/*
* Acquire synchronization TID
*/
static
int
hammer_ioc_synctid(hammer_transaction_t trans, hammer_inode_t ip,
struct hammer_ioc_synctid *std)
{
hammer_mount_t hmp = ip->hmp;
int error = 0;
switch(std->op) {
case HAMMER_SYNCTID_NONE:
std->tid = hmp->flusher.tid; /* inaccurate */
break;
case HAMMER_SYNCTID_ASYNC:
hammer_queue_inodes_flusher(hmp, MNT_NOWAIT);
hammer_flusher_async(hmp, NULL);
std->tid = hmp->flusher.tid; /* inaccurate */
break;
case HAMMER_SYNCTID_SYNC1:
hammer_queue_inodes_flusher(hmp, MNT_WAIT);
hammer_flusher_sync(hmp);
std->tid = hmp->flusher.tid;
break;
case HAMMER_SYNCTID_SYNC2:
hammer_queue_inodes_flusher(hmp, MNT_WAIT);
hammer_flusher_sync(hmp);
std->tid = hmp->flusher.tid;
hammer_flusher_sync(hmp);
break;
default:
error = EOPNOTSUPP;
break;
}
return(error);
}
/*
* Clean up the internal mount structure and disassociate it from the mount.
* This may issue I/O.
*
* Called with fs_token held.
*/
static void
hammer_free_hmp(struct mount *mp)
{
hammer_mount_t hmp = (void *)mp->mnt_data;
hammer_flush_group_t flg;
int count;
int dummy;
/*
* Flush anything dirty. This won't even run if the
* filesystem errored-out.
*/
count = 0;
while (hammer_flusher_haswork(hmp)) {
hammer_flusher_sync(hmp);
++count;
if (count >= 5) {
if (count == 5)
kprintf("HAMMER: umount flushing.");
else
kprintf(".");
tsleep(&dummy, 0, "hmrufl", hz);
}
if (count == 30) {
kprintf("giving up\n");
break;
}
}
if (count >= 5 && count < 30)
kprintf("\n");
/*
* If the mount had a critical error we have to destroy any
* remaining inodes before we can finish cleaning up the flusher.
*/
if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
hammer_destroy_inode_callback, NULL);
}
/*
* There shouldn't be any inodes left now and any left over
* flush groups should now be empty.
*/
KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
KKASSERT(RB_EMPTY(&flg->flush_tree));
if (flg->refs) {
kprintf("HAMMER: Warning, flush_group %p was "
"not empty on umount!\n", flg);
}
kfree(flg, hmp->m_misc);
}
/*
* We can finally destroy the flusher
*/
hammer_flusher_destroy(hmp);
/*
* We may have held recovered buffers due to a read-only mount.
* These must be discarded.
*/
if (hmp->ronly)
hammer_recover_flush_buffers(hmp, NULL, -1);
/*
* Unload buffers and then volumes
*/
RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
hammer_unload_buffer, NULL);
RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
hammer_unload_volume, NULL);
mp->mnt_data = NULL;
mp->mnt_flag &= ~MNT_LOCAL;
hmp->mp = NULL;
hammer_destroy_objid_cache(hmp);
hammer_destroy_dedup_cache(hmp);
if (hmp->dedup_free_cache != NULL) {
kfree(hmp->dedup_free_cache, hmp->m_misc);
hmp->dedup_free_cache = NULL;
}
kmalloc_destroy(&hmp->m_misc);
kmalloc_destroy(&hmp->m_inodes);
lwkt_reltoken(&hmp->fs_token);
kfree(hmp, M_HAMMER);
}
static int
hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
struct ucred *cred)
{
struct hammer_mount_info info;
hammer_mount_t hmp;
hammer_volume_t rootvol;
struct vnode *rootvp;
struct vnode *devvp = NULL;
const char *upath; /* volume name in userspace */
char *path; /* volume name in system space */
int error;
int i;
int master_id;
char *next_volume_ptr = NULL;
/*
* Accept hammer_mount_info. mntpt is NULL for root mounts at boot.
