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_nex(ip, HAMMER2_RESOLVE_NEVER);
213 * Check for deleted inodes and recycle immediately.
215 * WARNING: nvtruncbuf() can only be safely called without the inode
216 * lock held due to the way our write thread works.
218 if (hammer2_cluster_isunlinked(cluster)) {
222 nblksize = hammer2_calc_logical(ip, 0, &lbase, NULL);
223 hammer2_inode_unlock_ex(ip, cluster);
224 nvtruncbuf(vp, 0, nblksize, 0, 0);
227 hammer2_inode_unlock_ex(ip, cluster);
234 * Reclaim a vnode so that it can be reused; after the inode is
235 * disassociated, the filesystem must manage it alone.
239 hammer2_vop_reclaim(struct vop_reclaim_args *ap)
241 hammer2_cluster_t *cluster;
255 * Inode must be locked for reclaim.
258 cluster = hammer2_inode_lock_nex(ip, HAMMER2_RESOLVE_NEVER);
261 * The final close of a deleted file or directory marks it for
262 * destruction. The DELETED flag allows the flusher to shortcut
263 * any modified blocks still unflushed (that is, just ignore them).
265 * HAMMER2 usually does not try to optimize the freemap by returning
266 * deleted blocks to it as it does not usually know how many snapshots
267 * might be referencing portions of the file/dir.
273 * NOTE! We do not attempt to flush chains here, flushing is
274 * really fragile and could also deadlock.
279 * A reclaim can occur at any time so we cannot safely start a
280 * transaction to handle reclamation of unlinked files. Instead,
281 * the ip is left with a reference and placed on a linked list and
284 if (hammer2_cluster_isunlinked(cluster)) {
285 hammer2_inode_unlink_t *ipul;
287 ipul = kmalloc(sizeof(*ipul), pmp->minode, M_WAITOK | M_ZERO);
290 hammer2_spin_ex(&pmp->list_spin);
291 TAILQ_INSERT_TAIL(&pmp->unlinkq, ipul, entry);
292 hammer2_spin_unex(&pmp->list_spin);
293 hammer2_inode_unlock_ex(ip, cluster); /* unlock */
294 /* retain ref from vp for ipul */
296 hammer2_inode_unlock_ex(ip, cluster); /* unlock */
297 hammer2_inode_drop(ip); /* vp ref */
299 /* cluster no longer referenced */
300 /* cluster = NULL; not needed */
303 * XXX handle background sync when ip dirty, kernel will no longer
304 * notify us regarding this inode because there is no longer a
305 * vnode attached to it.
314 hammer2_vop_fsync(struct vop_fsync_args *ap)
317 hammer2_trans_t trans;
318 hammer2_cluster_t *cluster;
326 /* XXX can't do this yet */
327 hammer2_trans_init(&trans, ip->pmp, HAMMER2_TRANS_ISFLUSH);
328 vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
330 hammer2_trans_init(&trans, ip->pmp, 0);
331 vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
334 * Calling chain_flush here creates a lot of duplicative
335 * COW operations due to non-optimal vnode ordering.
337 * Only do it for an actual fsync() syscall. The other forms
338 * which call this function will eventually call chain_flush
339 * on the volume root as a catch-all, which is far more optimal.
341 cluster = hammer2_inode_lock_ex(ip);
342 atomic_clear_int(&ip->flags, HAMMER2_INODE_MODIFIED);
344 if (ip->flags & (HAMMER2_INODE_RESIZED|HAMMER2_INODE_MTIME))
345 hammer2_inode_fsync(&trans, ip, cluster);
349 * XXX creates discontinuity w/modify_tid
351 if (ap->a_flags & VOP_FSYNC_SYSCALL) {
352 hammer2_flush(&trans, cluster);
355 hammer2_inode_unlock_ex(ip, cluster);
356 hammer2_trans_done(&trans);
364 hammer2_vop_access(struct vop_access_args *ap)
366 hammer2_inode_t *ip = VTOI(ap->a_vp);
367 const hammer2_inode_data_t *ripdata;
368 hammer2_cluster_t *cluster;
374 cluster = hammer2_inode_lock_sh(ip);
375 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
376 uid = hammer2_to_unix_xid(&ripdata->uid);
377 gid = hammer2_to_unix_xid(&ripdata->gid);
378 error = vop_helper_access(ap, uid, gid, ripdata->mode, ripdata->uflags);
379 hammer2_inode_unlock_sh(ip, cluster);
387 hammer2_vop_getattr(struct vop_getattr_args *ap)
389 const hammer2_inode_data_t *ripdata;
390 hammer2_cluster_t *cluster;
403 cluster = hammer2_inode_lock_sh(ip);
404 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
405 KKASSERT(hammer2_cluster_type(cluster) == HAMMER2_BREF_TYPE_INODE);
407 vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0];
408 vap->va_fileid = ripdata->inum;
409 vap->va_mode = ripdata->mode;
410 vap->va_nlink = ripdata->nlinks;
411 vap->va_uid = hammer2_to_unix_xid(&ripdata->uid);
412 vap->va_gid = hammer2_to_unix_xid(&ripdata->gid);
415 vap->va_size = ip->size; /* protected by shared lock */
416 vap->va_blocksize = HAMMER2_PBUFSIZE;
417 vap->va_flags = ripdata->uflags;
418 hammer2_time_to_timespec(ripdata->ctime, &vap->va_ctime);
419 hammer2_time_to_timespec(ripdata->mtime, &vap->va_mtime);
420 hammer2_time_to_timespec(ripdata->mtime, &vap->va_atime);
422 vap->va_bytes = vap->va_size; /* XXX */
423 vap->va_type = hammer2_get_vtype(ripdata);
425 vap->va_uid_uuid = ripdata->uid;
426 vap->va_gid_uuid = ripdata->gid;
427 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
430 hammer2_inode_unlock_sh(ip, cluster);
438 hammer2_vop_setattr(struct vop_setattr_args *ap)
440 const hammer2_inode_data_t *ripdata;
441 hammer2_inode_data_t *wipdata;
443 hammer2_cluster_t *cluster;
444 hammer2_trans_t trans;
456 hammer2_update_time(&ctime);
460 if (ip->pmp->ronly) {
465 hammer2_pfs_memory_wait(ip->pmp);
466 hammer2_trans_init(&trans, ip->pmp, 0);
467 cluster = hammer2_inode_lock_ex(ip);
468 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
471 if (vap->va_flags != VNOVAL) {
474 flags = ripdata->uflags;
475 error = vop_helper_setattr_flags(&flags, vap->va_flags,
476 hammer2_to_unix_xid(&ripdata->uid),
479 if (ripdata->uflags != flags) {
480 wipdata = hammer2_cluster_modify_ip(&trans, ip,
482 wipdata->uflags = flags;
483 wipdata->ctime = ctime;
484 kflags |= NOTE_ATTRIB;
488 if (ripdata->uflags & (IMMUTABLE | APPEND)) {
495 if (ripdata->uflags & (IMMUTABLE | APPEND)) {
499 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
500 mode_t cur_mode = ripdata->mode;
501 uid_t cur_uid = hammer2_to_unix_xid(&ripdata->uid);
502 gid_t cur_gid = hammer2_to_unix_xid(&ripdata->gid);
506 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
508 &cur_uid, &cur_gid, &cur_mode);
510 hammer2_guid_to_uuid(&uuid_uid, cur_uid);
511 hammer2_guid_to_uuid(&uuid_gid, cur_gid);
512 if (bcmp(&uuid_uid, &ripdata->uid, sizeof(uuid_uid)) ||
513 bcmp(&uuid_gid, &ripdata->gid, sizeof(uuid_gid)) ||
514 ripdata->mode != cur_mode
516 wipdata = hammer2_cluster_modify_ip(&trans, ip,
518 wipdata->uid = uuid_uid;
519 wipdata->gid = uuid_gid;
520 wipdata->mode = cur_mode;
521 wipdata->ctime = ctime;
525 kflags |= NOTE_ATTRIB;
532 if (vap->va_size != VNOVAL && ip->size != vap->va_size) {
535 if (vap->va_size == ip->size)
537 hammer2_inode_unlock_ex(ip, cluster);
538 if (vap->va_size < ip->size) {
539 hammer2_truncate_file(ip, vap->va_size);
541 hammer2_extend_file(ip, vap->va_size);
543 cluster = hammer2_inode_lock_ex(ip);
545 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
554 /* atime not supported */
555 if (vap->va_atime.tv_sec != VNOVAL) {
556 wipdata = hammer2_cluster_modify_ip(&trans, ip, cluster, 0);
557 wipdata->atime = hammer2_timespec_to_time(&vap->va_atime);
558 kflags |= NOTE_ATTRIB;
563 if (vap->va_mtime.tv_sec != VNOVAL) {
564 wipdata = hammer2_cluster_modify_ip(&trans, ip, cluster, 0);
565 wipdata->mtime = hammer2_timespec_to_time(&vap->va_mtime);
566 kflags |= NOTE_ATTRIB;
571 if (vap->va_mode != (mode_t)VNOVAL) {
572 mode_t cur_mode = ripdata->mode;
573 uid_t cur_uid = hammer2_to_unix_xid(&ripdata->uid);
574 gid_t cur_gid = hammer2_to_unix_xid(&ripdata->gid);
576 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
577 cur_uid, cur_gid, &cur_mode);
578 if (error == 0 && ripdata->mode != cur_mode) {
579 wipdata = hammer2_cluster_modify_ip(&trans, ip,
581 wipdata->mode = cur_mode;
582 wipdata->ctime = ctime;
583 kflags |= NOTE_ATTRIB;
590 * If a truncation occurred we must call inode_fsync() now in order
591 * to trim the related data chains, otherwise a later expansion can
595 hammer2_cluster_modsync(cluster);
598 hammer2_inode_fsync(&trans, ip, cluster);
601 * Cleanup. If domtime is set an additional inode modification
602 * must be flagged. All other modifications will have already
603 * set INODE_MODIFIED and called vsetisdirty().
