NFS: merge _full and _partial write rpc_ops
authorFred Isaman <iisaman@netapp.com>
Fri, 20 Apr 2012 18:47:47 +0000 (14:47 -0400)
committerTrond Myklebust <Trond.Myklebust@netapp.com>
Fri, 27 Apr 2012 18:10:37 +0000 (14:10 -0400)
Decouple nfs_pgio_header and nfs_write_data, and have (possibly
multiple) nfs_write_datas each take a refcount on nfs_pgio_header.

For the moment keeps nfs_write_header as a way to preallocate a single
nfs_write_data with the nfs_pgio_header.  The code doesn't need this,
and would be prettier without, but given the amount of churn I am
already introducing I didn't want to play with tuning new mempools.

This also fixes bug in pnfs_ld_handle_write_error.  In the case of
desc->pg_bsize < PAGE_CACHE_SIZE, the pages list was empty, causing
replay attempt to do nothing.

Signed-off-by: Fred Isaman <iisaman@netapp.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
fs/nfs/direct.c
fs/nfs/internal.h
fs/nfs/nfs4filelayout.c
fs/nfs/nfs4proc.c
fs/nfs/pnfs.c
fs/nfs/write.c
include/linux/nfs_xdr.h

index 90b00ce..22a40c4 100644 (file)
@@ -768,11 +768,17 @@ static ssize_t nfs_direct_write_schedule_segment(struct nfs_direct_req *dreq,
                bytes = min(wsize,count);
 
                result = -ENOMEM;
-               whdr = nfs_writehdr_alloc(nfs_page_array_len(pgbase, bytes));
+               whdr = nfs_writehdr_alloc();
                if (unlikely(!whdr))
                        break;
 
-               data = &whdr->rpc_data;
+               data = nfs_writedata_alloc(&whdr->header, nfs_page_array_len(pgbase, bytes));
+               if (!data) {
+                       nfs_writehdr_free(&whdr->header);
+                       break;
+               }
+               data->header = &whdr->header;
+               atomic_inc(&data->header->refcnt);
                pages = &data->pages;
 
                down_read(&current->mm->mmap_sem);
index 33af5e5..16bc9c4 100644 (file)
@@ -319,10 +319,14 @@ extern void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor *pgio);
 extern void nfs_readdata_release(struct nfs_read_data *rdata);
 
 /* write.c */
-extern struct nfs_write_header *nfs_writehdr_alloc(unsigned int npages);
+extern void nfs_async_write_error(struct list_head *head);
+extern struct nfs_write_header *nfs_writehdr_alloc(void);
 extern void nfs_writehdr_free(struct nfs_pgio_header *hdr);
+extern struct nfs_write_data *nfs_writedata_alloc(struct nfs_pgio_header *hdr,
+                                                 unsigned int pagecount);
+extern void nfs_write_completion(struct nfs_pgio_header *hdr);
 extern int nfs_generic_flush(struct nfs_pageio_descriptor *desc,
-               struct list_head *head);
+                            struct nfs_pgio_header *hdr);
 extern void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor *pgio,
                                  struct inode *inode, int ioflags);
 extern void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio);
index 333e765..02d8170 100644 (file)
@@ -314,7 +314,6 @@ static void filelayout_write_release(void *data)
 {
        struct nfs_write_data *wdata = data;
 
-       put_lseg(wdata->header->lseg);
        wdata->header->mds_ops->rpc_release(data);
 }
 
index ce31ab2..87af80d 100644 (file)
@@ -3431,8 +3431,6 @@ void nfs4_reset_write(struct rpc_task *task, struct nfs_write_data *data)
        struct inode *inode = hdr->inode;
 
        dprintk("%s Reset task for i/o through\n", __func__);
-       put_lseg(hdr->lseg);
-       hdr->lseg        = NULL;
        data->ds_clp     = NULL;
        data->write_done_cb = nfs4_write_done_cb;
        data->args.fh       = NFS_FH(inode);
@@ -3448,7 +3446,7 @@ static void nfs4_proc_write_setup(struct nfs_write_data *data, struct rpc_messag
 {
        struct nfs_server *server = NFS_SERVER(data->header->inode);
 