*/
if (mntpt == NULL) {
bzero(&info, sizeof(info));
info.asof = 0;
info.hflags = 0;
info.nvolumes = 1;
next_volume_ptr = mp->mnt_stat.f_mntfromname;
/* Count number of volumes separated by ':' */
for (char *p = next_volume_ptr; *p != '\0'; ++p) {
if (*p == ':') {
++info.nvolumes;
}
}
mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
} else {
if ((error = copyin(data, &info, sizeof(info))) != 0)
return (error);
}
/*
* updating or new mount
*/
if (mp->mnt_flag & MNT_UPDATE) {
hmp = (void *)mp->mnt_data;
KKASSERT(hmp != NULL);
} else {
if (info.nvolumes <= 0 || info.nvolumes > HAMMER_MAX_VOLUMES)
return (EINVAL);
hmp = NULL;
}
/*
* master-id validation. The master id may not be changed by a
* mount update.
*/
if (info.hflags & HMNT_MASTERID) {
if (hmp && hmp->master_id != info.master_id) {
kprintf("hammer: cannot change master id "
"with mount update\n");
return(EINVAL);
}
master_id = info.master_id;
if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
return (EINVAL);
} else {
if (hmp)
master_id = hmp->master_id;
else
master_id = 0;
}
/*
* Internal mount data structure
*/
if (hmp == NULL) {
hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
mp->mnt_data = (qaddr_t)hmp;
hmp->mp = mp;
/*TAILQ_INIT(&hmp->recycle_list);*/
/*
* Make sure kmalloc type limits are set appropriately.
*
* Our inode kmalloc group is sized based on maxvnodes
* (controlled by the system, not us).
*/
kmalloc_create(&hmp->m_misc, "HAMMER-others");
kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
kmalloc_raise_limit(hmp->m_inodes, 0); /* unlimited */
hmp->root_btree_beg.localization = 0x00000000U;
hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
hmp->root_btree_beg.key = -0x8000000000000000LL;
hmp->root_btree_beg.create_tid = 1;
hmp->root_btree_beg.delete_tid = 1;
hmp->root_btree_beg.rec_type = 0;
hmp->root_btree_beg.obj_type = 0;
hmp->root_btree_end.localization = 0xFFFFFFFFU;
hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
hmp->root_btree_end.delete_tid = 0; /* special case */
hmp->root_btree_end.rec_type = 0xFFFFU;
hmp->root_btree_end.obj_type = 0;
hmp->krate.freq = 1; /* maximum reporting rate (hz) */
hmp->krate.count = -16; /* initial burst */
hmp->sync_lock.refs = 1;
hmp->free_lock.refs = 1;
hmp->undo_lock.refs = 1;
hmp->blkmap_lock.refs = 1;
hmp->snapshot_lock.refs = 1;
hmp->volume_lock.refs = 1;
TAILQ_INIT(&hmp->delay_list);
TAILQ_INIT(&hmp->flush_group_list);
TAILQ_INIT(&hmp->objid_cache_list);
TAILQ_INIT(&hmp->undo_lru_list);
TAILQ_INIT(&hmp->reclaim_list);
RB_INIT(&hmp->rb_dedup_crc_root);
RB_INIT(&hmp->rb_dedup_off_root);
TAILQ_INIT(&hmp->dedup_lru_list);
}
hmp->hflags &= ~HMNT_USERFLAGS;
hmp->hflags |= info.hflags & HMNT_USERFLAGS;
hmp->master_id = master_id;
if (info.asof) {
mp->mnt_flag |= MNT_RDONLY;
hmp->asof = info.asof;
} else {
hmp->asof = HAMMER_MAX_TID;
}
hmp->volume_to_remove = -1;
/*
* Re-open read-write if originally read-only, or vise-versa.
*
* When going from read-only to read-write execute the stage2
* recovery if it has not already been run.