607 atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED |
608 HAMMER2_INODE_MTIME);
612 hammer2_cluster_modsync(cluster);
613 hammer2_inode_unlock_ex(ip, cluster);
614 hammer2_trans_done(&trans);
615 hammer2_knote(ip->vp, kflags);
623 hammer2_vop_readdir(struct vop_readdir_args *ap)
625 const hammer2_inode_data_t *ripdata;
627 hammer2_inode_t *xip;
628 hammer2_cluster_t *cparent;
629 hammer2_cluster_t *cluster;
630 hammer2_cluster_t *xcluster;
631 hammer2_blockref_t bref;
633 hammer2_key_t key_next;
648 saveoff = uio->uio_offset;
651 * Setup cookies directory entry cookies if requested
653 if (ap->a_ncookies) {
654 ncookies = uio->uio_resid / 16 + 1;
657 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
664 cparent = hammer2_inode_lock_sh(ip);
665 ripdata = &hammer2_cluster_rdata(cparent)->ipdata;
668 * Handle artificial entries. To ensure that only positive 64 bit
669 * quantities are returned to userland we always strip off bit 63.
670 * The hash code is designed such that codes 0x0000-0x7FFF are not
671 * used, allowing us to use these codes for articial entries.
673 * Entry 0 is used for '.' and entry 1 is used for '..'. Do not
674 * allow '..' to cross the mount point into (e.g.) the super-root.
677 cluster = (void *)(intptr_t)-1; /* non-NULL for early goto done case */
680 inum = ripdata->inum & HAMMER2_DIRHASH_USERMSK;
681 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
685 cookies[cookie_index] = saveoff;
688 if (cookie_index == ncookies)
694 * Be careful with lockorder when accessing ".."
696 * (ip is the current dir. xip is the parent dir).
698 inum = ripdata->inum & HAMMER2_DIRHASH_USERMSK;
699 while (ip->pip != NULL && ip != ip->pmp->iroot) {
701 hammer2_inode_ref(xip);
702 hammer2_inode_unlock_sh(ip, cparent);
703 xcluster = hammer2_inode_lock_sh(xip);
704 cparent = hammer2_inode_lock_sh(ip);
705 hammer2_inode_drop(xip);
706 ripdata = &hammer2_cluster_rdata(cparent)->ipdata;
707 if (xip == ip->pip) {
708 inum = hammer2_cluster_rdata(xcluster)->
709 ipdata.inum & HAMMER2_DIRHASH_USERMSK;
710 hammer2_inode_unlock_sh(xip, xcluster);
713 hammer2_inode_unlock_sh(xip, xcluster);
715 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
719 cookies[cookie_index] = saveoff;
722 if (cookie_index == ncookies)
726 lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
727 if (hammer2_debug & 0x0020)
728 kprintf("readdir: lkey %016jx\n", lkey);
731 * parent is the inode cluster, already locked for us. Don't
732 * double lock shared locks as this will screw up upgrades.
737 cluster = hammer2_cluster_lookup(cparent, &key_next, lkey, lkey,
738 HAMMER2_LOOKUP_SHARED, &ddflag);
739 if (cluster == NULL) {
740 cluster = hammer2_cluster_lookup(cparent, &key_next,
741 lkey, (hammer2_key_t)-1,
742 HAMMER2_LOOKUP_SHARED, &ddflag);
745 hammer2_cluster_bref(cluster, &bref);
747 if (hammer2_debug & 0x0020)
748 kprintf("readdir: p=%p chain=%p %016jx (next %016jx)\n",
749 cparent->focus, cluster->focus,
752 if (bref.type == HAMMER2_BREF_TYPE_INODE) {
753 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
754 dtype = hammer2_get_dtype(ripdata);
755 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
756 r = vop_write_dirent(&error, uio,
758 HAMMER2_DIRHASH_USERMSK,
765 cookies[cookie_index] = saveoff;
768 /* XXX chain error */
769 kprintf("bad chain type readdir %d\n", bref.type);
773 * Keys may not be returned in order so once we have a
774 * placemarker (cluster) the scan must allow the full range
775 * or some entries will be missed.
777 cluster = hammer2_cluster_next(cparent, cluster, &key_next,
778 key_next, (hammer2_key_t)-1,
779 HAMMER2_LOOKUP_SHARED);
781 hammer2_cluster_bref(cluster, &bref);
782 saveoff = (bref.key & HAMMER2_DIRHASH_USERMSK) + 1;
784 saveoff = (hammer2_key_t)-1;
786 if (cookie_index == ncookies)
790 hammer2_cluster_unlock(cluster);
792 hammer2_inode_unlock_sh(ip, cparent);
794 *ap->a_eofflag = (cluster == NULL);
795 if (hammer2_debug & 0x0020)
796 kprintf("readdir: done at %016jx\n", saveoff);
797 uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
798 if (error && cookie_index == 0) {
800 kfree(cookies, M_TEMP);
802 *ap->a_cookies = NULL;
806 *ap->a_ncookies = cookie_index;
807 *ap->a_cookies = cookies;
815 * hammer2_vop_readlink { vp, uio, cred }
819 hammer2_vop_readlink(struct vop_readlink_args *ap)
826 if (vp->v_type != VLNK)
830 error = hammer2_read_file(ip, ap->a_uio, 0);
836 hammer2_vop_read(struct vop_read_args *ap)
846 * Read operations supported on this vnode?
849 if (vp->v_type != VREG)
859 seqcount = ap->a_ioflag >> 16;
860 bigread = (uio->uio_resid > 100 * 1024 * 1024);
862 error = hammer2_read_file(ip, uio, seqcount);
868 hammer2_vop_write(struct vop_write_args *ap)
871 hammer2_trans_t trans;
880 * Read operations supported on this vnode?
