690 hammer2_trans_init(&trans, dip->pmp, HAMMER2_TRANS_NEWINODE);
1693 nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1695 &ncluster, 0, &error);
1697 KKASSERT(nip == NULL);
1700 *ap->a_vpp = hammer2_igetv(nip, ncluster, &error);
1701 hammer2_inode_unlock(nip, ncluster);
1703 hammer2_trans_done(&trans);
1706 cache_setunresolved(ap->a_nch);
1707 cache_setvp(ap->a_nch, *ap->a_vpp);
1714 * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1718 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
1720 hammer2_inode_t *dip;
1721 hammer2_inode_t *nip;
1722 hammer2_cluster_t *ncparent;
1723 hammer2_trans_t trans;
1724 struct namecache *ncp;
1725 const uint8_t *name;
1729 dip = VTOI(ap->a_dvp);
1730 if (dip->pmp->ronly)
1733 ncp = ap->a_nch->ncp;
1734 name = ncp->nc_name;
1735 name_len = ncp->nc_nlen;
1736 hammer2_pfs_memory_wait(dip->pmp);
1737 hammer2_trans_init(&trans, dip->pmp, HAMMER2_TRANS_NEWINODE);
1740 ap->a_vap->va_type = VLNK; /* enforce type */
1742 nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1744 &ncparent, 0, &error);
1746 KKASSERT(nip == NULL);
1748 hammer2_trans_done(&trans);
1751 *ap->a_vpp = hammer2_igetv(nip, ncparent, &error);
1754 * Build the softlink (~like file data) and finalize the namecache.
1760 hammer2_inode_data_t *nipdata;
1762 nipdata = &hammer2_cluster_wdata(ncparent)->ipdata;
1763 /* nipdata = &nip->chain->data->ipdata;XXX */
1764 bytes = strlen(ap->a_target);
1766 if (bytes <= HAMMER2_EMBEDDED_BYTES) {
1767 KKASSERT(nipdata->op_flags &
1768 HAMMER2_OPFLAG_DIRECTDATA);
1769 bcopy(ap->a_target, nipdata->u.data, bytes);
1770 nipdata->size = bytes;
1772 hammer2_cluster_modsync(ncparent);
1773 hammer2_inode_unlock(nip, ncparent);
1774 /* nipdata = NULL; not needed */
1776 hammer2_inode_unlock(nip, ncparent);
1777 /* nipdata = NULL; not needed */
1778 bzero(&auio, sizeof(auio));
1779 bzero(&aiov, sizeof(aiov));
1780 auio.uio_iov = &aiov;
1781 auio.uio_segflg = UIO_SYSSPACE;
1782 auio.uio_rw = UIO_WRITE;
1783 auio.uio_resid = bytes;
1784 auio.uio_iovcnt = 1;
1785 auio.uio_td = curthread;
1786 aiov.iov_base = ap->a_target;
1787 aiov.iov_len = bytes;
1788 error = hammer2_write_file(nip, &auio, IO_APPEND, 0);
1789 /* XXX handle error */
1793 hammer2_inode_unlock(nip, ncparent);
1795 hammer2_trans_done(&trans);
1798 * Finalize namecache
1801 cache_setunresolved(ap->a_nch);
1802 cache_setvp(ap->a_nch, *ap->a_vpp);
1803 /* hammer2_knote(ap->a_dvp, NOTE_WRITE); */
1809 * hammer2_vop_nremove { nch, dvp, cred }
1813 hammer2_vop_nremove(struct vop_nremove_args *ap)
1815 hammer2_inode_t *dip;
1816 hammer2_trans_t trans;
1817 struct namecache *ncp;
1818 const uint8_t *name;
1823 dip = VTOI(ap->a_dvp);
1824 if (dip->pmp->ronly) {
1829 ncp = ap->a_nch->ncp;
1830 name = ncp->nc_name;
1831 name_len = ncp->nc_nlen;
1833 hammer2_pfs_memory_wait(dip->pmp);
1834 hammer2_trans_init(&trans, dip->pmp, 0);
1835 error = hammer2_unlink_file(&trans, dip, name, name_len,
1836 0, NULL, ap->a_nch, -1);
1837 hammer2_run_unlinkq(&trans, dip->pmp);
1838 hammer2_trans_done(&trans);
1840 cache_unlink(ap->a_nch);