*/
if (mp->mnt_flag & MNT_UPDATE) {
lwkt_gettoken(&hmp->fs_token);
error = 0;
if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
kprintf("HAMMER read-only -> read-write\n");
hmp->ronly = 0;
RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
hammer_adjust_volume_mode, NULL);
rootvol = hammer_get_root_volume(hmp, &error);
if (rootvol) {
hammer_recover_flush_buffers(hmp, rootvol, 1);
error = hammer_recover_stage2(hmp, rootvol);
bcopy(rootvol->ondisk->vol0_blockmap,
hmp->blockmap,
sizeof(hmp->blockmap));
hammer_rel_volume(rootvol, 0);
}
RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
hammer_reload_inode, NULL);
/* kernel clears MNT_RDONLY */
} else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
kprintf("HAMMER read-write -> read-only\n");
hmp->ronly = 1; /* messy */
RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
hammer_reload_inode, NULL);
hmp->ronly = 0;
hammer_flusher_sync(hmp);
hammer_flusher_sync(hmp);
hammer_flusher_sync(hmp);
hmp->ronly = 1;
RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
hammer_adjust_volume_mode, NULL);
}
lwkt_reltoken(&hmp->fs_token);
return(error);
}
RB_INIT(&hmp->rb_vols_root);
RB_INIT(&hmp->rb_inos_root);
RB_INIT(&hmp->rb_redo_root);
RB_INIT(&hmp->rb_nods_root);
RB_INIT(&hmp->rb_undo_root);
RB_INIT(&hmp->rb_resv_root);
RB_INIT(&hmp->rb_bufs_root);
RB_INIT(&hmp->rb_pfsm_root);
hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
RB_INIT(&hmp->volu_root);
RB_INIT(&hmp->undo_root);
RB_INIT(&hmp->data_root);
RB_INIT(&hmp->meta_root);
RB_INIT(&hmp->lose_root);
TAILQ_INIT(&hmp->iorun_list);
lwkt_token_init(&hmp->fs_token, "hammerfs");
lwkt_token_init(&hmp->io_token, "hammerio");
lwkt_gettoken(&hmp->fs_token);
/*
* Load volumes
*/
path = objcache_get(namei_oc, M_WAITOK);
hmp->nvolumes = -1;
for (i = 0; i < info.nvolumes; ++i) {
if (mntpt == NULL) {
/*
* Root mount.
*/
KKASSERT(next_volume_ptr != NULL);
strcpy(path, "");
if (*next_volume_ptr != '/') {
/* relative path */
strcpy(path, "/dev/");
}
int k;
for (k = strlen(path); k < MAXPATHLEN-1; ++k) {
if (*next_volume_ptr == '\0') {
break;
} else if (*next_volume_ptr == ':') {
++next_volume_ptr;
break;
} else {
path[k] = *next_volume_ptr;
++next_volume_ptr;
}
}
path[k] = '\0';
error = 0;
cdev_t dev = kgetdiskbyname(path);
error = bdevvp(dev, &devvp);
if (error) {
kprintf("hammer_mountroot: can't find devvp\n");
}
} else {
error = copyin(&info.volumes[i], &upath,
sizeof(char *));
if (error == 0)
error = copyinstr(upath, path,
MAXPATHLEN, NULL);
}
if (error == 0)
error = hammer_install_volume(hmp, path, devvp);
if (error)
break;
}
objcache_put(namei_oc, path);
/*
* Make sure we found a root volume
*/
if (error == 0 && hmp->rootvol == NULL) {
kprintf("hammer_mount: No root volume found!\n");
error = EINVAL;
}
/*
* Check that all required volumes are available
*/
if (error == 0 && hammer_mountcheck_volumes(hmp)) {
kprintf("hammer_mount: Missing volumes, cannot mount!\n");
error = EINVAL;
}
if (error) {
/* called with fs_token held */
hammer_free_hmp(mp);
return (error);
}
/*
* No errors, setup enough of the mount point so we can lookup the
* root vnode.
*/
mp->mnt_iosize_max = MAXPHYS;
mp->mnt_kern_flag |= MNTK_FSMID;
mp->mnt_kern_flag |= MNTK_THR_SYNC; /* new vsyncscan semantics */
/*
* MPSAFE code. Note that VOPs and VFSops which are not MPSAFE
* will acquire a per-mount token prior to entry and release it
* on return, so even if we do not specify it we no longer get
* the BGL regardlless of how we are flagged.
*/
mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
/*MNTK_RD_MPSAFE | MNTK_GA_MPSAFE | MNTK_IN_MPSAFE;*/
/*
* note: f_iosize is used by vnode_pager_haspage() when constructing
* its VOP_BMAP call.
