Archive-Team
/
zfs
2
0
Fork 0
Code Issues Pull Requests Packages Projects Releases 103 Wiki Activity

Merge branch 'fix-stack-traverse_visitbp' into refs/top-bases/fix-branch

This commit is contained in:
Brian Behlendorf 2010-06-29 11:04:51 -07:00
parent 96bd9f434d 34229a2f2a
commit 46b2d46053
1 changed files with 177 additions and 96 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

273
module/zfs/dmu_traverse.c
View File

@ -57,6 +57,33 @@ struct traverse_data {
void *td_arg; void *td_arg;
}; };
struct traverse_visitbp_data {
/* Function arguments */
struct traverse_data *tv_td;
const dnode_phys_t *tv_dnp;
arc_buf_t *tv_pbuf;
blkptr_t *tv_bp;
const zbookmark_t *tv_zb;
/* Local variables */
struct prefetch_data *tv_pd;
zbookmark_t tv_czb;
arc_buf_t *tv_buf;
boolean_t tv_hard;
objset_phys_t *tv_osp;
dnode_phys_t *tv_ldnp;
blkptr_t *tv_cbp;
uint32_t tv_flags;
int tv_err;
int tv_lasterr;
int tv_i;
int tv_epb;
#ifdef DEBUG
int tv_depth;
#endif
};
static inline int traverse_visitbp(struct traverse_data *td, const
dnode_phys_t *dnp, arc_buf_t *pbuf, blkptr_t *bp, const zbookmark_t *zb);
static int traverse_dnode(struct traverse_data *td, const dnode_phys_t *dnp, static int traverse_dnode(struct traverse_data *td, const dnode_phys_t *dnp,
arc_buf_t *buf, uint64_t objset, uint64_t object); arc_buf_t *buf, uint64_t objset, uint64_t object);
@ -128,137 +155,191 @@ traverse_zil(struct traverse_data *td, zil_header_t *zh)
zil_free(zilog); zil_free(zilog);
} }
static int #define TRAVERSE_VISITBP_MAX_DEPTH 20
traverse_visitbp(struct traverse_data *td, const dnode_phys_t *dnp,
arc_buf_t *pbuf, blkptr_t *bp, const zbookmark_t *zb)
{
zbookmark_t czb;
int err = 0, lasterr = 0;
arc_buf_t *buf = NULL;
struct prefetch_data *pd = td->td_pfd;
boolean_t hard = td->td_flags & TRAVERSE_HARD;
if (bp->blk_birth == 0) { static void
err = td->td_func(td->td_spa, NULL, NULL, pbuf, zb, dnp, __traverse_visitbp_init(struct traverse_visitbp_data *tv,
td->td_arg); struct traverse_data *td, const dnode_phys_t *dnp,
return (err); arc_buf_t *pbuf, blkptr_t *bp, const zbookmark_t *zb, int depth)
{
tv->tv_td = td;
tv->tv_dnp = dnp;
tv->tv_pbuf = pbuf;
tv->tv_bp = bp;
tv->tv_zb = zb;
tv->tv_err = 0;
tv->tv_lasterr = 0;
tv->tv_buf = NULL;
tv->tv_pd = td->td_pfd;
tv->tv_hard = td->td_flags & TRAVERSE_HARD;
tv->tv_flags = ARC_WAIT;
tv->tv_depth = depth;
}
static noinline int
__traverse_visitbp(struct traverse_visitbp_data *tv)
{
ASSERT3S(tv->tv_depth, <, TRAVERSE_VISITBP_MAX_DEPTH);
if (tv->tv_bp->blk_birth == 0) {
tv->tv_err = tv->tv_td->td_func(tv->tv_td->td_spa, NULL, NULL,
tv->tv_pbuf, tv->tv_zb, tv->tv_dnp, tv->tv_td->td_arg);
return (tv->tv_err);
} }
if (bp->blk_birth <= td->td_min_txg) if (tv->tv_bp->blk_birth <= tv->tv_td->td_min_txg)
return (0); return (0);
if (pd && !pd->pd_exited && if (tv->tv_pd && !tv->tv_pd->pd_exited &&
((pd->pd_flags & TRAVERSE_PREFETCH_DATA) || ((tv->tv_pd->pd_flags & TRAVERSE_PREFETCH_DATA) ||
BP_GET_TYPE(bp) == DMU_OT_DNODE || BP_GET_LEVEL(bp) > 0)) { BP_GET_TYPE(tv->tv_bp) == DMU_OT_DNODE ||
mutex_enter(&pd->pd_mtx); BP_GET_LEVEL(tv->tv_bp) > 0)) {
ASSERT(pd->pd_blks_fetched >= 0); mutex_enter(&tv->tv_pd->pd_mtx);
while (pd->pd_blks_fetched == 0 && !pd->pd_exited) ASSERT(tv->tv_pd->pd_blks_fetched >= 0);
cv_wait(&pd->pd_cv, &pd->pd_mtx); while (tv->tv_pd->pd_blks_fetched == 0 && !tv->tv_pd->pd_exited)