1846 * hammer2_vop_nrmdir { nch, dvp, cred }
1850 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
1852 hammer2_inode_t *dip;
1853 hammer2_trans_t trans;
1854 struct namecache *ncp;
1855 const uint8_t *name;
1860 dip = VTOI(ap->a_dvp);
1861 if (dip->pmp->ronly) {
1866 ncp = ap->a_nch->ncp;
1867 name = ncp->nc_name;
1868 name_len = ncp->nc_nlen;
1870 hammer2_pfs_memory_wait(dip->pmp);
1871 hammer2_trans_init(&trans, dip->pmp, 0);
1872 hammer2_run_unlinkq(&trans, dip->pmp);
1873 error = hammer2_unlink_file(&trans, dip, name, name_len,
1874 1, NULL, ap->a_nch, -1);
1875 hammer2_trans_done(&trans);
1877 cache_unlink(ap->a_nch);
1883 * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1887 hammer2_vop_nrename(struct vop_nrename_args *ap)
1889 struct namecache *fncp;
1890 struct namecache *tncp;
1891 hammer2_inode_t *cdip;
1892 hammer2_inode_t *fdip;
1893 hammer2_inode_t *tdip;
1894 hammer2_inode_t *ip;
1895 hammer2_cluster_t *cluster;
1896 hammer2_cluster_t *fdcluster;
1897 hammer2_cluster_t *tdcluster;
1898 hammer2_cluster_t *cdcluster;
1899 hammer2_trans_t trans;
1900 const uint8_t *fname;
1902 const uint8_t *tname;
1908 if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
1910 if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
1913 fdip = VTOI(ap->a_fdvp); /* source directory */
1914 tdip = VTOI(ap->a_tdvp); /* target directory */
1916 if (fdip->pmp->ronly)
1920 fncp = ap->a_fnch->ncp; /* entry name in source */
1921 fname = fncp->nc_name;
1922 fname_len = fncp->nc_nlen;
1924 tncp = ap->a_tnch->ncp; /* entry name in target */
1925 tname = tncp->nc_name;
1926 tname_len = tncp->nc_nlen;
1928 hammer2_pfs_memory_wait(tdip->pmp);
1929 hammer2_trans_init(&trans, tdip->pmp, 0);
1932 * ip is the inode being renamed. If this is a hardlink then
1933 * ip represents the actual file and not the hardlink marker.
1935 ip = VTOI(fncp->nc_vp);
1940 * The common parent directory must be locked first to avoid deadlocks.
1941 * Also note that fdip and/or tdip might match cdip.
1943 * WARNING! fdip may not match ip->pip. That is, if the source file
1944 * is already a hardlink then what we are renaming is the
1945 * hardlink pointer, not the hardlink itself. The hardlink
1946 * directory (ip->pip) will already be at a common parent
1949 * Be sure to use ip->pip when finding the common parent
1950 * against tdip or we might accidently move the hardlink
1951 * target into a subdirectory that makes it inaccessible to
1954 cdip = hammer2_inode_common_parent(ip->pip, tdip);
1955 cdcluster = hammer2_inode_lock(cdip, HAMMER2_RESOLVE_ALWAYS);
1956 fdcluster = hammer2_inode_lock(fdip, HAMMER2_RESOLVE_ALWAYS);
1957 tdcluster = hammer2_inode_lock(tdip, HAMMER2_RESOLVE_ALWAYS);
1960 * Keep a tight grip on the inode so the temporary unlinking from
1961 * the source location prior to linking to the target location
1962 * does not cause the cluster to be destroyed.
1964 * NOTE: To avoid deadlocks we cannot lock (ip) while we are
1965 * unlinking elements from their directories. Locking
1966 * the nlinks field does not lock the whole inode.
