Move more pnfs-isms out of the generic commit code.
Bugfixes:
- filelayout_scan_commit_lists doesn't need to get/put the lseg.
In fact since it is run under the inode->i_lock, the lseg_put()
can deadlock.
- Ensure that we distinguish between what needs to be done for
commit-to-data server and what needs to be done for commit-to-MDS
using the new flag PG_COMMIT_TO_DS. Otherwise we may end up calling
put_lseg() on a bucket for a struct nfs_page that got written
through the MDS.
- Fix a case where we were using list_del() on an nfs_page->wb_list
instead of list_del_init().
- filelayout_initiate_commit needs to call filelayout_commit_release
on error instead of the mds_ops->rpc_release(). Otherwise it won't
clear the commit lock.
Cleanups:
- Let the files layout manage the commit lists for the pNFS case.
Don't expose stuff like pnfs_choose_commit_list, and the fact
that the commit buckets hold references to the layout segment
in common code.
- Cast out the put_lseg() calls for the struct nfs_read/write_data->lseg
into the pNFS layer from whence they came.
- Let the pNFS layer manage the NFS_INO_PNFS_COMMIT bit.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Cc: Fred Isaman <iisaman@netapp.com>
extern void nfs_readdata_release(struct nfs_read_data *rdata);
/* write.c */
-extern int nfs_scan_commit_list(struct list_head *src, struct list_head *dst,
- int max);
extern int nfs_generic_flush(struct nfs_pageio_descriptor *desc,
struct list_head *head);
extern void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor *pgio,
void nfs_commit_clear_lock(struct nfs_inode *nfsi);
void nfs_commitdata_release(void *data);
void nfs_commit_release_pages(struct nfs_write_data *data);
+void nfs_request_add_commit_list(struct nfs_page *req, struct list_head *head);
+void nfs_request_remove_commit_list(struct nfs_page *req);
#ifdef CONFIG_MIGRATION
extern int nfs_migrate_page(struct address_space *,
{
struct nfs_read_data *rdata = (struct nfs_read_data *)data;
+ put_lseg(rdata->lseg);
rdata->mds_ops->rpc_release(data);
}
{
struct nfs_write_data *wdata = (struct nfs_write_data *)data;
+ put_lseg(wdata->lseg);
wdata->mds_ops->rpc_release(data);
}
nfs_commit_release_pages(wdata);
if (atomic_dec_and_test(&NFS_I(wdata->inode)->commits_outstanding))
nfs_commit_clear_lock(NFS_I(wdata->inode));
+ put_lseg(wdata->lseg);
nfs_commitdata_release(wdata);
}
/* The generic layer is about to remove the req from the commit list.
* If this will make the bucket empty, it will need to put the lseg reference.
- * Note inode lock is held, so we can't do the put here.
*/
-static struct pnfs_layout_segment *
-filelayout_remove_commit_req(struct nfs_page *req)
+static void
+filelayout_clear_request_commit(struct nfs_page *req)
{
+ struct pnfs_layout_segment *freeme = NULL;
+ struct inode *inode = req->wb_context->dentry->d_inode;
+
+ spin_lock(&inode->i_lock);
+ if (!test_and_clear_bit(PG_COMMIT_TO_DS, &req->wb_flags))
+ goto out;
if (list_is_singular(&req->wb_list)) {
struct inode *inode = req->wb_context->dentry->d_inode;
struct pnfs_layout_segment *lseg;
* 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)
- return lseg;
+ if (lseg->pls_range.iomode == IOMODE_RW) {
+ freeme = lseg;
+ break;
+ }
}
}
- return NULL;
+out:
+ nfs_request_remove_commit_list(req);
+ spin_unlock(&inode->i_lock);
+ put_lseg(freeme);
}
static struct list_head *
*/
get_lseg(lseg);
}
+ set_bit(PG_COMMIT_TO_DS, &req->wb_flags);
return list;
}
+static void
+filelayout_mark_request_commit(struct nfs_page *req,
+ struct pnfs_layout_segment *lseg)
+{
+ struct list_head *list;
+
+ list = filelayout_choose_commit_list(req, lseg);
+ nfs_request_add_commit_list(req, list);
+}
+
static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
{
struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
prepare_to_resend_writes(data);
- data->mds_ops->rpc_release(data);
+ filelayout_commit_release(data);
return -EAGAIN;
}
dprintk("%s ino %lu, how %d\n", __func__, data->inode->i_ino, how);
list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list)
if (lseg->pls_range.iomode == IOMODE_RW)
- return get_lseg(lseg);
+ return lseg;
return NULL;
}
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)
+{
+ struct list_head *src = &bucket->written;
+ struct list_head *dst = &bucket->committing;
+ struct nfs_page *req, *tmp;
+ int ret = 0;
+
+ list_for_each_entry_safe(req, tmp, src, wb_list) {
+ if (!nfs_lock_request(req))
+ continue;
+ nfs_request_remove_commit_list(req);
+ clear_bit(PG_COMMIT_TO_DS, &req->wb_flags);
+ nfs_list_add_request(req, dst);
+ ret++;
+ if (ret == max)
+ break;
+ }
+ return ret;
+}
+
/* Move reqs from written to committing lists, returning count of number moved.