883 if (vp->v_type != VREG)
892 if (ip->pmp->ronly) {
896 seqcount = ap->a_ioflag >> 16;
897 bigwrite = (uio->uio_resid > 100 * 1024 * 1024);
900 * Check resource limit
902 if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc &&
903 uio->uio_offset + uio->uio_resid >
904 td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
905 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
909 bigwrite = (uio->uio_resid > 100 * 1024 * 1024);
912 * The transaction interlocks against flushes initiations
913 * (note: but will run concurrently with the actual flush).
915 hammer2_trans_init(&trans, ip->pmp, 0);
916 error = hammer2_write_file(ip, uio, ap->a_ioflag, seqcount);
917 hammer2_trans_done(&trans);
923 * Perform read operations on a file or symlink given an UNLOCKED
926 * The passed ip is not locked.
930 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
941 * WARNING! Assumes that the kernel interlocks size changes at the
944 hammer2_mtx_sh(&ip->lock);
946 hammer2_mtx_unlock(&ip->lock);
948 while (uio->uio_resid > 0 && uio->uio_offset < size) {
955 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
958 error = cluster_read(ip->vp, leof, lbase, lblksize,
959 uio->uio_resid, seqcount * BKVASIZE,
964 loff = (int)(uio->uio_offset - lbase);
966 if (n > uio->uio_resid)
968 if (n > size - uio->uio_offset)
969 n = (int)(size - uio->uio_offset);
970 bp->b_flags |= B_AGE;
971 uiomove((char *)bp->b_data + loff, n, uio);
978 * Write to the file represented by the inode via the logical buffer cache.
979 * The inode may represent a regular file or a symlink.
981 * The inode must not be locked.
985 hammer2_write_file(hammer2_inode_t *ip,
986 struct uio *uio, int ioflag, int seqcount)
988 hammer2_key_t old_eof;
989 hammer2_key_t new_eof;
998 * WARNING! Assumes that the kernel interlocks size changes at the
1001 hammer2_mtx_ex(&ip->lock);
1002 if (ioflag & IO_APPEND)
1003 uio->uio_offset = ip->size;
1005 hammer2_mtx_unlock(&ip->lock);
1008 * Extend the file if necessary. If the write fails at some point
1009 * we will truncate it back down to cover as much as we were able
1012 * Doing this now makes it easier to calculate buffer sizes in
1019 if (uio->uio_offset + uio->uio_resid > old_eof) {
1020 new_eof = uio->uio_offset + uio->uio_resid;
1022 hammer2_extend_file(ip, new_eof);
1023 kflags |= NOTE_EXTEND;
1031 while (uio->uio_resid > 0) {
1032 hammer2_key_t lbase;
1040 * Don't allow the buffer build to blow out the buffer
1043 if ((ioflag & IO_RECURSE) == 0)
1044 bwillwrite(HAMMER2_PBUFSIZE);
1047 * This nominally tells us how much we can cluster and
1048 * what the logical buffer size needs to be. Currently
1049 * we don't try to cluster the write and just handle one
1052 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
1054 loff = (int)(uio->uio_offset - lbase);
1056 KKASSERT(lblksize <= 65536);
1059 * Calculate bytes to copy this transfer and whether the
1060 * copy completely covers the buffer or not.
1063 n = lblksize - loff;
1064 if (n > uio->uio_resid) {
1066 if (loff == lbase && uio->uio_offset + n == new_eof)
1078 if (uio->uio_segflg == UIO_NOCOPY) {
1080 * Issuing a write with the same data backing the
1081 * buffer. Instantiate the buffer to collect the
1082 * backing vm pages, then read-in any missing bits.
1084 * This case is used by vop_stdputpages().
1086 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
1087 if ((bp->b_flags & B_CACHE) == 0) {
1089 error = bread(ip->vp, lbase, lblksize, &bp);
1091 } else if (trivial) {
1093 * Even though we are entirely overwriting the buffer
1094 * we may still have to zero it out to avoid a
1095 * mmap/write visibility issue.
1097 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
1098 if ((bp->b_flags & B_CACHE) == 0)
1102 * Partial overwrite, read in any missing bits then
1103 * replace the portion being written.
1105 * (The strategy code will detect zero-fill physical
1106 * blocks for this case).
1108 error = bread(ip->vp, lbase, lblksize, &bp);
1119 * Ok, copy the data in
1121 error = uiomove(bp->b_data + loff, n, uio);
1122 kflags |= NOTE_WRITE;
1130 * WARNING: Pageout daemon will issue UIO_NOCOPY writes
1131 * with IO_SYNC or IO_ASYNC set. These writes
1132 * must be handled as the pageout daemon expects.
1134 if (ioflag & IO_SYNC) {
1136 } else if ((ioflag & IO_DIRECT) && endofblk) {
1138 } else if (ioflag & IO_ASYNC) {
1146 * Cleanup. If we extended the file EOF but failed to write through
1147 * the entire write is a failure and we have to back-up.
1149 if (error && new_eof != old_eof) {
1150 hammer2_truncate_file(ip, old_eof);
1151 } else if (modified) {
1152 hammer2_mtx_ex(&ip->lock);
1153 hammer2_update_time(&ip->mtime);
1154 atomic_set_int(&ip->flags, HAMMER2_INODE_MTIME);
1155 hammer2_mtx_unlock(&ip->lock);
1157 atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED);
1158 hammer2_knote(ip->vp, kflags);
1159 vsetisdirty(ip->vp);
1165 * Truncate the size of a file. The inode must not be locked.
1167 * NOTE: Caller handles setting HAMMER2_INODE_MODIFIED
1169 * WARNING: nvtruncbuf() can only be safely called without the inode lock
1170 * held due to the way our write thread works.
1172 * WARNING! Assumes that the kernel interlocks size changes at the
1177 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1179 hammer2_key_t lbase;
1184 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1185 nvtruncbuf(ip->vp, nsize,
1186 nblksize, (int)nsize & (nblksize - 1),
1189 hammer2_mtx_ex(&ip->lock);
1191 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1192 hammer2_mtx_unlock(&ip->lock);
1197 * Extend the size of a file. The inode must not be locked.
1199 * WARNING! Assumes that the kernel interlocks size changes at the
1202 * NOTE: Caller handles setting HAMMER2_INODE_MODIFIED
1206 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1208 hammer2_key_t lbase;
1209 hammer2_key_t osize;
1214 hammer2_mtx_ex(&ip->lock);
1217 hammer2_mtx_unlock(&ip->lock);
1220 oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL);
1221 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1227 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1233 hammer2_vop_nresolve(struct vop_nresolve_args *ap)
1235 hammer2_inode_t *ip;
1236 hammer2_inode_t *dip;
1237 hammer2_cluster_t *cparent;
1238 hammer2_cluster_t *cluster;
1239 const hammer2_inode_data_t *ripdata;
1240 hammer2_key_t key_next;
1242 struct namecache *ncp;
1243 const uint8_t *name;
1250 dip = VTOI(ap->a_dvp);
1251 ncp = ap->a_nch->ncp;
1252 name = ncp->nc_name;
1253 name_len = ncp->nc_nlen;
1254 lhc = hammer2_dirhash(name, name_len);
1257 * Note: In DragonFly the kernel handles '.' and '..'.