1968 hammer2_inode_ref(ip);
1971 * Remove target if it exists.
1973 error = hammer2_unlink_file(&trans, tdip, tname, tname_len,
1974 -1, NULL, ap->a_tnch, -1);
1976 if (error && error != ENOENT)
1980 * When renaming a hardlinked file we may have to re-consolidate
1981 * the location of the hardlink target.
1983 * If ip represents a regular file the consolidation code essentially
1984 * does nothing other than return the same locked cluster that was
1987 * The returned cluster will be locked.
1989 * WARNING! We do not currently have a local copy of ipdata but
1990 * we do use one later remember that it must be reloaded
1991 * on any modification to the inode, including connects.
1993 cluster = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS);
1994 error = hammer2_hardlink_consolidate(&trans, ip, &cluster,
1995 cdip, cdcluster, 0);
2000 * Disconnect (fdip, fname) from the source directory. This will
2001 * disconnect (ip) if it represents a direct file. If (ip) represents
2002 * a hardlink the HARDLINK pointer object will be removed but the
2003 * hardlink will stay intact.
2005 * Always pass nch as NULL because we intend to reconnect the inode,
2006 * so we don't want hammer2_unlink_file() to rename it to the hidden
2007 * open-but-unlinked directory.
2009 * The target cluster may be marked DELETED but will not be destroyed
2010 * since we retain our hold on ip and cluster.
2012 * NOTE: We pass nlinks as 0 (not -1) in order to retain the file's
2015 error = hammer2_unlink_file(&trans, fdip, fname, fname_len,
2016 -1, &hlink, NULL, 0);
2017 KKASSERT(error != EAGAIN);
2022 * Reconnect ip to target directory using cluster. Chains cannot
2023 * actually be moved, so this will duplicate the cluster in the new
2024 * spot and assign it to the ip, replacing the old cluster.
2026 * WARNING: Because recursive locks are allowed and we unlinked the
2027 * file that we have a cluster-in-hand for just above, the
2028 * cluster might have been delete-duplicated. We must
2029 * refactor the cluster.
2031 * WARNING: Chain locks can lock buffer cache buffers, to avoid
2032 * deadlocks we want to unlock before issuing a cache_*()
2033 * op (that might have to lock a vnode).
2035 * NOTE: Pass nlinks as 0 because we retained the link count from
2036 * the unlink, so we do not have to modify it.
2038 error = hammer2_inode_connect(&trans, &cluster, hlink,
2040 tname, tname_len, 0);
2042 KKASSERT(cluster != NULL);
2043 hammer2_inode_repoint(ip, (hlink ? ip->pip : tdip), cluster);
2046 hammer2_inode_unlock(ip, cluster);
2047 hammer2_inode_unlock(tdip, tdcluster);
2048 hammer2_inode_unlock(fdip, fdcluster);
2049 hammer2_inode_unlock(cdip, cdcluster);
2050 hammer2_inode_drop(ip);
2051 hammer2_inode_drop(cdip);
2052 hammer2_run_unlinkq(&trans, fdip->pmp);
2053 hammer2_trans_done(&trans);
2056 * Issue the namecache update after unlocking all the internal
2057 * hammer structures, otherwise we might deadlock.
2059 if (tnch_error == 0) {
2060 cache_unlink(ap->a_tnch);
2061 cache_setunresolved(ap->a_tnch);
2064 cache_rename(ap->a_fnch, ap->a_tnch);
2071 * Strategy code (async logical file buffer I/O from system)
2073 * WARNING: The strategy code cannot safely use hammer2 transactions
2074 * as this can deadlock against vfs_sync's vfsync() call
2075 * if multiple flushes are queued. All H2 structures must
2076 * already be present and ready for the DIO.
2078 * Reads can be initiated asynchronously, writes have to be
2079 * spooled to a separate thread for action to avoid deadlocks.