*/
mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
mp->mnt_maxsymlinklen = 255;
mp->mnt_flag |= MNT_LOCAL;
vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
/*
* The root volume's ondisk pointer is only valid if we hold a
* reference to it.
*/
rootvol = hammer_get_root_volume(hmp, &error);
if (error)
goto failed;
/*
* Perform any necessary UNDO operations. The recovery code does
* call hammer_undo_lookup() so we have to pre-cache the blockmap,
* and then re-copy it again after recovery is complete.
*
* If this is a read-only mount the UNDO information is retained
* in memory in the form of dirty buffer cache buffers, and not
* written back to the media.
*/
bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
sizeof(hmp->blockmap));
/*
* Check filesystem version
*/
hmp->version = rootvol->ondisk->vol_version;
if (hmp->version < HAMMER_VOL_VERSION_MIN ||
hmp->version > HAMMER_VOL_VERSION_MAX) {
kprintf("HAMMER: mount unsupported fs version %d\n",
hmp->version);
error = ERANGE;
goto done;
}
/*
* The undo_rec_limit limits the size of flush groups to avoid
* blowing out the UNDO FIFO. This calculation is typically in
* the tens of thousands and is designed primarily when small
* HAMMER filesystems are created.
*/
hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
if (hammer_debug_general & 0x0001)
kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
/*
* NOTE: Recover stage1 not only handles meta-data recovery, it
* also sets hmp->undo_seqno for HAMMER VERSION 4+ filesystems.
*/
error = hammer_recover_stage1(hmp, rootvol);
if (error) {
kprintf("Failed to recover HAMMER filesystem on mount\n");
goto done;
}
/*
* Finish setup now that we have a good root volume.
*
* The top 16 bits of fsid.val[1] is a pfs id.
*/
ksnprintf(mp->mnt_stat.f_mntfromname,
sizeof(mp->mnt_stat.f_mntfromname), "%s",
rootvol->ondisk->vol_name);
mp->mnt_stat.f_fsid.val[0] =
crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
mp->mnt_stat.f_fsid.val[1] =
crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
sizeof(mp->mnt_vstat.f_fsid_uuid));
/*
* Certain often-modified fields in the root volume are cached in
* the hammer_mount structure so we do not have to generate lots
* of little UNDO structures for them.
*
* Recopy after recovery. This also has the side effect of
* setting our cached undo FIFO's first_offset, which serves to
* placemark the FIFO start for the NEXT flush cycle while the
* on-disk first_offset represents the LAST flush cycle.
*/
hmp->next_tid = rootvol->ondisk->vol0_next_tid;
hmp->flush_tid1 = hmp->next_tid;
hmp->flush_tid2 = hmp->next_tid;
bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
sizeof(hmp->blockmap));
hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
hammer_flusher_create(hmp);
/*
* Locate the root directory using the root cluster's B-Tree as a
* starting point. The root directory uses an obj_id of 1.
*
* FUTURE: Leave the root directory cached referenced but unlocked
* in hmp->rootvp (need to flush it on unmount).
*/
error = hammer_vfs_vget(mp, NULL, 1, &rootvp);
if (error)
goto done;
vput(rootvp);
/*vn_unlock(hmp->rootvp);*/
if (hmp->ronly == 0)
error = hammer_recover_stage2(hmp, rootvol);
/*
* If the stage2 recovery fails be sure to clean out all cached
* vnodes before throwing away the mount structure or bad things
* will happen.
*/
if (error)
vflush(mp, 0, 0);
done:
if ((mp->mnt_flag & MNT_UPDATE) == 0) {
/* New mount */
/* Populate info for mount point (NULL pad)*/
bzero(mp->mnt_stat.f_mntonname, MNAMELEN);
size_t size;
if (mntpt) {
copyinstr(mntpt, mp->mnt_stat.f_mntonname,
MNAMELEN -1, &size);
} else { /* Root mount */
mp->mnt_stat.f_mntonname[0] = '/';
}
}
(void)VFS_STATFS(mp, &mp->mnt_stat, cred);
hammer_rel_volume(rootvol, 0);
failed:
/*
* Cleanup and return.