1259 cparent = hammer2_inode_lock_sh(dip);
1260 cluster = hammer2_cluster_lookup(cparent, &key_next,
1261 lhc, lhc + HAMMER2_DIRHASH_LOMASK,
1262 HAMMER2_LOOKUP_SHARED, &ddflag);
1264 if (hammer2_cluster_type(cluster) == HAMMER2_BREF_TYPE_INODE) {
1265 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
1266 if (ripdata->name_len == name_len &&
1267 bcmp(ripdata->filename, name, name_len) == 0) {
1271 cluster = hammer2_cluster_next(cparent, cluster, &key_next,
1273 lhc + HAMMER2_DIRHASH_LOMASK,
1274 HAMMER2_LOOKUP_SHARED);
1276 hammer2_inode_unlock_sh(dip, cparent);
1279 * Resolve hardlink entries before acquiring the inode.
1282 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
1283 if (ripdata->type == HAMMER2_OBJTYPE_HARDLINK) {
1284 hammer2_tid_t inum = ripdata->inum;
1285 error = hammer2_hardlink_find(dip, NULL, cluster);
1287 kprintf("hammer2: unable to find hardlink "
1288 "0x%016jx\n", inum);
1289 hammer2_cluster_unlock(cluster);
1297 * nresolve needs to resolve hardlinks, the original cluster is not
1301 ip = hammer2_inode_get(dip->pmp, dip, cluster);
1302 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
1303 if (ripdata->type == HAMMER2_OBJTYPE_HARDLINK) {
1304 kprintf("nresolve: fixup hardlink\n");
1305 hammer2_inode_ref(ip);
1306 hammer2_inode_unlock_ex(ip, NULL);
1307 hammer2_cluster_unlock(cluster);
1308 cluster = hammer2_inode_lock_ex(ip);
1309 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
1310 hammer2_inode_drop(ip);
1311 kprintf("nresolve: fixup to type %02x\n",
1320 * Deconsolidate any hardlink whos nlinks == 1. Ignore errors.
1321 * If an error occurs chain and ip are left alone.
1323 * XXX upgrade shared lock?
1325 if (ochain && chain &&
1326 chain->data->ipdata.nlinks == 1 && !dip->pmp->ronly) {
1327 kprintf("hammer2: need to unconsolidate hardlink for %s\n",
1328 chain->data->ipdata.filename);
1329 /* XXX retain shared lock on dip? (currently not held) */
1330 hammer2_trans_init(&trans, dip->pmp, 0);
1331 hammer2_hardlink_deconsolidate(&trans, dip, &chain, &ochain);
1332 hammer2_trans_done(&trans);
1337 * Acquire the related vnode
1339 * NOTE: For error processing, only ENOENT resolves the namecache
1340 * entry to NULL, otherwise we just return the error and
1341 * leave the namecache unresolved.
1343 * NOTE: multiple hammer2_inode structures can be aliased to the
1344 * same chain element, for example for hardlinks. This
1345 * use case does not 'reattach' inode associations that
1346 * might already exist, but always allocates a new one.
1348 * WARNING: inode structure is locked exclusively via inode_get
1349 * but chain was locked shared. inode_unlock_ex()
1350 * will handle it properly.
1353 vp = hammer2_igetv(ip, cluster, &error);
1356 cache_setvp(ap->a_nch, vp);
1357 } else if (error == ENOENT) {
1358 cache_setvp(ap->a_nch, NULL);
1360 hammer2_inode_unlock_ex(ip, cluster);
1363 * The vp should not be released until after we've disposed
1364 * of our locks, because it might cause vop_inactive() to
1371 cache_setvp(ap->a_nch, NULL);
1373 KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
1374 ("resolve error %d/%p ap %p\n",
1375 error, ap->a_nch->ncp->nc_vp, ap));
1382 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
1384 hammer2_inode_t *dip;
1385 hammer2_inode_t *ip;
1386 hammer2_cluster_t *cparent;
1390 dip = VTOI(ap->a_dvp);
1392 if ((ip = dip->pip) == NULL) {
1397 cparent = hammer2_inode_lock_ex(ip);
1398 *ap->a_vpp = hammer2_igetv(ip, cparent, &error);
1399 hammer2_inode_unlock_ex(ip, cparent);
1407 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
1409 hammer2_inode_t *dip;
1410 hammer2_inode_t *nip;
1411 hammer2_trans_t trans;
1412 hammer2_cluster_t *cluster;
1413 struct namecache *ncp;
1414 const uint8_t *name;
1419 dip = VTOI(ap->a_dvp);
1420 if (dip->pmp->ronly) {
1425 ncp = ap->a_nch->ncp;
1426 name = ncp->nc_name;
1427 name_len = ncp->nc_nlen;
1430 hammer2_pfs_memory_wait(dip->pmp);
1431 hammer2_trans_init(&trans, dip->pmp, HAMMER2_TRANS_NEWINODE);
1432 nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1434 &cluster, 0, &error);
1436 KKASSERT(nip == NULL);
1439 *ap->a_vpp = hammer2_igetv(nip, cluster, &error);
1440 hammer2_inode_unlock_ex(nip, cluster);
1442 hammer2_trans_done(&trans);
1445 cache_setunresolved(ap->a_nch);
1446 cache_setvp(ap->a_nch, *ap->a_vpp);
1453 * Return the largest contiguous physical disk range for the logical
1454 * request, in bytes.
1456 * (struct vnode *vp, off_t loffset, off_t *doffsetp, int *runp, int *runb)
1458 * Basically disabled, the logical buffer write thread has to deal with
1459 * buffers one-at-a-time.
1463 hammer2_vop_bmap(struct vop_bmap_args *ap)
1465 *ap->a_doffsetp = NOOFFSET;
1470 return (EOPNOTSUPP);
1475 hammer2_vop_open(struct vop_open_args *ap)
1477 return vop_stdopen(ap);
1481 * hammer2_vop_advlock { vp, id, op, fl, flags }
1485 hammer2_vop_advlock(struct vop_advlock_args *ap)
1487 hammer2_inode_t *ip = VTOI(ap->a_vp);
1488 const hammer2_inode_data_t *ripdata;
1489 hammer2_cluster_t *cparent;
1492 cparent = hammer2_inode_lock_sh(ip);
1493 ripdata = &hammer2_cluster_rdata(cparent)->ipdata;
1494 size = ripdata->size;
1495 hammer2_inode_unlock_sh(ip, cparent);
1496 return (lf_advlock(ap, &ip->advlock, size));
1502 hammer2_vop_close(struct vop_close_args *ap)
1504 return vop_stdclose(ap);
1508 * hammer2_vop_nlink { nch, dvp, vp, cred }
1510 * Create a hardlink from (vp) to {dvp, nch}.
1514 hammer2_vop_nlink(struct vop_nlink_args *ap)
1516 hammer2_inode_t *fdip; /* target directory to create link in */
1517 hammer2_inode_t *tdip; /* target directory to create link in */
1518 hammer2_inode_t *cdip; /* common parent directory */
1519 hammer2_inode_t *ip; /* inode we are hardlinking to */
1520 hammer2_cluster_t *cluster;
1521 hammer2_cluster_t *fdcluster;
1522 hammer2_cluster_t *tdcluster;
1523 hammer2_cluster_t *cdcluster;
1524 hammer2_trans_t trans;
1525 struct namecache *ncp;
1526 const uint8_t *name;
1531 tdip = VTOI(ap->a_dvp);
1532 if (tdip->pmp->ronly) {
1537 ncp = ap->a_nch->ncp;
1538 name = ncp->nc_name;
1539 name_len = ncp->nc_nlen;
1542 * ip represents the file being hardlinked. The file could be a
1543 * normal file or a hardlink target if it has already been hardlinked.
1544 * If ip is a hardlinked target then ip->pip represents the location
1545 * of the hardlinked target, NOT the location of the hardlink pointer.
1547 * Bump nlinks and potentially also create or move the hardlink
1548 * target in the parent directory common to (ip) and (tdip). The
1549 * consolidation code can modify ip->cluster and ip->pip. The
1550 * returned cluster is locked.