2081 static int hammer2_strategy_read(struct vop_strategy_args *ap);
2082 static int hammer2_strategy_write(struct vop_strategy_args *ap);
2083 static void hammer2_strategy_read_callback(hammer2_iocb_t *iocb);
2087 hammer2_vop_strategy(struct vop_strategy_args *ap)
2098 error = hammer2_strategy_read(ap);
2099 ++hammer2_iod_file_read;
2102 error = hammer2_strategy_write(ap);
2103 ++hammer2_iod_file_write;
2106 bp->b_error = error = EINVAL;
2107 bp->b_flags |= B_ERROR;
2115 * Logical buffer I/O, async read.
2119 hammer2_strategy_read(struct vop_strategy_args *ap)
2124 hammer2_inode_t *ip;
2125 hammer2_cluster_t *cparent;
2126 hammer2_cluster_t *cluster;
2127 hammer2_key_t key_dummy;
2128 hammer2_key_t lbase;
2133 ip = VTOI(ap->a_vp);
2134 nbio = push_bio(bio);
2136 lbase = bio->bio_offset;
2137 KKASSERT(((int)lbase & HAMMER2_PBUFMASK) == 0);
2140 * Lookup the file offset.
2142 cparent = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS |
2143 HAMMER2_RESOLVE_SHARED);
2144 cluster = hammer2_cluster_lookup(cparent, &key_dummy,
2146 HAMMER2_LOOKUP_NODATA |
2147 HAMMER2_LOOKUP_SHARED);
2148 hammer2_inode_unlock(ip, cparent);
2151 * Data is zero-fill if no cluster could be found
2152 * (XXX or EIO on a cluster failure).
2154 if (cluster == NULL) {
2157 bzero(bp->b_data, bp->b_bcount);
2163 * Cluster elements must be type INODE or type DATA, but the
2164 * compression mode (or not) for DATA chains can be different for
2165 * each chain. This will be handled by the callback.
2167 * If the cluster already has valid data the callback will be made
2168 * immediately/synchronously.
2170 btype = hammer2_cluster_type(cluster);
2171 if (btype != HAMMER2_BREF_TYPE_INODE &&
2172 btype != HAMMER2_BREF_TYPE_DATA) {
2173 panic("READ PATH: hammer2_strategy_read: unknown bref type");
2175 hammer2_cluster_load_async(cluster, hammer2_strategy_read_callback,
2181 * Read callback for hammer2_cluster_load_async(). The load function may
2182 * start several actual I/Os but will only make one callback, typically with
2183 * the first valid I/O XXX
2187 hammer2_strategy_read_callback(hammer2_iocb_t *iocb)
2189 struct bio *bio = iocb->ptr; /* original logical buffer */
2190 struct buf *bp = bio->bio_buf; /* original logical buffer */
2191 hammer2_chain_t *chain;
2192 hammer2_cluster_t *cluster;
2198 * Extract data and handle iteration on I/O failure. iocb->off
2199 * is the cluster index for iteration.
2201 cluster = iocb->cluster;
2202 dio = iocb->dio; /* can be NULL if iocb not in progress */
2205 * Work to do if INPROG set, else dio is already good or dio is
2206 * NULL (which is the shortcut case if chain->data is already good).
2208 if (iocb->flags & HAMMER2_IOCB_INPROG) {
2210 * Read attempt not yet made. Issue an asynchronous read
2211 * if necessary and return, operation will chain back to
2214 if ((iocb->flags & HAMMER2_IOCB_READ) == 0) {
2215 if (dio->bp == NULL ||
2216 (dio->bp->b_flags & B_CACHE) == 0) {
2221 iocb->flags |= HAMMER2_IOCB_READ;
2222 breadcb(dio->hmp->devvp,
2223 dio->pbase, dio->psize,
2224 hammer2_io_callback, iocb);
2231 * If we have a DIO it is now done, check for an error and
2232 * calculate the data.