-       if (data->header->lseg) {
+       if (data->ds_clp) {
                data->args.bitmask = NULL;
                data->res.fattr = NULL;
        } else
index d1a91db..d515f00 100644 (file)
@@ -1199,7 +1199,9 @@ static void pnfs_ld_handle_write_error(struct nfs_write_data *data)
                clear_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(hdr->inode)->flags);
                pnfs_return_layout(hdr->inode);
        }
-       data->task.tk_status = pnfs_write_done_resend_to_mds(hdr->inode, &hdr->pages);
+       if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
+               data->task.tk_status = pnfs_write_done_resend_to_mds(hdr->inode,
+                                                               &hdr->pages);
 }
 
 /*
@@ -1214,7 +1216,6 @@ void pnfs_ld_write_done(struct nfs_write_data *data)
                hdr->mds_ops->rpc_call_done(&data->task, data);
        } else
                pnfs_ld_handle_write_error(data);
-       put_lseg(hdr->lseg);
        hdr->mds_ops->rpc_release(data);
 }
 EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
@@ -1225,12 +1226,11 @@ pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
 {
        struct nfs_pgio_header *hdr = data->header;
 
-       list_splice_tail_init(&hdr->pages, &desc->pg_list);
-       if (hdr->req && list_empty(&hdr->req->wb_list))
-               nfs_list_add_request(hdr->req, &desc->pg_list);
-       nfs_pageio_reset_write_mds(desc);
-       desc->pg_recoalesce = 1;
-       put_lseg(hdr->lseg);
+       if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
+               list_splice_tail_init(&hdr->pages, &desc->pg_list);
+               nfs_pageio_reset_write_mds(desc);
+               desc->pg_recoalesce = 1;
+       }
        nfs_writedata_release(data);
 }
 
@@ -1246,18 +1246,12 @@ pnfs_try_to_write_data(struct nfs_write_data *wdata,
        struct nfs_server *nfss = NFS_SERVER(inode);
 
        hdr->mds_ops = call_ops;
-       hdr->lseg = get_lseg(lseg);
 
        dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
                inode->i_ino, wdata->args.count, wdata->args.offset, how);
-
        trypnfs = nfss->pnfs_curr_ld->write_pagelist(wdata, how);
-       if (trypnfs == PNFS_NOT_ATTEMPTED) {
-               put_lseg(hdr->lseg);
-               hdr->lseg = NULL;
-       } else
+       if (trypnfs != PNFS_NOT_ATTEMPTED)
                nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
-
        dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
        return trypnfs;
 }
@@ -1273,7 +1267,7 @@ pnfs_do_multiple_writes(struct nfs_pageio_descriptor *desc, struct list_head *he
        while (!list_empty(head)) {
                enum pnfs_try_status trypnfs;
 
-               data = list_entry(head->next, struct nfs_write_data, list);
+               data = list_first_entry(head, struct nfs_write_data, list);
                list_del_init(&data->list);
 
                trypnfs = pnfs_try_to_write_data(data, call_ops, lseg, how);
@@ -1283,20 +1277,40 @@ pnfs_do_multiple_writes(struct nfs_pageio_descriptor *desc, struct list_head *he
        put_lseg(lseg);
 }
 
+static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
+{
+       put_lseg(hdr->lseg);
+       nfs_writehdr_free(hdr);
+}
+
 int
 pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
 {
-       LIST_HEAD(head);
+       struct nfs_write_header *whdr;
+       struct nfs_pgio_header *hdr;
        int ret;
 