1552 ip = VTOI(ap->a_vp);
1553 hammer2_pfs_memory_wait(ip->pmp);
1554 hammer2_trans_init(&trans, ip->pmp, HAMMER2_TRANS_NEWINODE);
1557 * The common parent directory must be locked first to avoid deadlocks.
1558 * Also note that fdip and/or tdip might match cdip.
1561 cdip = hammer2_inode_common_parent(fdip, tdip);
1562 cdcluster = hammer2_inode_lock_ex(cdip);
1563 fdcluster = hammer2_inode_lock_ex(fdip);
1564 tdcluster = hammer2_inode_lock_ex(tdip);
1565 cluster = hammer2_inode_lock_ex(ip);
1566 error = hammer2_hardlink_consolidate(&trans, ip, &cluster,
1567 cdip, cdcluster, 1);
1572 * Create a directory entry connected to the specified cluster.
1574 * WARNING! chain can get moved by the connect (indirectly due to
1575 * potential indirect block creation).
1577 error = hammer2_inode_connect(&trans, &cluster, 1,
1581 cache_setunresolved(ap->a_nch);
1582 cache_setvp(ap->a_nch, ap->a_vp);
1585 hammer2_inode_unlock_ex(ip, cluster);
1586 hammer2_inode_unlock_ex(tdip, tdcluster);
1587 hammer2_inode_unlock_ex(fdip, fdcluster);
1588 hammer2_inode_unlock_ex(cdip, cdcluster);
1589 hammer2_inode_drop(cdip);
1590 hammer2_trans_done(&trans);
1597 * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1599 * The operating system has already ensured that the directory entry
1600 * does not exist and done all appropriate namespace locking.
1604 hammer2_vop_ncreate(struct vop_ncreate_args *ap)
1606 hammer2_inode_t *dip;
1607 hammer2_inode_t *nip;
1608 hammer2_trans_t trans;
1609 hammer2_cluster_t *ncluster;
1610 struct namecache *ncp;
1611 const uint8_t *name;
1616 dip = VTOI(ap->a_dvp);
1617 if (dip->pmp->ronly) {
1622 ncp = ap->a_nch->ncp;
1623 name = ncp->nc_name;
1624 name_len = ncp->nc_nlen;
1625 hammer2_pfs_memory_wait(dip->pmp);
1626 hammer2_trans_init(&trans, dip->pmp, HAMMER2_TRANS_NEWINODE);
1629 nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1631 &ncluster, 0, &error);
1633 KKASSERT(nip == NULL);
1636 *ap->a_vpp = hammer2_igetv(nip, ncluster, &error);
1637 hammer2_inode_unlock_ex(nip, ncluster);
1639 hammer2_trans_done(&trans);
1642 cache_setunresolved(ap->a_nch);
1643 cache_setvp(ap->a_nch, *ap->a_vpp);
1650 * Make a device node (typically a fifo)
1654 hammer2_vop_nmknod(struct vop_nmknod_args *ap)
1656 hammer2_inode_t *dip;
1657 hammer2_inode_t *nip;
1658 hammer2_trans_t trans;
1659 hammer2_cluster_t *ncluster;
1660 struct namecache *ncp;
1661 const uint8_t *name;
1666 dip = VTOI(ap->a_dvp);
1667 if (dip->pmp->ronly) {
1672 ncp = ap->a_nch->ncp;
1673 name = ncp->nc_name;
1674 name_len = ncp->nc_nlen;
1675 hammer2_pfs_memory_wait(dip->pmp);
1676 hammer2_trans_init(&trans, dip->pmp, HAMMER2_TRANS_NEWINODE);
1679 nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1681 &ncluster, 0, &error);
1683 KKASSERT(nip == NULL);
1686 *ap->a_vpp = hammer2_igetv(nip, ncluster, &error);
1687 hammer2_inode_unlock_ex(nip, ncluster);
1689 hammer2_trans_done(&trans);
1692 cache_setunresolved(ap->a_nch);
1693 cache_setvp(ap->a_nch, *ap->a_vpp);
1700 * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1704 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
1706 hammer2_inode_t *dip;
1707 hammer2_inode_t *nip;
1708 hammer2_cluster_t *ncparent;
1709 hammer2_trans_t trans;
1710 struct namecache *ncp;
1711 const uint8_t *name;
1715 dip = VTOI(ap->a_dvp);
1716 if (dip->pmp->ronly)
1719 ncp = ap->a_nch->ncp;
1720 name = ncp->nc_name;
1721 name_len = ncp->nc_nlen;
1722 hammer2_pfs_memory_wait(dip->pmp);
1723 hammer2_trans_init(&trans, dip->pmp, HAMMER2_TRANS_NEWINODE);
1726 ap->a_vap->va_type = VLNK; /* enforce type */
1728 nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1730 &ncparent, 0, &error);
1732 KKASSERT(nip == NULL);
1734 hammer2_trans_done(&trans);
1737 *ap->a_vpp = hammer2_igetv(nip, ncparent, &error);
1740 * Build the softlink (~like file data) and finalize the namecache.
1746 hammer2_inode_data_t *nipdata;
1748 nipdata = &hammer2_cluster_wdata(ncparent)->ipdata;
1749 /* nipdata = &nip->chain->data->ipdata;XXX */
1750 bytes = strlen(ap->a_target);
1752 if (bytes <= HAMMER2_EMBEDDED_BYTES) {
1753 KKASSERT(nipdata->op_flags &
1754 HAMMER2_OPFLAG_DIRECTDATA);
1755 bcopy(ap->a_target, nipdata->u.data, bytes);
1756 nipdata->size = bytes;
1758 hammer2_cluster_modsync(ncparent);
1759 hammer2_inode_unlock_ex(nip, ncparent);
1760 /* nipdata = NULL; not needed */
1762 hammer2_inode_unlock_ex(nip, ncparent);
1763 /* nipdata = NULL; not needed */
1764 bzero(&auio, sizeof(auio));
1765 bzero(&aiov, sizeof(aiov));
1766 auio.uio_iov = &aiov;
1767 auio.uio_segflg = UIO_SYSSPACE;
1768 auio.uio_rw = UIO_WRITE;
1769 auio.uio_resid = bytes;
1770 auio.uio_iovcnt = 1;
1771 auio.uio_td = curthread;
1772 aiov.iov_base = ap->a_target;
1773 aiov.iov_len = bytes;
1774 error = hammer2_write_file(nip, &auio, IO_APPEND, 0);
1775 /* XXX handle error */
1779 hammer2_inode_unlock_ex(nip, ncparent);
1781 hammer2_trans_done(&trans);
1784 * Finalize namecache
1787 cache_setunresolved(ap->a_nch);
1788 cache_setvp(ap->a_nch, *ap->a_vpp);
1789 /* hammer2_knote(ap->a_dvp, NOTE_WRITE); */
1795 * hammer2_vop_nremove { nch, dvp, cred }
1799 hammer2_vop_nremove(struct vop_nremove_args *ap)
1801 hammer2_inode_t *dip;
1802 hammer2_trans_t trans;
1803 struct namecache *ncp;
1804 const uint8_t *name;
1809 dip = VTOI(ap->a_dvp);
1810 if (dip->pmp->ronly) {
1815 ncp = ap->a_nch->ncp;
1816 name = ncp->nc_name;
1817 name_len = ncp->nc_nlen;
1819 hammer2_pfs_memory_wait(dip->pmp);
1820 hammer2_trans_init(&trans, dip->pmp, 0);
1821 error = hammer2_unlink_file(&trans, dip, name, name_len,
1822 0, NULL, ap->a_nch, -1);
1823 hammer2_run_unlinkq(&trans, dip->pmp);
1824 hammer2_trans_done(&trans);
1826 cache_unlink(ap->a_nch);
1832 * hammer2_vop_nrmdir { nch, dvp, cred }
1836 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
1838 hammer2_inode_t *dip;
1839 hammer2_trans_t trans;
1840 struct namecache *ncp;
1841 const uint8_t *name;
1846 dip = VTOI(ap->a_dvp);
1847 if (dip->pmp->ronly) {
1852 ncp = ap->a_nch->ncp;
1853 name = ncp->nc_name;
1854 name_len = ncp->nc_nlen;
1856 hammer2_pfs_memory_wait(dip->pmp);
1857 hammer2_trans_init(&trans, dip->pmp, 0);
1858 hammer2_run_unlinkq(&trans, dip->pmp);
1859 error = hammer2_unlink_file(&trans, dip, name, name_len,
1860 1, NULL, ap->a_nch, -1);
1861 hammer2_trans_done(&trans);
1863 cache_unlink(ap->a_nch);
1869 * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1873 hammer2_vop_nrename(struct vop_nrename_args *ap)
1875 struct namecache *fncp;
1876 struct namecache *tncp;
1877 hammer2_inode_t *cdip;
1878 hammer2_inode_t *fdip;
1879 hammer2_inode_t *tdip;
1880 hammer2_inode_t *ip;
1881 hammer2_cluster_t *cluster;
1882 hammer2_cluster_t *fdcluster;
1883 hammer2_cluster_t *tdcluster;
1884 hammer2_cluster_t *cdcluster;
1885 hammer2_trans_t trans;
1886 const uint8_t *fname;
1888 const uint8_t *tname;
1894 if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
1896 if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
1899 fdip = VTOI(ap->a_fdvp); /* source directory */
1900 tdip = VTOI(ap->a_tdvp); /* target directory */
1902 if (fdip->pmp->ronly)
1906 fncp = ap->a_fnch->ncp; /* entry name in source */
1907 fname = fncp->nc_name;
1908 fname_len = fncp->nc_nlen;
1910 tncp = ap->a_tnch->ncp; /* entry name in target */
1911 tname = tncp->nc_name;
1912 tname_len = tncp->nc_nlen;
1914 hammer2_pfs_memory_wait(tdip->pmp);
1915 hammer2_trans_init(&trans, tdip->pmp, 0);
1918 * ip is the inode being renamed. If this is a hardlink then
1919 * ip represents the actual file and not the hardlink marker.