2234 * If there is no DIO it is an optimization by
2235 * hammer2_cluster_load_async(), the data is available in
2239 if (dio->bp->b_flags & B_ERROR) {
2240 i = (int)iocb->lbase + 1;
2241 if (i >= cluster->nchains) {
2242 bp->b_flags |= B_ERROR;
2243 bp->b_error = dio->bp->b_error;
2244 hammer2_io_complete(iocb);
2246 hammer2_cluster_unlock(cluster);
2248 hammer2_io_complete(iocb); /* XXX */
2249 chain = cluster->array[i].chain;
2250 kprintf("hammer2: IO CHAIN-%d %p\n", i, chain);
2251 hammer2_adjreadcounter(&chain->bref,
2253 iocb->chain = chain;
2254 iocb->lbase = (off_t)i;
2257 hammer2_io_getblk(chain->hmp,
2258 chain->bref.data_off,
2264 chain = iocb->chain;
2265 data = hammer2_io_data(dio, chain->bref.data_off);
2268 * Special synchronous case, data present in chain->data.
2270 chain = iocb->chain;
2271 data = (void *)chain->data;
2274 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) {
2276 * Data is embedded in the inode (copy from inode).
2278 bcopy(((hammer2_inode_data_t *)data)->u.data,
2279 bp->b_data, HAMMER2_EMBEDDED_BYTES);
2280 bzero(bp->b_data + HAMMER2_EMBEDDED_BYTES,
2281 bp->b_bcount - HAMMER2_EMBEDDED_BYTES);
2284 } else if (chain->bref.type == HAMMER2_BREF_TYPE_DATA) {
2286 * Data is on-media, issue device I/O and copy.
2288 * XXX direct-IO shortcut could go here XXX.
2290 switch (HAMMER2_DEC_COMP(chain->bref.methods)) {
2291 case HAMMER2_COMP_LZ4:
2292 hammer2_decompress_LZ4_callback(data, chain->bytes,
2295 case HAMMER2_COMP_ZLIB:
2296 hammer2_decompress_ZLIB_callback(data, chain->bytes,
2299 case HAMMER2_COMP_NONE:
2300 KKASSERT(chain->bytes <= bp->b_bcount);
2301 bcopy(data, bp->b_data, chain->bytes);
2302 if (chain->bytes < bp->b_bcount) {
2303 bzero(bp->b_data + chain->bytes,
2304 bp->b_bcount - chain->bytes);
2306 bp->b_flags |= B_NOTMETA;
2311 panic("hammer2_strategy_read: "
2312 "unknown compression type");
2315 /* bqrelse the dio to help stabilize the call to panic() */
2317 hammer2_io_bqrelse(&dio);
2318 panic("hammer2_strategy_read: unknown bref type");
2322 * Once the iocb is cleaned up the DIO (if any) will no longer be
2323 * in-progress but will still have a ref. Be sure to release
2326 hammer2_io_complete(iocb); /* physical management */
2327 if (dio) /* physical dio & buffer */
2328 hammer2_io_bqrelse(&dio);
2329 hammer2_cluster_unlock(cluster); /* cluster management */
2330 biodone(bio); /* logical buffer */
2335 hammer2_strategy_write(struct vop_strategy_args *ap)
2340 hammer2_inode_t *ip;
2344 ip = VTOI(ap->a_vp);
2347 hammer2_lwinprog_ref(pmp);
2348 hammer2_mtx_ex(&pmp->wthread_mtx);
2349 if (TAILQ_EMPTY(&pmp->wthread_bioq.queue)) {
2350 bioq_insert_tail(&pmp->wthread_bioq, ap->a_bio);
2351 hammer2_mtx_unlock(&pmp->wthread_mtx);
2352 wakeup(&pmp->wthread_bioq);
2354 bioq_insert_tail(&pmp->wthread_bioq, ap->a_bio);
2355 hammer2_mtx_unlock(&pmp->wthread_mtx);
2357 hammer2_lwinprog_wait(pmp);
2363 * hammer2_vop_ioctl { vp, command, data, fflag, cred }
2367 hammer2_vop_ioctl(struct vop_ioctl_args *ap)
2369 hammer2_inode_t *ip;
2373 ip = VTOI(ap->a_vp);
2375 error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
2376 ap->a_fflag, ap->a_cred);
2383 hammer2_vop_mountctl(struct vop_mountctl_args *ap)
2391 case (MOUNTCTL_SET_EXPORT):
2392 mp = ap->a_head.a_ops->head.vv_mount;
2395 if (ap->a_ctllen != sizeof(struct export_args))
2398 rc = vfs_export(mp, &pmp->export,
2399 (const struct export_args *)ap->a_ctl);
2402 rc = vop_stdmountctl(ap);
2410 * This handles unlinked open files after the vnode is finally dereferenced.