pd->pd_blks_fetched--; cv_wait(&tv->tv_pd->pd_cv, &tv->tv_pd->pd_mtx);
cv_broadcast(&pd->pd_cv); tv->tv_pd->pd_blks_fetched--;
mutex_exit(&pd->pd_mtx); cv_broadcast(&tv->tv_pd->pd_cv);
mutex_exit(&tv->tv_pd->pd_mtx);
} }
if (td->td_flags & TRAVERSE_PRE) { if (tv->tv_td->td_flags & TRAVERSE_PRE) {
err = td->td_func(td->td_spa, NULL, bp, pbuf, zb, dnp, tv->tv_err = tv->tv_td->td_func(tv->tv_td->td_spa, NULL,
td->td_arg); tv->tv_bp, tv->tv_pbuf, tv->tv_zb, tv->tv_dnp,
if (err) tv->tv_td->td_arg);
return (err); if (tv->tv_err)
return (tv->tv_err);
} }
if (BP_GET_LEVEL(bp) > 0) { if (BP_GET_LEVEL(tv->tv_bp) > 0) {
uint32_t flags = ARC_WAIT; tv->tv_epb = BP_GET_LSIZE(tv->tv_bp) >> SPA_BLKPTRSHIFT;
int i;
blkptr_t *cbp;
int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
err = dsl_read(NULL, td->td_spa, bp, pbuf, tv->tv_err = dsl_read(NULL, tv->tv_td->td_spa, tv->tv_bp,
arc_getbuf_func, &buf, tv->tv_pbuf, arc_getbuf_func, &tv->tv_buf,
ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
if (err) &tv->tv_flags, tv->tv_zb);
return (err); if (tv->tv_err)
return (tv->tv_err);
/* recursively visitbp() blocks below this */ /* recursively visitbp() blocks below this */
cbp = buf->b_data; tv->tv_cbp = tv->tv_buf->b_data;
for (i = 0; i < epb; i++, cbp++) { for (tv->tv_i = 0; tv->tv_i < tv->tv_epb;
SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object, tv->tv_i++, tv->tv_cbp++) {
zb->zb_level - 1, SET_BOOKMARK(&tv->tv_czb, tv->tv_zb->zb_objset,
zb->zb_blkid * epb + i); tv->tv_zb->zb_object, tv->tv_zb->zb_level - 1,
err = traverse_visitbp(td, dnp, buf, cbp, &czb); tv->tv_zb->zb_blkid * tv->tv_epb + tv->tv_i);
if (err) { __traverse_visitbp_init(tv + 1, tv->tv_td,
if (!hard) tv->tv_dnp, tv->tv_buf, tv->tv_cbp,
&tv->tv_czb, tv->tv_depth + 1);
tv->tv_err = __traverse_visitbp(tv + 1);
if (tv->tv_err) {
if (!tv->tv_hard)
break; break;
lasterr = err; tv->tv_lasterr = tv->tv_err;
} }
} }
} else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) { } else if (BP_GET_TYPE(tv->tv_bp) == DMU_OT_DNODE) {
uint32_t flags = ARC_WAIT; tv->tv_epb = BP_GET_LSIZE(tv->tv_bp) >> DNODE_SHIFT;
int i;
int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT;
err = dsl_read(NULL, td->td_spa, bp, pbuf, tv->tv_err = dsl_read(NULL, tv->tv_td->td_spa, tv->tv_bp,
arc_getbuf_func, &buf, tv->tv_pbuf, arc_getbuf_func, &tv->tv_buf,
ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
if (err) &tv->tv_flags, tv->tv_zb);
return (err); if (tv->tv_err)
return (tv->tv_err);
/* recursively visitbp() blocks below this */ /* recursively visitbp() blocks below this */
dnp = buf->b_data; tv->tv_dnp = tv->tv_buf->b_data;
for (i = 0; i < epb; i++, dnp++) { for (tv->tv_i = 0; tv->tv_i < tv->tv_epb;
err = traverse_dnode(td, dnp, buf, zb->zb_objset, tv->tv_i++, tv->tv_dnp++) {
zb->zb_blkid * epb + i); tv->tv_err = traverse_dnode(tv->tv_td, tv->tv_dnp,
if (err) { tv->tv_buf, tv->tv_zb->zb_objset,
if (!hard) tv->tv_zb->zb_blkid * tv->tv_epb + tv->tv_i);
if (tv->tv_err) {
if (!tv->tv_hard)
break; break;
lasterr = err; tv->tv_lasterr = tv->tv_err;
} }
} }
} else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) { } else if (BP_GET_TYPE(tv->tv_bp) == DMU_OT_OBJSET) {
uint32_t flags = ARC_WAIT;
objset_phys_t *osp;
dnode_phys_t *dnp;
err = dsl_read_nolock(NULL, td->td_spa, bp, tv->tv_err = dsl_read_nolock(NULL, tv->tv_td->td_spa,