-       ret = nfs_generic_flush(desc, &head);
-       if (ret != 0) {
+       whdr = nfs_writehdr_alloc();
+       if (!whdr) {
+               nfs_async_write_error(&desc->pg_list);
                put_lseg(desc->pg_lseg);
                desc->pg_lseg = NULL;
-               return ret;
+               return -ENOMEM;
        }
-       pnfs_do_multiple_writes(desc, &head, desc->pg_ioflags);
-       return 0;
+       hdr = &whdr->header;
+       nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
+       hdr->lseg = get_lseg(desc->pg_lseg);
+       atomic_inc(&hdr->refcnt);
+       ret = nfs_generic_flush(desc, hdr);
+       if (ret != 0) {
+               put_lseg(desc->pg_lseg);
+               desc->pg_lseg = NULL;
+               set_bit(NFS_IOHDR_REDO, &hdr->flags);
+       } else
+               pnfs_do_multiple_writes(desc, &hdr->rpc_list, desc->pg_ioflags);
+       if (atomic_dec_and_test(&hdr->refcnt))
+               nfs_write_completion(hdr);
+       return ret;
 }
 EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
 
index 2efae04..076075e 100644 (file)
@@ -42,8 +42,7 @@
 static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc,
                                  struct inode *inode, int ioflags);
 static void nfs_redirty_request(struct nfs_page *req);
-static const struct rpc_call_ops nfs_write_partial_ops;
-static const struct rpc_call_ops nfs_write_full_ops;
+static const struct rpc_call_ops nfs_write_common_ops;
 static const struct rpc_call_ops nfs_commit_ops;
 
 static struct kmem_cache *nfs_wdata_cachep;
@@ -69,26 +68,47 @@ void nfs_commit_free(struct nfs_commit_data *p)
 }
 EXPORT_SYMBOL_GPL(nfs_commit_free);
 
-struct nfs_write_header *nfs_writehdr_alloc(unsigned int pagecount)
+struct nfs_write_header *nfs_writehdr_alloc(void)
 {
        struct nfs_write_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS);
 
        if (p) {
                struct nfs_pgio_header *hdr = &p->header;
-               struct nfs_write_data *data = &p->rpc_data;
 
                memset(p, 0, sizeof(*p));
                INIT_LIST_HEAD(&hdr->pages);
-               INIT_LIST_HEAD(&data->list);
-               data->header = hdr;
-               if (!nfs_pgarray_set(&data->pages, pagecount)) {
-                       mempool_free(p, nfs_wdata_mempool);
-                       p = NULL;
-               }
+               INIT_LIST_HEAD(&hdr->rpc_list);
+               spin_lock_init(&hdr->lock);
+               atomic_set(&hdr->refcnt, 0);
        }
        return p;
 }
 
+struct nfs_write_data *nfs_writedata_alloc(struct nfs_pgio_header *hdr,
+                                          unsigned int pagecount)
+{
+       struct nfs_write_data *data, *prealloc;
+
+       prealloc = &container_of(hdr, struct nfs_write_header, header)->rpc_data;
+       if (prealloc->header == NULL)
+               data = prealloc;
+       else
+               data = kzalloc(sizeof(*data), GFP_KERNEL);
+       if (!data)
+               goto out;
+
+       if (nfs_pgarray_set(&data->pages, pagecount)) {
+               data->header = hdr;
+               atomic_inc(&hdr->refcnt);
+       } else {
+               if (data != prealloc)
+                       kfree(data);
+               data = NULL;
+       }
+out:
+       return data;
+}
+
 void nfs_writehdr_free(struct nfs_pgio_header *hdr)
 {
        struct nfs_write_header *whdr = container_of(hdr, struct nfs_write_header, header);
@@ -97,10 +117,18 @@ void nfs_writehdr_free(struct nfs_pgio_header *hdr)
 
 void nfs_writedata_release(struct nfs_write_data *wdata)
 {
+       struct nfs_pgio_header *hdr = wdata->header;
+       struct nfs_write_header *write_header = container_of(hdr, struct nfs_write_header, header);
+
        put_nfs_open_context(wdata->args.context);
        if (wdata->pages.pagevec != wdata->pages.page_array)
                kfree(wdata->pages.pagevec);
-       nfs_writehdr_free(wdata->header);
+       if (wdata != &write_header->rpc_data)
+               kfree(wdata);
+       else
+               wdata->header = NULL;
+       if (atomic_dec_and_test(&hdr->refcnt))
+               nfs_write_completion(hdr);
 }
 
 static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error)
@@ -511,20 +539,6 @@ int nfs_write_need_commit(struct nfs_write_data *data)
        return data->verf.committed != NFS_FILE_SYNC;
 }
 