2411 * To avoid deadlocks it cannot be called from the normal vnode recycling
2412 * path, so we call it (1) after a unlink, rmdir, or rename, (2) on every
2413 * flush, and (3) on umount.
2416 hammer2_run_unlinkq(hammer2_trans_t *trans, hammer2_pfs_t *pmp)
2418 const hammer2_inode_data_t *ripdata;
2419 hammer2_inode_unlink_t *ipul;
2420 hammer2_inode_t *ip;
2421 hammer2_cluster_t *cluster;
2422 hammer2_cluster_t *cparent;
2424 if (TAILQ_EMPTY(&pmp->unlinkq))
2428 hammer2_spin_ex(&pmp->list_spin);
2429 while ((ipul = TAILQ_FIRST(&pmp->unlinkq)) != NULL) {
2430 TAILQ_REMOVE(&pmp->unlinkq, ipul, entry);
2431 hammer2_spin_unex(&pmp->list_spin);
2433 kfree(ipul, pmp->minode);
2435 cluster = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS);
2436 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
2437 if (hammer2_debug & 0x400) {
2438 kprintf("hammer2: unlink on reclaim: %s refs=%d\n",
2439 ripdata->filename, ip->refs);
2441 KKASSERT(ripdata->nlinks == 0);
2443 cparent = hammer2_cluster_parent(cluster);
2444 hammer2_cluster_delete(trans, cparent, cluster,
2445 HAMMER2_DELETE_PERMANENT);
2446 hammer2_cluster_unlock(cparent);
2447 hammer2_inode_unlock(ip, cluster); /* inode lock */
2448 hammer2_inode_drop(ip); /* ipul ref */
2450 hammer2_spin_ex(&pmp->list_spin);
2452 hammer2_spin_unex(&pmp->list_spin);
2460 static void filt_hammer2detach(struct knote *kn);
2461 static int filt_hammer2read(struct knote *kn, long hint);
2462 static int filt_hammer2write(struct knote *kn, long hint);
2463 static int filt_hammer2vnode(struct knote *kn, long hint);
2465 static struct filterops hammer2read_filtops =
2466 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2467 NULL, filt_hammer2detach, filt_hammer2read };
2468 static struct filterops hammer2write_filtops =
2469 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2470 NULL, filt_hammer2detach, filt_hammer2write };
2471 static struct filterops hammer2vnode_filtops =
2472 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2473 NULL, filt_hammer2detach, filt_hammer2vnode };
2477 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap)
2479 struct vnode *vp = ap->a_vp;
2480 struct knote *kn = ap->a_kn;
2482 switch (kn->kn_filter) {
2484 kn->kn_fop = &hammer2read_filtops;
2487 kn->kn_fop = &hammer2write_filtops;
2490 kn->kn_fop = &hammer2vnode_filtops;
2493 return (EOPNOTSUPP);
2496 kn->kn_hook = (caddr_t)vp;
2498 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2504 filt_hammer2detach(struct knote *kn)
2506 struct vnode *vp = (void *)kn->kn_hook;
2508 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2512 filt_hammer2read(struct knote *kn, long hint)
2514 struct vnode *vp = (void *)kn->kn_hook;
2515 hammer2_inode_t *ip = VTOI(vp);
2518 if (hint == NOTE_REVOKE) {
2519 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2522 off = ip->size - kn->kn_fp->f_offset;
2523 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2524 if (kn->kn_sfflags & NOTE_OLDAPI)
2526 return (kn->kn_data != 0);