1921 ip = VTOI(fncp->nc_vp);
1926 * The common parent directory must be locked first to avoid deadlocks.
1927 * Also note that fdip and/or tdip might match cdip.
1929 * WARNING! fdip may not match ip->pip. That is, if the source file
1930 * is already a hardlink then what we are renaming is the
1931 * hardlink pointer, not the hardlink itself. The hardlink
1932 * directory (ip->pip) will already be at a common parent
1935 * Be sure to use ip->pip when finding the common parent
1936 * against tdip or we might accidently move the hardlink
1937 * target into a subdirectory that makes it inaccessible to
1940 cdip = hammer2_inode_common_parent(ip->pip, tdip);
1941 cdcluster = hammer2_inode_lock_ex(cdip);
1942 fdcluster = hammer2_inode_lock_ex(fdip);
1943 tdcluster = hammer2_inode_lock_ex(tdip);
1946 * Keep a tight grip on the inode so the temporary unlinking from
1947 * the source location prior to linking to the target location
1948 * does not cause the cluster to be destroyed.
1950 * NOTE: To avoid deadlocks we cannot lock (ip) while we are
1951 * unlinking elements from their directories. Locking
1952 * the nlinks field does not lock the whole inode.
1954 hammer2_inode_ref(ip);
1957 * Remove target if it exists.
1959 error = hammer2_unlink_file(&trans, tdip, tname, tname_len,
1960 -1, NULL, ap->a_tnch, -1);
1962 if (error && error != ENOENT)
1966 * When renaming a hardlinked file we may have to re-consolidate
1967 * the location of the hardlink target.
1969 * If ip represents a regular file the consolidation code essentially
1970 * does nothing other than return the same locked cluster that was
1973 * The returned cluster will be locked.
1975 * WARNING! We do not currently have a local copy of ipdata but
1976 * we do use one later remember that it must be reloaded
1977 * on any modification to the inode, including connects.
1979 cluster = hammer2_inode_lock_ex(ip);
1980 error = hammer2_hardlink_consolidate(&trans, ip, &cluster,
1981 cdip, cdcluster, 0);
1986 * Disconnect (fdip, fname) from the source directory. This will
1987 * disconnect (ip) if it represents a direct file. If (ip) represents
1988 * a hardlink the HARDLINK pointer object will be removed but the
1989 * hardlink will stay intact.
1991 * Always pass nch as NULL because we intend to reconnect the inode,
1992 * so we don't want hammer2_unlink_file() to rename it to the hidden
1993 * open-but-unlinked directory.
1995 * The target cluster may be marked DELETED but will not be destroyed
1996 * since we retain our hold on ip and cluster.
1998 * NOTE: We pass nlinks as 0 (not -1) in order to retain the file's
2001 error = hammer2_unlink_file(&trans, fdip, fname, fname_len,
2002 -1, &hlink, NULL, 0);
2003 KKASSERT(error != EAGAIN);
2008 * Reconnect ip to target directory using cluster. Chains cannot
2009 * actually be moved, so this will duplicate the cluster in the new
2010 * spot and assign it to the ip, replacing the old cluster.
2012 * WARNING: Because recursive locks are allowed and we unlinked the
2013 * file that we have a cluster-in-hand for just above, the
2014 * cluster might have been delete-duplicated. We must
2015 * refactor the cluster.
2017 * WARNING: Chain locks can lock buffer cache buffers, to avoid
2018 * deadlocks we want to unlock before issuing a cache_*()
2019 * op (that might have to lock a vnode).
2021 * NOTE: Pass nlinks as 0 because we retained the link count from
2022 * the unlink, so we do not have to modify it.
2024 error = hammer2_inode_connect(&trans, &cluster, hlink,
2026 tname, tname_len, 0);
2028 KKASSERT(cluster != NULL);
2029 hammer2_inode_repoint(ip, (hlink ? ip->pip : tdip), cluster);
2032 hammer2_inode_unlock_ex(ip, cluster);
2033 hammer2_inode_unlock_ex(tdip, tdcluster);
2034 hammer2_inode_unlock_ex(fdip, fdcluster);
2035 hammer2_inode_unlock_ex(cdip, cdcluster);
2036 hammer2_inode_drop(ip);
2037 hammer2_inode_drop(cdip);
2038 hammer2_run_unlinkq(&trans, fdip->pmp);
2039 hammer2_trans_done(&trans);
2042 * Issue the namecache update after unlocking all the internal
2043 * hammer structures, otherwise we might deadlock.
2045 if (tnch_error == 0) {
2046 cache_unlink(ap->a_tnch);
2047 cache_setunresolved(ap->a_tnch);
2050 cache_rename(ap->a_fnch, ap->a_tnch);
2057 * Strategy code (async logical file buffer I/O from system)
2059 * WARNING: The strategy code cannot safely use hammer2 transactions
2060 * as this can deadlock against vfs_sync's vfsync() call
2061 * if multiple flushes are queued. All H2 structures must
2062 * already be present and ready for the DIO.
2064 * Reads can be initiated asynchronously, writes have to be
2065 * spooled to a separate thread for action to avoid deadlocks.
2067 static int hammer2_strategy_read(struct vop_strategy_args *ap);
2068 static int hammer2_strategy_write(struct vop_strategy_args *ap);
2069 static void hammer2_strategy_read_callback(hammer2_iocb_t *iocb);
2073 hammer2_vop_strategy(struct vop_strategy_args *ap)
2084 error = hammer2_strategy_read(ap);
2085 ++hammer2_iod_file_read;
2088 error = hammer2_strategy_write(ap);
2089 ++hammer2_iod_file_write;
2092 bp->b_error = error = EINVAL;
2093 bp->b_flags |= B_ERROR;
2101 * Logical buffer I/O, async read.