* Note called with i_lock held.
*/
lseg = find_only_write_lseg_locked(inode);
if (!lseg)
- return 0;
+ goto out_done;
fl = FILELAYOUT_LSEG(lseg);
if (fl->commit_through_mds)
- goto out_put;
- for (i = 0; i < fl->number_of_buckets; i++) {
- if (list_empty(&fl->commit_buckets[i].written))
- continue;
- cnt = nfs_scan_commit_list(&fl->commit_buckets[i].written,
- &fl->commit_buckets[i].committing,
- max);
+ goto out_done;
+ for (i = 0; i < fl->number_of_buckets && max != 0; i++) {
+ cnt = filelayout_scan_ds_commit_list(&fl->commit_buckets[i], max);
max -= cnt;
rv += cnt;
}
-out_put:
- put_lseg(lseg);
+out_done:
return rv;
}
.free_lseg = filelayout_free_lseg,
.pg_read_ops = &filelayout_pg_read_ops,
.pg_write_ops = &filelayout_pg_write_ops,
- .choose_commit_list = filelayout_choose_commit_list,
- .remove_commit_req = filelayout_remove_commit_req,
+ .mark_request_commit = filelayout_mark_request_commit,
+ .clear_request_commit = filelayout_clear_request_commit,
.scan_commit_lists = filelayout_scan_commit_lists,
.commit_pagelist = filelayout_commit_pagelist,
.read_pagelist = filelayout_read_pagelist,
}
data->task.tk_status = pnfs_write_done_resend_to_mds(data->inode, &data->pages);
}
+ put_lseg(data->lseg);
data->mds_ops->rpc_release(data);
}
EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
nfs_list_add_request(data->req, &desc->pg_list);
nfs_pageio_reset_write_mds(desc);
desc->pg_recoalesce = 1;
+ put_lseg(data->lseg);
nfs_writedata_release(data);
}
data->mds_ops->rpc_call_done(&data->task, data);
} else
pnfs_ld_handle_read_error(data);
+ put_lseg(data->lseg);
data->mds_ops->rpc_release(data);
}
EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
const struct nfs_pageio_ops *pg_read_ops;
const struct nfs_pageio_ops *pg_write_ops;
- struct list_head * (*choose_commit_list) (struct nfs_page *req,
+ void (*mark_request_commit) (struct nfs_page *req,
struct pnfs_layout_segment *lseg);
- struct pnfs_layout_segment *(*remove_commit_req) (struct nfs_page *req);
+ void (*clear_request_commit) (struct nfs_page *req);
int (*scan_commit_lists) (struct inode *inode, int max);
int (*commit_pagelist)(struct inode *inode, struct list_head *mds_pages, int how);
return NFS_SERVER(inode)->pnfs_curr_ld->commit_pagelist(inode, mds_pages, how);
}
-static inline struct list_head *
-pnfs_choose_commit_list(struct nfs_page *req, struct pnfs_layout_segment *lseg)
+static inline bool
+pnfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg)
{
struct inode *inode = req->wb_context->dentry->d_inode;
- struct list_head *rv;
+ struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
- if (lseg && NFS_SERVER(inode)->pnfs_curr_ld->choose_commit_list)
- rv = NFS_SERVER(inode)->pnfs_curr_ld->choose_commit_list(req, lseg);
- else
- rv = &NFS_I(inode)->commit_list;
- return rv;
+ if (lseg == NULL || ld->mark_request_commit == NULL)
+ return false;
+ ld->mark_request_commit(req, lseg);
+ return true;
}
-static inline struct pnfs_layout_segment *
+static inline bool
pnfs_clear_request_commit(struct nfs_page *req)
{
struct inode *inode = req->wb_context->dentry->d_inode;
+ struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
- if (NFS_SERVER(inode)->pnfs_curr_ld &&
- NFS_SERVER(inode)->pnfs_curr_ld->remove_commit_req)
- return NFS_SERVER(inode)->pnfs_curr_ld->remove_commit_req(req);
- else
- return NULL;
+ if (ld == NULL || ld->clear_request_commit == NULL)
+ return false;
+ ld->clear_request_commit(req);
+ return true;
}
static inline int
pnfs_scan_commit_lists(struct inode *inode, int max)
{
- if (NFS_SERVER(inode)->pnfs_curr_ld &&
- NFS_SERVER(inode)->pnfs_curr_ld->scan_commit_lists)
- return NFS_SERVER(inode)->pnfs_curr_ld->scan_commit_lists(inode, max);
- else
+ struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
+ int ret;
+
+ if (ld == NULL || ld->scan_commit_lists == NULL)
return 0;
+ ret = ld->scan_commit_lists(inode, max);
+ if (ret != 0)
+ set_bit(NFS_INO_PNFS_COMMIT, &NFS_I(inode)->flags);
+ return ret;
}
/* Should the pNFS client commit and return the layout upon a setattr */
return PNFS_NOT_ATTEMPTED;
}
-static inline struct list_head *
-pnfs_choose_commit_list(struct nfs_page *req, struct pnfs_layout_segment *lseg)
+static inline bool
+pnfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg)
{
- struct inode *inode = req->wb_context->dentry->d_inode;
-
- return &NFS_I(inode)->commit_list;
+ return false;
}
-static inline struct pnfs_layout_segment *
+static inline bool
pnfs_clear_request_commit(struct nfs_page *req)
{
- return NULL;
+ return false;
}
static inline int
void nfs_readdata_release(struct nfs_read_data *rdata)
{
- put_lseg(rdata->lseg);
put_nfs_open_context(rdata->args.context);
nfs_readdata_free(rdata);
}
void nfs_writedata_release(struct nfs_write_data *wdata)
{
- put_lseg(wdata->lseg);
put_nfs_open_context(wdata->args.context);
nfs_writedata_free(wdata);
}
spin_unlock(&inode->i_lock);
}
-static struct pnfs_layout_segment *nfs_clear_request_commit(struct nfs_page *req);
-
/*
* Remove a write request from an inode
*/
{
struct inode *inode = req->wb_context->dentry->d_inode;
struct nfs_inode *nfsi = NFS_I(inode);
- struct pnfs_layout_segment *lseg;
BUG_ON (!NFS_WBACK_BUSY(req));
spin_lock(&inode->i_lock);
- lseg = nfs_clear_request_commit(req);
set_page_private(req->wb_page, 0);
ClearPagePrivate(req->wb_page);
clear_bit(PG_MAPPED, &req->wb_flags);
nfsi->npages--;
spin_unlock(&inode->i_lock);
- put_lseg(lseg);
nfs_release_request(req);
}
}
#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
-/*
- * Add a request to the inode's commit list.
+/**
+ * nfs_request_add_commit_list - add request to a commit list
+ * @req: pointer to a struct nfs_page
+ * @head: commit list head
+ *
+ * This sets the PG_CLEAN bit, updates the inode global count of
+ * number of outstanding requests requiring a commit as well as
+ * the MM page stats.
+ *
+ * The caller must _not_ hold the inode->i_lock, but must be
+ * holding the nfs_page lock.
*/
-static void
-nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg)
+void
+nfs_request_add_commit_list(struct nfs_page *req, struct list_head *head)
{
struct inode *inode = req->wb_context->dentry->d_inode;
- struct nfs_inode *nfsi = NFS_I(inode);
- struct list_head *clist;
- clist = pnfs_choose_commit_list(req, lseg);
- spin_lock(&inode->i_lock);
set_bit(PG_CLEAN, &(req)->wb_flags);
- nfs_list_add_request(req, clist);
- nfsi->ncommit++;
+ spin_lock(&inode->i_lock);
+ nfs_list_add_request(req, head);
+ NFS_I(inode)->ncommit++;
spin_unlock(&inode->i_lock);
inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE);
__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
}
+EXPORT_SYMBOL_GPL(nfs_request_add_commit_list);
+
+/**
+ * nfs_request_remove_commit_list - Remove request from a commit list
+ * @req: pointer to a nfs_page
+ *
+ * This clears the PG_CLEAN bit, and updates the inode global count of
+ * number of outstanding requests requiring a commit
+ * It does not update the MM page stats.
+ *
+ * The caller _must_ hold the inode->i_lock and the nfs_page lock.
+ */
+void
+nfs_request_remove_commit_list(struct nfs_page *req)
+{
+ struct inode *inode = req->wb_context->dentry->d_inode;
+
+ if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags))
+ return;
+ nfs_list_remove_request(req);
+ NFS_I(inode)->ncommit--;
+}
+EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list);
+
+
+/*
+ * Add a request to the inode's commit list.
