Also create a commit_info structure to hold the bucket array and push
it up from the lseg to the layout where it really belongs.
While we are at it, fix a refcounting bug due to an (incorrect)
implicit assumption that filelayout_scan_ds_commit_list always
completely emptied the src list.
This clarifies refcounting, removes the ugly find_only_write_lseg
functions, and pushes the file layout commit code along on the path to
supporting multiple lsegs.
Signed-off-by: Fred Isaman <iisaman@xxxxxxxxxx>
---
fs/nfs/nfs4filelayout.c | 148 ++++++++++++++++++++++++----------------------
fs/nfs/nfs4filelayout.h | 20 ++++++-
2 files changed, 95 insertions(+), 73 deletions(-)
diff --git a/fs/nfs/nfs4filelayout.c b/fs/nfs/nfs4filelayout.c
index a8623bb..4ce0a62 100644
--- a/fs/nfs/nfs4filelayout.c
+++ b/fs/nfs/nfs4filelayout.c
@@ -650,7 +650,14 @@ filelayout_free_lseg(struct pnfs_layout_segment *lseg)
dprintk("--> %s\n", __func__);
nfs4_fl_put_deviceid(fl->dsaddr);
- kfree(fl->commit_buckets);
+ /* This assumes a single RW lseg */
+ if (lseg->pls_range.iomode == IOMODE_RW) {
+ struct nfs4_filelayout *flo;
+
+ flo = FILELAYOUT_FROM_HDR(lseg->pls_layout);
+ kfree(flo->commit_info);
+ flo->commit_info = NULL;
+ }
_filelayout_free_lseg(fl);
}
@@ -681,19 +688,27 @@ filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid,
* to filelayout_write_pagelist().
* */
if ((!fl->commit_through_mds) && (lgr->range.iomode == IOMODE_RW)) {
+ struct nfs4_filelayout *flo = FILELAYOUT_FROM_HDR(layoutid);
int i;
int size = (fl->stripe_type == STRIPE_SPARSE) ?
fl->dsaddr->ds_num : fl->dsaddr->stripe_count;
- fl->commit_buckets = kcalloc(size, sizeof(struct nfs4_fl_commit_bucket), gfp_flags);
- if (!fl->commit_buckets) {
+ if (flo->commit_info) {
+ /* Shouldn't happen if only one IOMODE_RW lseg */
+ _filelayout_free_lseg(fl);
+ return NULL;
+ }
+ flo->commit_info = kzalloc(sizeof(*flo->commit_info) +
+ size * sizeof(struct nfs4_fl_commit_bucket),
+ gfp_flags);
+ if (!flo->commit_info) {
filelayout_free_lseg(&fl->generic_hdr);
return NULL;
}
- fl->number_of_buckets = size;
+ flo->commit_info->nbuckets = size;
for (i = 0; i < size; i++) {
- INIT_LIST_HEAD(&fl->commit_buckets[i].written);
- INIT_LIST_HEAD(&fl->commit_buckets[i].committing);
+ INIT_LIST_HEAD(&flo->commit_info->buckets[i].written);
+ INIT_LIST_HEAD(&flo->commit_info->buckets[i].committing);
}
}
return &fl->generic_hdr;
@@ -793,17 +808,13 @@ filelayout_clear_request_commit(struct nfs_page *req)
if (!test_and_clear_bit(PG_COMMIT_TO_DS, &req->wb_flags))
goto out;
if (list_is_singular(&req->wb_list)) {
- struct pnfs_layout_segment *lseg;
+ struct nfs4_fl_commit_bucket *bucket;
- /* From here we can find the bucket, but for the moment,
- * since there is only one relevant lseg...