2105 hammer2_strategy_read(struct vop_strategy_args *ap)
2110 hammer2_inode_t *ip;
2111 hammer2_cluster_t *cparent;
2112 hammer2_cluster_t *cluster;
2113 hammer2_key_t key_dummy;
2114 hammer2_key_t lbase;
2120 ip = VTOI(ap->a_vp);
2121 nbio = push_bio(bio);
2123 lbase = bio->bio_offset;
2124 KKASSERT(((int)lbase & HAMMER2_PBUFMASK) == 0);
2127 * Lookup the file offset.
2129 cparent = hammer2_inode_lock_sh(ip);
2130 cluster = hammer2_cluster_lookup(cparent, &key_dummy,
2132 HAMMER2_LOOKUP_NODATA |
2133 HAMMER2_LOOKUP_SHARED,
2135 hammer2_inode_unlock_sh(ip, cparent);
2138 * Data is zero-fill if no cluster could be found
2139 * (XXX or EIO on a cluster failure).
2141 if (cluster == NULL) {
2144 bzero(bp->b_data, bp->b_bcount);
2150 * Cluster elements must be type INODE or type DATA, but the
2151 * compression mode (or not) for DATA chains can be different for
2152 * each chain. This will be handled by the callback.
2154 * If the cluster already has valid data the callback will be made
2155 * immediately/synchronously.
2157 btype = hammer2_cluster_type(cluster);
2158 if (btype != HAMMER2_BREF_TYPE_INODE &&
2159 btype != HAMMER2_BREF_TYPE_DATA) {
2160 panic("READ PATH: hammer2_strategy_read: unknown bref type");
2162 hammer2_cluster_load_async(cluster, hammer2_strategy_read_callback,
2168 * Read callback for hammer2_cluster_load_async(). The load function may
2169 * start several actual I/Os but will only make one callback, typically with
2170 * the first valid I/O XXX
2174 hammer2_strategy_read_callback(hammer2_iocb_t *iocb)
2176 struct bio *bio = iocb->ptr; /* original logical buffer */
2177 struct buf *bp = bio->bio_buf; /* original logical buffer */
2178 hammer2_chain_t *chain;
2179 hammer2_cluster_t *cluster;
2185 * Extract data and handle iteration on I/O failure. iocb->off
2186 * is the cluster index for iteration.
2188 cluster = iocb->cluster;
2189 dio = iocb->dio; /* can be NULL if iocb not in progress */
2192 * Work to do if INPROG set, else dio is already good or dio is
2193 * NULL (which is the shortcut case if chain->data is already good).
2195 if (iocb->flags & HAMMER2_IOCB_INPROG) {
2197 * Read attempt not yet made. Issue an asynchronous read
2198 * if necessary and return, operation will chain back to
2201 if ((iocb->flags & HAMMER2_IOCB_READ) == 0) {
2202 if (dio->bp == NULL ||
2203 (dio->bp->b_flags & B_CACHE) == 0) {
2208 iocb->flags |= HAMMER2_IOCB_READ;
2209 breadcb(dio->hmp->devvp,
2210 dio->pbase, dio->psize,
2211 hammer2_io_callback, iocb);
2218 * If we have a DIO it is now done, check for an error and
2219 * calculate the data.
2221 * If there is no DIO it is an optimization by
2222 * hammer2_cluster_load_async(), the data is available in
2226 if (dio->bp->b_flags & B_ERROR) {
2227 i = (int)iocb->lbase + 1;
2228 if (i >= cluster->nchains) {
2229 bp->b_flags |= B_ERROR;
2230 bp->b_error = dio->bp->b_error;
2231 hammer2_io_complete(iocb);
2233 hammer2_cluster_unlock(cluster);
2235 hammer2_io_complete(iocb); /* XXX */
2236 chain = cluster->array[i].chain;
2237 kprintf("hammer2: IO CHAIN-%d %p\n", i, chain);
2238 hammer2_adjreadcounter(&chain->bref,
2240 iocb->chain = chain;
2241 iocb->lbase = (off_t)i;
2244 hammer2_io_getblk(chain->hmp,
2245 chain->bref.data_off,
2251 chain = iocb->chain;
2252 data = hammer2_io_data(dio, chain->bref.data_off);
2255 * Special synchronous case, data present in chain->data.
2257 chain = iocb->chain;
2258 data = (void *)chain->data;
2261 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) {
2263 * Data is embedded in the inode (copy from inode).
2265 bcopy(((hammer2_inode_data_t *)data)->u.data,
2266 bp->b_data, HAMMER2_EMBEDDED_BYTES);
2267 bzero(bp->b_data + HAMMER2_EMBEDDED_BYTES,
2268 bp->b_bcount - HAMMER2_EMBEDDED_BYTES);
2271 } else if (chain->bref.type == HAMMER2_BREF_TYPE_DATA) {
2273 * Data is on-media, issue device I/O and copy.
2275 * XXX direct-IO shortcut could go here XXX.
2277 switch (HAMMER2_DEC_COMP(chain->bref.methods)) {
2278 case HAMMER2_COMP_LZ4:
2279 hammer2_decompress_LZ4_callback(data, chain->bytes,
2282 case HAMMER2_COMP_ZLIB:
2283 hammer2_decompress_ZLIB_callback(data, chain->bytes,
2286 case HAMMER2_COMP_NONE:
2287 KKASSERT(chain->bytes <= bp->b_bcount);
2288 bcopy(data, bp->b_data, chain->bytes);
2289 if (chain->bytes < bp->b_bcount) {
2290 bzero(bp->b_data + chain->bytes,
2291 bp->b_bcount - chain->bytes);
2293 bp->b_flags |= B_NOTMETA;
2298 panic("hammer2_strategy_read: "
2299 "unknown compression type");
2302 /* bqrelse the dio to help stabilize the call to panic() */
2304 hammer2_io_bqrelse(&dio);
2305 panic("hammer2_strategy_read: unknown bref type");
2309 * Once the iocb is cleaned up the DIO (if any) will no longer be
2310 * in-progress but will still have a ref. Be sure to release
2313 hammer2_io_complete(iocb); /* physical management */
2314 if (dio) /* physical dio & buffer */
2315 hammer2_io_bqrelse(&dio);
2316 hammer2_cluster_unlock(cluster); /* cluster management */
2317 biodone(bio); /* logical buffer */
2322 hammer2_strategy_write(struct vop_strategy_args *ap)
2327 hammer2_inode_t *ip;
2331 ip = VTOI(ap->a_vp);
2334 hammer2_lwinprog_ref(pmp);
2335 hammer2_mtx_ex(&pmp->wthread_mtx);
2336 if (TAILQ_EMPTY(&pmp->wthread_bioq.queue)) {
2337 bioq_insert_tail(&pmp->wthread_bioq, ap->a_bio);
2338 hammer2_mtx_unlock(&pmp->wthread_mtx);
2339 wakeup(&pmp->wthread_bioq);
2341 bioq_insert_tail(&pmp->wthread_bioq, ap->a_bio);
2342 hammer2_mtx_unlock(&pmp->wthread_mtx);
2344 hammer2_lwinprog_wait(pmp);
2350 * hammer2_vop_ioctl { vp, command, data, fflag, cred }
2354 hammer2_vop_ioctl(struct vop_ioctl_args *ap)
2356 hammer2_inode_t *ip;
2360 ip = VTOI(ap->a_vp);
2362 error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
2363 ap->a_fflag, ap->a_cred);
2370 hammer2_vop_mountctl(struct vop_mountctl_args *ap)
2378 case (MOUNTCTL_SET_EXPORT):
2379 mp = ap->a_head.a_ops->head.vv_mount;
2382 if (ap->a_ctllen != sizeof(struct export_args))
2385 rc = vfs_export(mp, &pmp->export,
2386 (const struct export_args *)ap->a_ctl);
2389 rc = vop_stdmountctl(ap);
2397 * This handles unlinked open files after the vnode is finally dereferenced.