*/
if (error) {
/* called with fs_token held */
hammer_free_hmp(mp);
} else {
lwkt_reltoken(&hmp->fs_token);
}
return (error);
}
int
hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip,
struct hammer_ioc_reblock *reblock)
{
struct hammer_cursor cursor;
hammer_btree_elm_t elm;
int checkspace_count;
int error;
int seq;
int slop;
/*
* A fill level <= 20% is considered an emergency. free_level is
* inverted from fill_level.
*/
if (reblock->free_level >= HAMMER_LARGEBLOCK_SIZE * 8 / 10)
slop = HAMMER_CHKSPC_EMERGENCY;
else
slop = HAMMER_CHKSPC_REBLOCK;
if ((reblock->key_beg.localization | reblock->key_end.localization) &
HAMMER_LOCALIZE_PSEUDOFS_MASK) {
return(EINVAL);
}
if (reblock->key_beg.obj_id >= reblock->key_end.obj_id)
return(EINVAL);
if (reblock->free_level < 0)
return(EINVAL);
reblock->key_cur = reblock->key_beg;
reblock->key_cur.localization &= HAMMER_LOCALIZE_MASK;
reblock->key_cur.localization += ip->obj_localization;
checkspace_count = 0;
seq = trans->hmp->flusher.done;
retry:
error = hammer_init_cursor(trans, &cursor, NULL, NULL);
if (error) {
hammer_done_cursor(&cursor);
goto failed;
}
cursor.key_beg.localization = reblock->key_cur.localization;
cursor.key_beg.obj_id = reblock->key_cur.obj_id;
cursor.key_beg.key = HAMMER_MIN_KEY;
cursor.key_beg.create_tid = 1;
cursor.key_beg.delete_tid = 0;
cursor.key_beg.rec_type = HAMMER_MIN_RECTYPE;
cursor.key_beg.obj_type = 0;
cursor.key_end.localization = (reblock->key_end.localization &
HAMMER_LOCALIZE_MASK) +
ip->obj_localization;
cursor.key_end.obj_id = reblock->key_end.obj_id;
cursor.key_end.key = HAMMER_MAX_KEY;
cursor.key_end.create_tid = HAMMER_MAX_TID - 1;
cursor.key_end.delete_tid = 0;
cursor.key_end.rec_type = HAMMER_MAX_RECTYPE;
cursor.key_end.obj_type = 0;
cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
cursor.flags |= HAMMER_CURSOR_BACKEND;
cursor.flags |= HAMMER_CURSOR_NOSWAPCACHE;
/*
* This flag allows the btree scan code to return internal nodes,
* so we can reblock them in addition to the leafs. Only specify it
* if we intend to reblock B-Tree nodes.
*/
if (reblock->head.flags & HAMMER_IOC_DO_BTREE)
cursor.flags |= HAMMER_CURSOR_REBLOCKING;
error = hammer_btree_first(&cursor);
while (error == 0) {
/*
* Internal or Leaf node
*/
KKASSERT(cursor.index < cursor.node->ondisk->count);
elm = &cursor.node->ondisk->elms[cursor.index];
reblock->key_cur.obj_id = elm->base.obj_id;
reblock->key_cur.localization = elm->base.localization;
/*
* Yield to more important tasks
*/
if ((error = hammer_signal_check(trans->hmp)) != 0)
break;
/*
* If there is insufficient free space it may be due to
* reserved bigblocks, which flushing might fix.
*
* We must force a retest in case the unlocked cursor is
* moved to the end of the leaf, or moved to an internal
* node.
*
* WARNING: See warnings in hammer_unlock_cursor() function.
*/
if (hammer_checkspace(trans->hmp, slop)) {
if (++checkspace_count == 10) {
error = ENOSPC;
break;
}
hammer_unlock_cursor(&cursor);
cursor.flags |= HAMMER_CURSOR_RETEST;
hammer_flusher_wait(trans->hmp, seq);
hammer_lock_cursor(&cursor);
seq = hammer_flusher_async(trans->hmp, NULL);
goto skip;
}
/*
* Acquiring the sync_lock prevents the operation from
* crossing a synchronization boundary.
*
* NOTE: cursor.node may have changed on return.
*
* WARNING: See warnings in hammer_unlock_cursor() function.