- */
- list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
- if (lseg->pls_range.iomode == IOMODE_RW) {
- freeme = lseg;
- break;
- }
- }
+ bucket = list_first_entry(&req->wb_list,
+ struct nfs4_fl_commit_bucket,
+ written);
+ freeme = bucket->wlseg;
+ bucket->wlseg = NULL;
}
out:
nfs_request_remove_commit_list(req);
@@ -818,6 +829,7 @@ filelayout_choose_commit_list(struct nfs_page *req,
struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
u32 i, j;
struct list_head *list;
+ struct nfs4_fl_commit_info *fl_cinfo;
if (fl->commit_through_mds)
return &NFS_I(req->wb_context->dentry->d_inode)->commit_list;
@@ -831,15 +843,16 @@ filelayout_choose_commit_list(struct nfs_page *req,
j = nfs4_fl_calc_j_index(lseg,
(loff_t)req->wb_index << PAGE_CACHE_SHIFT);
i = select_bucket_index(fl, j);
- list = &fl->commit_buckets[i].written;
+ fl_cinfo = FILELAYOUT_FROM_HDR(lseg->pls_layout)->commit_info;
+ list = &fl_cinfo->buckets[i].written;
if (list_empty(list)) {
/* Non-empty buckets hold a reference on the lseg. That ref
* is normally transferred to the COMMIT call and released
* there. It could also be released if the last req is pulled
* off due to a rewrite, in which case it will be done in
- * filelayout_remove_commit_req
+ * filelayout_clear_request_commit
*/
- get_lseg(lseg);
+ fl_cinfo->buckets[i].wlseg = get_lseg(lseg);
}
set_bit(PG_COMMIT_TO_DS, &req->wb_flags);
return list;
@@ -908,32 +921,6 @@ static int filelayout_initiate_commit(struct nfs_write_data *data, int how)
&filelayout_commit_call_ops, how);
}
-/*
- * This is only useful while we are using whole file layouts.
- */
-static struct pnfs_layout_segment *
-find_only_write_lseg_locked(struct inode *inode)
-{
- struct pnfs_layout_segment *lseg;
-
- list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list)
- if (lseg->pls_range.iomode == IOMODE_RW)
- return lseg;
- return NULL;
-}
-
-static struct pnfs_layout_segment *find_only_write_lseg(struct inode *inode)
-{
- struct pnfs_layout_segment *rv;
-
- spin_lock(&inode->i_lock);
- rv = find_only_write_lseg_locked(inode);
- if (rv)
- get_lseg(rv);
- spin_unlock(&inode->i_lock);
- return rv;
-}
-
static int
filelayout_scan_ds_commit_list(struct nfs4_fl_commit_bucket *bucket, int max,
spinlock_t *lock)
@@ -955,6 +942,13 @@ filelayout_scan_ds_commit_list(struct nfs4_fl_commit_bucket *bucket, int max,
if (ret == max)
break;
}
+ if (ret) {
+ bucket->clseg = bucket->wlseg;
+ if (list_empty(src))
+ bucket->wlseg = NULL;
+ else
+ get_lseg(bucket->clseg);
+ }
return ret;
}
@@ -964,18 +958,14 @@ filelayout_scan_ds_commit_list(struct nfs4_fl_commit_bucket *bucket, int max,
static int filelayout_scan_commit_lists(struct inode *inode, int max,
spinlock_t *lock)
{
- struct pnfs_layout_segment *lseg;
- struct nfs4_filelayout_segment *fl;
+ struct nfs4_fl_commit_info *fl_cinfo;
int i, rv = 0, cnt;
- lseg = find_only_write_lseg_locked(inode);
- if (!lseg)
+ fl_cinfo = FILELAYOUT_FROM_HDR(NFS_I(inode)->layout)->commit_info;
+ if (fl_cinfo == NULL)
goto out_done;
- fl = FILELAYOUT_LSEG(lseg);
- if (fl->commit_through_mds)
- goto out_done;
- for (i = 0; i < fl->number_of_buckets && max != 0; i++) {
- cnt = filelayout_scan_ds_commit_list(&fl->commit_buckets[i],
+ for (i = 0; i < fl_cinfo->nbuckets && max != 0; i++) {
+ cnt = filelayout_scan_ds_commit_list(&fl_cinfo->buckets[i],
max, lock);
max -= cnt;
rv += cnt;
@@ -987,38 +977,34 @@ out_done:
static unsigned int
alloc_ds_commits(struct inode *inode, struct list_head *list)
{
- struct pnfs_layout_segment *lseg;
- struct nfs4_filelayout_segment *fl;
+ struct nfs4_fl_commit_info *fl_cinfo;
+ struct nfs4_fl_commit_bucket *bucket;
struct nfs_write_data *data;
int i, j;
unsigned int nreq = 0;
- /* Won't need this when non-whole file layout segments are supported