+ */
+static void
+nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg)
+{
+ struct inode *inode = req->wb_context->dentry->d_inode;
+
+ if (pnfs_mark_request_commit(req, lseg))
+ return;
+ nfs_request_add_commit_list(req, &NFS_I(inode)->commit_list);
+}
static void
nfs_clear_page_commit(struct page *page)
dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE);
}
-static struct pnfs_layout_segment *
+static void
nfs_clear_request_commit(struct nfs_page *req)
{
- struct pnfs_layout_segment *lseg = NULL;
+ if (test_bit(PG_CLEAN, &req->wb_flags)) {
+ struct inode *inode = req->wb_context->dentry->d_inode;
- if (test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) {
+ if (!pnfs_clear_request_commit(req)) {
+ spin_lock(&inode->i_lock);
+ nfs_request_remove_commit_list(req);
+ spin_unlock(&inode->i_lock);
+ }
nfs_clear_page_commit(req->wb_page);
- lseg = pnfs_clear_request_commit(req);
- NFS_I(req->wb_context->dentry->d_inode)->ncommit--;
- list_del(&req->wb_list);
}
- return lseg;
}
static inline
return 0;
}
#else
-static inline void
+static void
nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg)
{
}
-static inline struct pnfs_layout_segment *
+static void
nfs_clear_request_commit(struct nfs_page *req)
{
- return NULL;
}
static inline
}
/* i_lock held by caller */
-int
+static int
nfs_scan_commit_list(struct list_head *src, struct list_head *dst, int max)
{
struct nfs_page *req, *tmp;
int ret = 0;
list_for_each_entry_safe(req, tmp, src, wb_list) {
- if (nfs_lock_request_dontget(req)) {
- kref_get(&req->wb_kref);
- list_move_tail(&req->wb_list, dst);
- clear_bit(PG_CLEAN, &(req)->wb_flags);
- ret++;
- if (ret == max)
- break;
- }
+ if (!nfs_lock_request(req))
+ continue;
+ nfs_request_remove_commit_list(req);
+ nfs_list_add_request(req, dst);
+ ret++;
+ if (ret == max)
+ break;
}
return ret;
}
-EXPORT_SYMBOL_GPL(nfs_scan_commit_list);
/*
* nfs_scan_commit - Scan an inode for commit requests
spin_lock(&inode->i_lock);
if (nfsi->ncommit > 0) {
+ const int max = INT_MAX;
int pnfs_ret;
- ret = nfs_scan_commit_list(&nfsi->commit_list, dst, INT_MAX);
- pnfs_ret = pnfs_scan_commit_lists(inode, INT_MAX - ret);
- if (pnfs_ret) {
- ret += pnfs_ret;
- set_bit(NFS_INO_PNFS_COMMIT, &nfsi->flags);
- }
+ ret = nfs_scan_commit_list(&nfsi->commit_list, dst, max);
+ pnfs_ret = pnfs_scan_commit_lists(inode, max - ret);
+ ret += pnfs_ret;
nfsi->ncommit -= ret;
}
spin_unlock(&inode->i_lock);
unsigned int rqend;
unsigned int end;
int error;
- struct pnfs_layout_segment *lseg = NULL;
if (!PagePrivate(page))
return NULL;
spin_lock(&inode->i_lock);
}
- lseg = nfs_clear_request_commit(req);
-
/* Okay, the request matches. Update the region */
if (offset < req->wb_offset) {
req->wb_offset = offset;
req->wb_bytes = rqend - req->wb_offset;
out_unlock:
spin_unlock(&inode->i_lock);
- put_lseg(lseg);
+ nfs_clear_request_commit(req);
return req;
out_flushme:
spin_unlock(&inode->i_lock);
{
struct nfs_write_data *wdata = data;
- put_lseg(wdata->lseg);
put_nfs_open_context(wdata->args.context);
nfs_commit_free(wdata);
}
if (req == NULL)
break;
if (nfs_lock_request_dontget(req)) {
+ nfs_clear_request_commit(req);
nfs_inode_remove_request(req);
/*
* In case nfs_inode_remove_request has marked the
PG_NEED_COMMIT,
PG_NEED_RESCHED,
PG_PARTIAL_READ_FAILED,
+ PG_COMMIT_TO_DS,
};
struct nfs_inode;
return !test_and_set_bit(PG_BUSY, &req->wb_flags);
}
+static inline int
+nfs_lock_request(struct nfs_page *req)
+{
+ if (test_and_set_bit(PG_BUSY, &req->wb_flags))
+ return 0;
+ kref_get(&req->wb_kref);
+ return 1;
+}
+
+
/**
* nfs_list_add_request - Insert a request into a list
* @req: request