*/
hammer_sync_lock_sh(trans);
error = hammer_reblock_helper(reblock, &cursor, elm);
hammer_sync_unlock(trans);
while (hammer_flusher_meta_halflimit(trans->hmp) ||
hammer_flusher_undo_exhausted(trans, 2)) {
hammer_unlock_cursor(&cursor);
hammer_flusher_wait(trans->hmp, seq);
hammer_lock_cursor(&cursor);
seq = hammer_flusher_async_one(trans->hmp);
}
/*
* Setup for iteration, our cursor flags may be modified by
* other threads while we are unlocked.
*/
cursor.flags |= HAMMER_CURSOR_ATEDISK;
/*
* We allocate data buffers, which atm we don't track
* dirty levels for because we allow the kernel to write
* them. But if we allocate too many we can still deadlock
* the buffer cache.
*
* WARNING: See warnings in hammer_unlock_cursor() function.
* (The cursor's node and element may change!)
*/
if (bd_heatup()) {
hammer_unlock_cursor(&cursor);
bwillwrite(HAMMER_XBUFSIZE);
hammer_lock_cursor(&cursor);
}
/* XXX vm_wait_nominal(); */
skip:
if (error == 0) {
error = hammer_btree_iterate(&cursor);
}
}
if (error == ENOENT)
error = 0;
hammer_done_cursor(&cursor);
if (error == EWOULDBLOCK) {
hammer_flusher_sync(trans->hmp);
goto retry;
}
if (error == EDEADLK)
goto retry;
if (error == EINTR) {
reblock->head.flags |= HAMMER_IOC_HEAD_INTR;
error = 0;
}
failed:
reblock->key_cur.localization &= HAMMER_LOCALIZE_MASK;
return(error);
}
/*
* Remove a volume.
*/
int
hammer_ioc_volume_del(hammer_transaction_t trans, hammer_inode_t ip,
struct hammer_ioc_volume *ioc)
{
struct hammer_mount *hmp = trans->hmp;
struct mount *mp = hmp->mp;
hammer_volume_t volume;
int error = 0;
if (mp->mnt_flag & MNT_RDONLY) {
kprintf("Cannot del volume from read-only HAMMER filesystem\n");
return (EINVAL);
}
if (hammer_lock_ex_try(&hmp->volume_lock) != 0) {
kprintf("Another volume operation is in progress!\n");
return (EAGAIN);
}
volume = NULL;
/*
* find volume by volname
*/
for (int vol_no = 0; vol_no < HAMMER_MAX_VOLUMES; ++vol_no) {
volume = hammer_get_volume(hmp, vol_no, &error);
if (volume == NULL && error == ENOENT) {
/*
* Skip unused volume numbers
*/
error = 0;
continue;
}
KKASSERT(volume != NULL && error == 0);
if (strcmp(volume->vol_name, ioc->device_name) == 0) {
break;
}
hammer_rel_volume(volume, 0);
volume = NULL;
}
if (volume == NULL) {
kprintf("Couldn't find volume\n");
error = EINVAL;
goto end;
}
if (volume == trans->rootvol) {
kprintf("Cannot remove root-volume\n");
hammer_rel_volume(volume, 0);
error = EINVAL;
goto end;
}
/*
*
*/
hmp->volume_to_remove = volume->vol_no;
struct hammer_ioc_reblock reblock;
bzero(&reblock, sizeof(reblock));
reblock.key_beg.localization = HAMMER_MIN_LOCALIZATION;
reblock.key_beg.obj_id = HAMMER_MIN_OBJID;
reblock.key_end.localization = HAMMER_MAX_LOCALIZATION;
reblock.key_end.obj_id = HAMMER_MAX_OBJID;
reblock.head.flags = HAMMER_IOC_DO_FLAGS;
reblock.free_level = 0;
error = hammer_ioc_reblock(trans, ip, &reblock);
if (reblock.head.flags & HAMMER_IOC_HEAD_INTR) {
error = EINTR;
}
if (error) {
if (error == EINTR) {
kprintf("reblock was interrupted\n");
} else {
kprintf("reblock failed: %d\n", error);
}
hmp->volume_to_remove = -1;
hammer_rel_volume(volume, 0);
goto end;
}
/*
* Sync filesystem
*/
int count = 0;
while (hammer_flusher_haswork(hmp)) {
hammer_flusher_sync(hmp);
++count;
if (count >= 5) {
if (count == 5)