-static inline
-int nfs_reschedule_unstable_write(struct nfs_page *req,
-                                 struct nfs_write_data *data)
-{
-       if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) {
-               nfs_mark_request_commit(req, data->header->lseg);
-               return 1;
-       }
-       if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) {
-               nfs_mark_request_dirty(req);
-               return 1;
-       }
-       return 0;
-}
 #else
 static void
 nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg)
@@ -542,13 +556,43 @@ int nfs_write_need_commit(struct nfs_write_data *data)
        return 0;
 }
 
-static inline
-int nfs_reschedule_unstable_write(struct nfs_page *req,
-                                 struct nfs_write_data *data)
+#endif
+
+void nfs_write_completion(struct nfs_pgio_header *hdr)
 {
-       return 0;
+       unsigned long bytes = 0;
+
+       if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
+               goto out;
+       while (!list_empty(&hdr->pages)) {
+               struct nfs_page *req = nfs_list_entry(hdr->pages.next);
+               struct page *page = req->wb_page;
+
+               bytes += req->wb_bytes;
+               nfs_list_remove_request(req);
+               if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) &&
+                   (hdr->good_bytes < bytes)) {
+                       nfs_set_pageerror(page);
+                       nfs_context_set_write_error(req->wb_context, hdr->error);
+                       goto remove_req;
+               }
+               if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) {
+                       nfs_mark_request_dirty(req);
+                       goto next;
+               }
+               if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) {
+                       nfs_mark_request_commit(req, hdr->lseg);
+                       goto next;
+               }
+remove_req:
+               nfs_inode_remove_request(req);
+next:
+               nfs_unlock_request(req);
+               nfs_end_page_writeback(page);
+       }
+out:
+       hdr->release(hdr);
 }
-#endif
 
 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
 static int
@@ -813,17 +857,6 @@ int nfs_updatepage(struct file *file, struct page *page,
        return status;
 }
 
-static void nfs_writepage_release(struct nfs_page *req,
-                                 struct nfs_write_data *data)
-{
-       struct page *page = req->wb_page;
-
-       if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req, data))
-               nfs_inode_remove_request(req);
-       nfs_unlock_request(req);
-       nfs_end_page_writeback(page);
-}
-
 static int flush_task_priority(int how)
 {
        switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
@@ -890,22 +923,16 @@ EXPORT_SYMBOL_GPL(nfs_initiate_write);
 /*
  * Set up the argument/result storage required for the RPC call.
  */
-static void nfs_write_rpcsetup(struct nfs_page *req,
-               struct nfs_write_data *data,
+static void nfs_write_rpcsetup(struct nfs_write_data *data,
                unsigned int count, unsigned int offset,
                int how)
 {
-       struct nfs_pgio_header *hdr = data->header;
-       struct inode *inode = req->wb_context->dentry->d_inode;
+       struct nfs_page *req = data->header->req;
 
        /* Set up the RPC argument and reply structs
         * NB: take care not to mess about with data->commit et al. */
 
-       hdr->req = req;
-       hdr->inode = inode = req->wb_context->dentry->d_inode;
-       hdr->cred = req->wb_context->cred;
-
-       data->args.fh     = NFS_FH(inode);
+       data->args.fh     = NFS_FH(data->header->inode);
        data->args.offset = req_offset(req) + offset;
        /* pnfs_set_layoutcommit needs this */
        data->mds_offset = data->args.offset;
@@ -919,7 +946,7 @@ static void nfs_write_rpcsetup(struct nfs_page *req,
        case 0:
                break;
        case FLUSH_COND_STABLE:
-               if (nfs_need_commit(NFS_I(inode)))
+               if (nfs_need_commit(NFS_I(data->header->inode)))
                        break;
        default:
                data->args.stable = NFS_FILE_SYNC;
@@ -950,7 +977,7 @@ static int nfs_do_multiple_writes(struct list_head *head,
        while (!list_empty(head)) {
                int ret2;
 