- * instead we will use a pnfs_layout_hdr structure */
- lseg = find_only_write_lseg(inode);
- if (!lseg)
- return 0;
- fl = FILELAYOUT_LSEG(lseg);
- for (i = 0; i < fl->number_of_buckets; i++) {
- if (list_empty(&fl->commit_buckets[i].committing))
+ fl_cinfo = FILELAYOUT_FROM_HDR(NFS_I(inode)->layout)->commit_info;
+ for (i = 0, bucket = fl_cinfo->buckets; i < fl_cinfo->nbuckets; i++, bucket++) {
+ if (list_empty(&bucket->committing))
continue;
data = nfs_commitdata_alloc();
if (!data)
break;
data->ds_commit_index = i;
- data->lseg = lseg;
+ data->lseg = bucket->clseg;
+ bucket->clseg = NULL;
list_add(&data->pages, list);
nreq++;
}
/* Clean up on error */
- for (j = i; j < fl->number_of_buckets; j++) {
- if (list_empty(&fl->commit_buckets[i].committing))
+ for (j = i; j < fl_cinfo->nbuckets; j++, bucket++) {
+ if (list_empty(&bucket->committing))
continue;
- nfs_retry_commit(&fl->commit_buckets[i].committing, lseg);
- put_lseg(lseg); /* associated with emptying bucket */
+ nfs_retry_commit(&bucket->committing, bucket->clseg);
+ put_lseg(bucket->clseg);
+ bucket->clseg = NULL;
}
- put_lseg(lseg);
/* Caller will clean up entries put on list */
return nreq;
}
@@ -1060,7 +1046,10 @@ filelayout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
nfs_initiate_commit(data, NFS_CLIENT(inode),
data->mds_ops, how);
} else {
- nfs_init_commit(data, &FILELAYOUT_LSEG(data->lseg)->commit_buckets[data->ds_commit_index].committing, data->lseg);
+ struct nfs4_fl_commit_info *fl_cinfo;
+
+ fl_cinfo = FILELAYOUT_FROM_HDR(data->lseg->pls_layout)->commit_info;
+ nfs_init_commit(data, &fl_cinfo->buckets[data->ds_commit_index].committing, data->lseg);
filelayout_initiate_commit(data, how);
}
}
@@ -1074,10 +1063,27 @@ filelayout_free_deveiceid_node(struct nfs4_deviceid_node *d)
nfs4_fl_free_deviceid(container_of(d, struct nfs4_file_layout_dsaddr, id_node));
}
+static struct pnfs_layout_hdr *
+filelayout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
+{
+ struct nfs4_filelayout *flo;
+
+ flo = kzalloc(sizeof(*flo), gfp_flags);
+ return &flo->generic_hdr;
+}
+
+static void
+filelayout_free_layout_hdr(struct pnfs_layout_hdr *lo)
+{
+ kfree(FILELAYOUT_FROM_HDR(lo));
+}
+
static struct pnfs_layoutdriver_type filelayout_type = {
.id = LAYOUT_NFSV4_1_FILES,
.name = "LAYOUT_NFSV4_1_FILES",
.owner = THIS_MODULE,
+ .alloc_layout_hdr = filelayout_alloc_layout_hdr,
+ .free_layout_hdr = filelayout_free_layout_hdr,
.alloc_lseg = filelayout_alloc_lseg,
.free_lseg = filelayout_free_lseg,
.pg_read_ops = &filelayout_pg_read_ops,
diff --git a/fs/nfs/nfs4filelayout.h b/fs/nfs/nfs4filelayout.h
index 21190bb..7d75c03 100644
--- a/fs/nfs/nfs4filelayout.h
+++ b/fs/nfs/nfs4filelayout.h
@@ -77,6 +77,13 @@ struct nfs4_file_layout_dsaddr {
struct nfs4_fl_commit_bucket {
struct list_head written;
struct list_head committing;
+ struct pnfs_layout_segment *wlseg;
+ struct pnfs_layout_segment *clseg;
+};
+
+struct nfs4_fl_commit_info {
+ int nbuckets;
+ struct nfs4_fl_commit_bucket buckets[0];
};
struct nfs4_filelayout_segment {
@@ -89,10 +96,19 @@ struct nfs4_filelayout_segment {
struct nfs4_file_layout_dsaddr *dsaddr; /* Point to GETDEVINFO data */
unsigned int num_fh;
struct nfs_fh **fh_array;
- struct nfs4_fl_commit_bucket *commit_buckets; /* Sort commits to ds */
- int number_of_buckets;
};
+struct nfs4_filelayout {
+ struct pnfs_layout_hdr generic_hdr;
+ struct nfs4_fl_commit_info *commit_info;
+};
+
+static inline struct nfs4_filelayout *
+FILELAYOUT_FROM_HDR(struct pnfs_layout_hdr *lo)
+{
+ return container_of(lo, struct nfs4_filelayout, generic_hdr);
+}
+
static inline struct nfs4_filelayout_segment *
FILELAYOUT_LSEG(struct pnfs_layout_segment *lseg)
{
--
1.7.2.1
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