kprintf("HAMMER: flushing.");
else
kprintf(".");
tsleep(&count, 0, "hmrufl", hz);
}
if (count == 30) {
kprintf("giving up");
break;
}
}
kprintf("\n");
hammer_sync_lock_sh(trans);
hammer_lock_ex(&hmp->blkmap_lock);
/*
* We use stat later to update rootvol's bigblock stats
*/
struct bigblock_stat stat;
error = hammer_free_freemap(trans, volume, &stat);
if (error) {
kprintf("Failed to free volume. Volume not empty!\n");
hmp->volume_to_remove = -1;
hammer_rel_volume(volume, 0);
hammer_unlock(&hmp->blkmap_lock);
hammer_sync_unlock(trans);
goto end;
}
hmp->volume_to_remove = -1;
hammer_rel_volume(volume, 0);
/*
* Unload buffers
*/
RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
hammer_unload_buffer, volume);
error = hammer_unload_volume(volume, NULL);
if (error == -1) {
kprintf("Failed to unload volume\n");
hammer_unlock(&hmp->blkmap_lock);
hammer_sync_unlock(trans);
goto end;
}
volume = NULL;
--hmp->nvolumes;
/*
* Set each volume's new value of the vol_count field.
*/
for (int vol_no = 0; vol_no < HAMMER_MAX_VOLUMES; ++vol_no) {
volume = hammer_get_volume(hmp, vol_no, &error);
if (volume == NULL && error == ENOENT) {
/*
* Skip unused volume numbers
*/
error = 0;
continue;
}
KKASSERT(volume != NULL && error == 0);
hammer_modify_volume_field(trans, volume, vol_count);
volume->ondisk->vol_count = hmp->nvolumes;
hammer_modify_volume_done(volume);
/*
* Only changes to the header of the root volume
* are automatically flushed to disk. For all
* other volumes that we modify we do it here.
*
* No interlock is needed, volume buffers are not
* messed with by bioops.
*/
if (volume != trans->rootvol && volume->io.modified) {
hammer_crc_set_volume(volume->ondisk);
hammer_io_flush(&volume->io, 0);
}
hammer_rel_volume(volume, 0);
}
/*
* Update the total number of bigblocks
*/
hammer_modify_volume_field(trans, trans->rootvol,
vol0_stat_bigblocks);
trans->rootvol->ondisk->vol0_stat_bigblocks -= stat.total_bigblocks;
hammer_modify_volume_done(trans->rootvol);
/*
* Update the number of free bigblocks
* (including the copy in hmp)
*/
hammer_modify_volume_field(trans, trans->rootvol,
vol0_stat_freebigblocks);
trans->rootvol->ondisk->vol0_stat_freebigblocks -= stat.total_free_bigblocks;
hmp->copy_stat_freebigblocks =
trans->rootvol->ondisk->vol0_stat_freebigblocks;
hammer_modify_volume_done(trans->rootvol);
/*
* Bigblock count changed so recompute the total number of blocks.
*/
mp->mnt_stat.f_blocks = trans->rootvol->ondisk->vol0_stat_bigblocks *
(HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE);
mp->mnt_vstat.f_blocks = trans->rootvol->ondisk->vol0_stat_bigblocks *
(HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE);
hammer_unlock(&hmp->blkmap_lock);
hammer_sync_unlock(trans);
/*
* Erase the volume header of the removed device.
*
* This is to not accidentally mount the volume again.
*/
struct vnode *devvp = NULL;
error = hammer_setup_device(&devvp, ioc->device_name, 0);
if (error) {
kprintf("Failed to open device: %s\n", ioc->device_name);
goto end;
}
KKASSERT(devvp);
error = hammer_clear_volume_header(devvp);
if (error) {
kprintf("Failed to clear volume header of device: %s\n",
ioc->device_name);
goto end;
}
hammer_close_device(&devvp, 0);
KKASSERT(error == 0);
end:
hammer_unlock(&hmp->volume_lock);
return (error);
}