-               data = list_entry(head->next, struct nfs_write_data, list);
+               data = list_first_entry(head, struct nfs_write_data, list);
                list_del_init(&data->list);
                
                ret2 = nfs_do_write(data, call_ops, how);
@@ -973,15 +1000,26 @@ static void nfs_redirty_request(struct nfs_page *req)
        nfs_end_page_writeback(page);
 }
 
+void nfs_async_write_error(struct list_head *head)
+{
+       struct nfs_page *req;
+
+       while (!list_empty(head)) {
+               req = nfs_list_entry(head->next);
+               nfs_list_remove_request(req);
+               nfs_redirty_request(req);
+       }
+}
+
 /*
  * Generate multiple small requests to write out a single
  * contiguous dirty area on one page.
  */
-static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, struct list_head *res)
+static int nfs_flush_multi(struct nfs_pageio_descriptor *desc,
+                          struct nfs_pgio_header *hdr)
 {
-       struct nfs_page *req = nfs_list_entry(desc->pg_list.next);
+       struct nfs_page *req = hdr->req;
        struct page *page = req->wb_page;
-       struct nfs_write_header *whdr;
        struct nfs_write_data *data;
        size_t wsize = desc->pg_bsize, nbytes;
        unsigned int offset;
@@ -989,6 +1027,7 @@ static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, struct list_head
        int ret = 0;
 
        nfs_list_remove_request(req);
+       nfs_list_add_request(req, &hdr->pages);
 
        if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
            (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit ||
@@ -1001,28 +1040,27 @@ static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, struct list_head
        do {
                size_t len = min(nbytes, wsize);
 
-               whdr = nfs_writehdr_alloc(1);
-               if (!whdr)
+               data = nfs_writedata_alloc(hdr, 1);
+               if (!data)
                        goto out_bad;
-               data = &whdr->rpc_data;
                data->pages.pagevec[0] = page;
-               nfs_write_rpcsetup(req, data, len, offset, desc->pg_ioflags);
-               list_add(&data->list, res);
+               nfs_write_rpcsetup(data, len, offset, desc->pg_ioflags);
+               list_add(&data->list, &hdr->rpc_list);
                requests++;
                nbytes -= len;
                offset += len;
        } while (nbytes != 0);
        atomic_set(&req->wb_complete, requests);
-       desc->pg_rpc_callops = &nfs_write_partial_ops;
+       desc->pg_rpc_callops = &nfs_write_common_ops;
        return ret;
 
 out_bad:
-       while (!list_empty(res)) {
-               data = list_entry(res->next, struct nfs_write_data, list);
+       while (!list_empty(&hdr->rpc_list)) {
+               data = list_first_entry(&hdr->rpc_list, struct nfs_write_data, list);
                list_del(&data->list);
                nfs_writedata_release(data);
        }
-       nfs_redirty_request(req);
+       nfs_async_write_error(&hdr->pages);
        return -ENOMEM;
 }
 
@@ -1034,64 +1072,74 @@ out_bad:
  * This is the case if nfs_updatepage detects a conflicting request
  * that has been written but not committed.
  */
-static int nfs_flush_one(struct nfs_pageio_descriptor *desc, struct list_head *res)
+static int nfs_flush_one(struct nfs_pageio_descriptor *desc,
+                        struct nfs_pgio_header *hdr)
 {
        struct nfs_page         *req;
        struct page             **pages;
-       struct nfs_write_header *whdr;
        struct nfs_write_data   *data;
        struct list_head *head = &desc->pg_list;
        int ret = 0;
 