2531 filt_hammer2write(struct knote *kn, long hint)
2533 if (hint == NOTE_REVOKE)
2534 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2540 filt_hammer2vnode(struct knote *kn, long hint)
2542 if (kn->kn_sfflags & hint)
2543 kn->kn_fflags |= hint;
2544 if (hint == NOTE_REVOKE) {
2545 kn->kn_flags |= (EV_EOF | EV_NODATA);
2548 return (kn->kn_fflags != 0);
2556 hammer2_vop_markatime(struct vop_markatime_args *ap)
2558 hammer2_inode_t *ip;
2571 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap)
2575 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2577 error = hammer2_vop_kqfilter(ap);
2584 struct vop_ops hammer2_vnode_vops = {
2585 .vop_default = vop_defaultop,
2586 .vop_fsync = hammer2_vop_fsync,
2587 .vop_getpages = vop_stdgetpages,
2588 .vop_putpages = vop_stdputpages,
2589 .vop_access = hammer2_vop_access,
2590 .vop_advlock = hammer2_vop_advlock,
2591 .vop_close = hammer2_vop_close,
2592 .vop_nlink = hammer2_vop_nlink,
2593 .vop_ncreate = hammer2_vop_ncreate,
2594 .vop_nsymlink = hammer2_vop_nsymlink,
2595 .vop_nremove = hammer2_vop_nremove,
2596 .vop_nrmdir = hammer2_vop_nrmdir,
2597 .vop_nrename = hammer2_vop_nrename,
2598 .vop_getattr = hammer2_vop_getattr,
2599 .vop_setattr = hammer2_vop_setattr,
2600 .vop_readdir = hammer2_vop_readdir,
2601 .vop_readlink = hammer2_vop_readlink,
2602 .vop_getpages = vop_stdgetpages,
2603 .vop_putpages = vop_stdputpages,
2604 .vop_read = hammer2_vop_read,
2605 .vop_write = hammer2_vop_write,
2606 .vop_open = hammer2_vop_open,
2607 .vop_inactive = hammer2_vop_inactive,
2608 .vop_reclaim = hammer2_vop_reclaim,
2609 .vop_nresolve = hammer2_vop_nresolve,
2610 .vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
2611 .vop_nmkdir = hammer2_vop_nmkdir,
2612 .vop_nmknod = hammer2_vop_nmknod,
2613 .vop_ioctl = hammer2_vop_ioctl,
2614 .vop_mountctl = hammer2_vop_mountctl,
2615 .vop_bmap = hammer2_vop_bmap,
2616 .vop_strategy = hammer2_vop_strategy,
2617 .vop_kqfilter = hammer2_vop_kqfilter
2620 struct vop_ops hammer2_spec_vops = {
2621 .vop_default = vop_defaultop,
2622 .vop_fsync = hammer2_vop_fsync,
2623 .vop_read = vop_stdnoread,
2624 .vop_write = vop_stdnowrite,
2625 .vop_access = hammer2_vop_access,
2626 .vop_close = hammer2_vop_close,
2627 .vop_markatime = hammer2_vop_markatime,
2628 .vop_getattr = hammer2_vop_getattr,
2629 .vop_inactive = hammer2_vop_inactive,
2630 .vop_reclaim = hammer2_vop_reclaim,
2631 .vop_setattr = hammer2_vop_setattr
2634 struct vop_ops hammer2_fifo_vops = {
2635 .vop_default = fifo_vnoperate,
2636 .vop_fsync = hammer2_vop_fsync,
2638 .vop_read = hammer2_vop_fiforead,
2639 .vop_write = hammer2_vop_fifowrite,
2641 .vop_access = hammer2_vop_access,
2643 .vop_close = hammer2_vop_fifoclose,
2645 .vop_markatime = hammer2_vop_markatime,
2646 .vop_getattr = hammer2_vop_getattr,
2647 .vop_inactive = hammer2_vop_inactive,
2648 .vop_reclaim = hammer2_vop_reclaim,
2649 .vop_setattr = hammer2_vop_setattr,
2650 .vop_kqfilter = hammer2_vop_fifokqfilter