2398 * To avoid deadlocks it cannot be called from the normal vnode recycling
2399 * path, so we call it (1) after a unlink, rmdir, or rename, (2) on every
2400 * flush, and (3) on umount.
2403 hammer2_run_unlinkq(hammer2_trans_t *trans, hammer2_pfs_t *pmp)
2405 const hammer2_inode_data_t *ripdata;
2406 hammer2_inode_unlink_t *ipul;
2407 hammer2_inode_t *ip;
2408 hammer2_cluster_t *cluster;
2409 hammer2_cluster_t *cparent;
2411 if (TAILQ_EMPTY(&pmp->unlinkq))
2415 hammer2_spin_ex(&pmp->list_spin);
2416 while ((ipul = TAILQ_FIRST(&pmp->unlinkq)) != NULL) {
2417 TAILQ_REMOVE(&pmp->unlinkq, ipul, entry);
2418 hammer2_spin_unex(&pmp->list_spin);
2420 kfree(ipul, pmp->minode);
2422 cluster = hammer2_inode_lock_ex(ip);
2423 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
2424 if (hammer2_debug & 0x400) {
2425 kprintf("hammer2: unlink on reclaim: %s refs=%d\n",
2426 ripdata->filename, ip->refs);
2428 KKASSERT(ripdata->nlinks == 0);
2430 cparent = hammer2_cluster_parent(cluster);
2431 hammer2_cluster_delete(trans, cparent, cluster,
2432 HAMMER2_DELETE_PERMANENT);
2433 hammer2_cluster_unlock(cparent);
2434 hammer2_inode_unlock_ex(ip, cluster); /* inode lock */
2435 hammer2_inode_drop(ip); /* ipul ref */
2437 hammer2_spin_ex(&pmp->list_spin);
2439 hammer2_spin_unex(&pmp->list_spin);
2447 static void filt_hammer2detach(struct knote *kn);
2448 static int filt_hammer2read(struct knote *kn, long hint);
2449 static int filt_hammer2write(struct knote *kn, long hint);
2450 static int filt_hammer2vnode(struct knote *kn, long hint);
2452 static struct filterops hammer2read_filtops =
2453 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2454 NULL, filt_hammer2detach, filt_hammer2read };
2455 static struct filterops hammer2write_filtops =
2456 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2457 NULL, filt_hammer2detach, filt_hammer2write };
2458 static struct filterops hammer2vnode_filtops =
2459 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2460 NULL, filt_hammer2detach, filt_hammer2vnode };
2464 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap)
2466 struct vnode *vp = ap->a_vp;
2467 struct knote *kn = ap->a_kn;
2469 switch (kn->kn_filter) {
2471 kn->kn_fop = &hammer2read_filtops;
2474 kn->kn_fop = &hammer2write_filtops;
2477 kn->kn_fop = &hammer2vnode_filtops;
2480 return (EOPNOTSUPP);
2483 kn->kn_hook = (caddr_t)vp;
2485 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2491 filt_hammer2detach(struct knote *kn)
2493 struct vnode *vp = (void *)kn->kn_hook;
2495 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2499 filt_hammer2read(struct knote *kn, long hint)
2501 struct vnode *vp = (void *)kn->kn_hook;
2502 hammer2_inode_t *ip = VTOI(vp);
2505 if (hint == NOTE_REVOKE) {
2506 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2509 off = ip->size - kn->kn_fp->f_offset;
2510 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2511 if (kn->kn_sfflags & NOTE_OLDAPI)
2513 return (kn->kn_data != 0);
2518 filt_hammer2write(struct knote *kn, long hint)
2520 if (hint == NOTE_REVOKE)
2521 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2527 filt_hammer2vnode(struct knote *kn, long hint)
2529 if (kn->kn_sfflags & hint)
2530 kn->kn_fflags |= hint;
2531 if (hint == NOTE_REVOKE) {
2532 kn->kn_flags |= (EV_EOF | EV_NODATA);
2535 return (kn->kn_fflags != 0);
2543 hammer2_vop_markatime(struct vop_markatime_args *ap)
2545 hammer2_inode_t *ip;
2558 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap)
2562 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2564 error = hammer2_vop_kqfilter(ap);
2571 struct vop_ops hammer2_vnode_vops = {
2572 .vop_default = vop_defaultop,
2573 .vop_fsync = hammer2_vop_fsync,
2574 .vop_getpages = vop_stdgetpages,
2575 .vop_putpages = vop_stdputpages,
2576 .vop_access = hammer2_vop_access,
2577 .vop_advlock = hammer2_vop_advlock,
2578 .vop_close = hammer2_vop_close,
2579 .vop_nlink = hammer2_vop_nlink,
2580 .vop_ncreate = hammer2_vop_ncreate,
2581 .vop_nsymlink = hammer2_vop_nsymlink,
2582 .vop_nremove = hammer2_vop_nremove,
2583 .vop_nrmdir = hammer2_vop_nrmdir,
2584 .vop_nrename = hammer2_vop_nrename,
2585 .vop_getattr = hammer2_vop_getattr,
2586 .vop_setattr = hammer2_vop_setattr,
2587 .vop_readdir = hammer2_vop_readdir,
2588 .vop_readlink = hammer2_vop_readlink,
2589 .vop_getpages = vop_stdgetpages,
2590 .vop_putpages = vop_stdputpages,
2591 .vop_read = hammer2_vop_read,
2592 .vop_write = hammer2_vop_write,
2593 .vop_open = hammer2_vop_open,
2594 .vop_inactive = hammer2_vop_inactive,
2595 .vop_reclaim = hammer2_vop_reclaim,
2596 .vop_nresolve = hammer2_vop_nresolve,
2597 .vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
2598 .vop_nmkdir = hammer2_vop_nmkdir,
2599 .vop_nmknod = hammer2_vop_nmknod,
2600 .vop_ioctl = hammer2_vop_ioctl,
2601 .vop_mountctl = hammer2_vop_mountctl,
2602 .vop_bmap = hammer2_vop_bmap,
2603 .vop_strategy = hammer2_vop_strategy,
2604 .vop_kqfilter = hammer2_vop_kqfilter
2607 struct vop_ops hammer2_spec_vops = {
2608 .vop_default = vop_defaultop,
2609 .vop_fsync = hammer2_vop_fsync,
2610 .vop_read = vop_stdnoread,
2611 .vop_write = vop_stdnowrite,
2612 .vop_access = hammer2_vop_access,
2613 .vop_close = hammer2_vop_close,
2614 .vop_markatime = hammer2_vop_markatime,
2615 .vop_getattr = hammer2_vop_getattr,
2616 .vop_inactive = hammer2_vop_inactive,
2617 .vop_reclaim = hammer2_vop_reclaim,
2618 .vop_setattr = hammer2_vop_setattr
2621 struct vop_ops hammer2_fifo_vops = {
2622 .vop_default = fifo_vnoperate,
2623 .vop_fsync = hammer2_vop_fsync,
2625 .vop_read = hammer2_vop_fiforead,
2626 .vop_write = hammer2_vop_fifowrite,
2628 .vop_access = hammer2_vop_access,
2630 .vop_close = hammer2_vop_fifoclose,
2632 .vop_markatime = hammer2_vop_markatime,
2633 .vop_getattr = hammer2_vop_getattr,
2634 .vop_inactive = hammer2_vop_inactive,
2635 .vop_reclaim = hammer2_vop_reclaim,
2636 .vop_setattr = hammer2_vop_setattr,
2637 .vop_kqfilter = hammer2_vop_fifokqfilter