arc_getbuf_func, &buf, tv->tv_bp, arc_getbuf_func, &tv->tv_buf,
ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb); ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
if (err) &tv->tv_flags, tv->tv_zb);
return (err); if (tv->tv_err)
return (tv->tv_err);
osp = buf->b_data; tv->tv_osp = tv->tv_buf->b_data;
traverse_zil(td, &osp->os_zil_header); traverse_zil(tv->tv_td, &tv->tv_osp->os_zil_header);
dnp = &osp->os_meta_dnode; tv->tv_ldnp = &tv->tv_osp->os_meta_dnode;
err = traverse_dnode(td, dnp, buf, zb->zb_objset, tv->tv_err = traverse_dnode(tv->tv_td, tv->tv_ldnp, tv->tv_buf,
DMU_META_DNODE_OBJECT); tv->tv_zb->zb_objset, DMU_META_DNODE_OBJECT);
if (err && hard) { if (tv->tv_err && tv->tv_hard) {
lasterr = err; tv->tv_lasterr = tv->tv_err;
err = 0; tv->tv_err = 0;
} }
if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) { if (tv->tv_err == 0 &&
dnp = &osp->os_userused_dnode; arc_buf_size(tv->tv_buf) >= sizeof (objset_phys_t)) {
err = traverse_dnode(td, dnp, buf, zb->zb_objset, tv->tv_ldnp = &tv->tv_osp->os_userused_dnode;
tv->tv_err = traverse_dnode(tv->tv_td, tv->tv_ldnp,
tv->tv_buf, tv->tv_zb->zb_objset,
DMU_USERUSED_OBJECT); DMU_USERUSED_OBJECT);
} }
if (err && hard) { if (tv->tv_err && tv->tv_hard) {
lasterr = err; tv->tv_lasterr = tv->tv_err;
err = 0; tv->tv_err = 0;
} }
if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) { if (tv->tv_err == 0 &&
dnp = &osp->os_groupused_dnode; arc_buf_size(tv->tv_buf) >= sizeof (objset_phys_t)) {
err = traverse_dnode(td, dnp, buf, zb->zb_objset, tv->tv_ldnp = &tv->tv_osp->os_groupused_dnode;
tv->tv_err = traverse_dnode(tv->tv_td, tv->tv_ldnp,
tv->tv_buf, tv->tv_zb->zb_objset,
DMU_GROUPUSED_OBJECT); DMU_GROUPUSED_OBJECT);
} }
} }
if (buf) if (tv->tv_buf)
(void) arc_buf_remove_ref(buf, &buf); (void) arc_buf_remove_ref(tv->tv_buf, &tv->tv_buf);
if (err == 0 && lasterr == 0 && (td->td_flags & TRAVERSE_POST)) { if (tv->tv_err == 0 && tv->tv_lasterr == 0 &&
err = td->td_func(td->td_spa, NULL, bp, pbuf, zb, dnp, (tv->tv_td->td_flags & TRAVERSE_POST)) {
td->td_arg); tv->tv_err = tv->tv_td->td_func(tv->tv_td->td_spa, NULL,
tv->tv_bp, tv->tv_pbuf, tv->tv_zb, tv->tv_dnp,
tv->tv_td->td_arg);
} }
return (err != 0 ? err : lasterr); return (tv->tv_err != 0 ? tv->tv_err : tv->tv_lasterr);
}
/*
* Due to limited stack space recursive functions are frowned upon in
* the Linux kernel. However, they often are the most elegant solution
* to a problem. The following code preserves the recursive function
* traverse_visitbp() but moves the local variables AND function
* arguments to the heap to minimize the stack frame size. Enough
* space is initially allocated on the stack for 16 levels of recursion.
* This change does ugly-up-the-code but it reduces the worst case
* usage from roughly 2496 bytes to 576 bytes on x86_64 archs.
*/
static int
traverse_visitbp(struct traverse_data *td, const dnode_phys_t *dnp,
arc_buf_t *pbuf, blkptr_t *bp, const zbookmark_t *zb)
{
struct traverse_visitbp_data *tv;
int error;
tv = kmem_zalloc(sizeof(struct traverse_visitbp_data) *
TRAVERSE_VISITBP_MAX_DEPTH, KM_SLEEP);
__traverse_visitbp_init(tv, td, dnp, pbuf, bp, zb, 0);
error = __traverse_visitbp(tv);
kmem_free(tv, sizeof(struct traverse_visitbp_data) *
TRAVERSE_VISITBP_MAX_DEPTH);
return (error);
} }
static int static int