-       whdr = nfs_writehdr_alloc(nfs_page_array_len(desc->pg_base,
-                                                    desc->pg_count));
-       if (!whdr) {
-               while (!list_empty(head)) {
-                       req = nfs_list_entry(head->next);
-                       nfs_list_remove_request(req);
-                       nfs_redirty_request(req);
-               }
+       data = nfs_writedata_alloc(hdr, nfs_page_array_len(desc->pg_base,
+                                                          desc->pg_count));
+       if (!data) {
+               nfs_async_write_error(head);
                ret = -ENOMEM;
                goto out;
        }
-       data = &whdr->rpc_data;
+
        pages = data->pages.pagevec;
        while (!list_empty(head)) {
                req = nfs_list_entry(head->next);
                nfs_list_remove_request(req);
-               nfs_list_add_request(req, &whdr->header.pages);
+               nfs_list_add_request(req, &hdr->pages);
                *pages++ = req->wb_page;
        }
-       req = nfs_list_entry(whdr->header.pages.next);
 
        if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
            (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit))
                desc->pg_ioflags &= ~FLUSH_COND_STABLE;
 
        /* Set up the argument struct */
-       nfs_write_rpcsetup(req, data, desc->pg_count, 0, desc->pg_ioflags);
-       list_add(&data->list, res);
-       desc->pg_rpc_callops = &nfs_write_full_ops;
+       nfs_write_rpcsetup(data, desc->pg_count, 0, desc->pg_ioflags);
+       list_add(&data->list, &hdr->rpc_list);
+       desc->pg_rpc_callops = &nfs_write_common_ops;
 out:
        return ret;
 }
 
-int nfs_generic_flush(struct nfs_pageio_descriptor *desc, struct list_head *head)
+int nfs_generic_flush(struct nfs_pageio_descriptor *desc,
+                     struct nfs_pgio_header *hdr)
 {
        if (desc->pg_bsize < PAGE_CACHE_SIZE)
-               return nfs_flush_multi(desc, head);
-       return nfs_flush_one(desc, head);
+               return nfs_flush_multi(desc, hdr);
+       return nfs_flush_one(desc, hdr);
 }
 
 static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
 {
-       LIST_HEAD(head);
+       struct nfs_write_header *whdr;
+       struct nfs_pgio_header *hdr;
        int ret;
 
-       ret = nfs_generic_flush(desc, &head);
+       whdr = nfs_writehdr_alloc();
+       if (!whdr) {
+               nfs_async_write_error(&desc->pg_list);
+               return -ENOMEM;
+       }
+       hdr = &whdr->header;
+       nfs_pgheader_init(desc, hdr, nfs_writehdr_free);
+       atomic_inc(&hdr->refcnt);
+       ret = nfs_generic_flush(desc, hdr);
        if (ret == 0)
-               ret = nfs_do_multiple_writes(&head, desc->pg_rpc_callops,
-                               desc->pg_ioflags);
+               ret = nfs_do_multiple_writes(&hdr->rpc_list,
+                                            desc->pg_rpc_callops,
+                                            desc->pg_ioflags);
+       else
+               set_bit(NFS_IOHDR_REDO, &hdr->flags);
+       if (atomic_dec_and_test(&hdr->refcnt))
+               nfs_write_completion(hdr);
        return ret;
 }
 
@@ -1121,62 +1169,6 @@ static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
                nfs_pageio_init_write_mds(pgio, inode, ioflags);
 }
 
-/*
- * Handle a write reply that flushed part of a page.
- */
-static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata)
-{
-       struct nfs_write_data   *data = calldata;
-
-       dprintk("NFS: %5u write(%s/%lld %d@%lld)",
-               task->tk_pid,
-               data->header->inode->i_sb->s_id,
-               (long long)
-                 NFS_FILEID(data->header->inode),
-               data->header->req->wb_bytes,
-               (long long)req_offset(data->header->req));
-
-       nfs_writeback_done(task, data);
-}
-
-static void nfs_writeback_release_partial(void *calldata)
-{
-       struct nfs_write_data   *data = calldata;
-       struct nfs_page         *req = data->header->req;
-       struct page             *page = req->wb_page;
-       int status = data->task.tk_status;
-
-       if (status < 0) {
-               nfs_set_pageerror(page);
-               nfs_context_set_write_error(req->wb_context, status);
-               dprintk(", error = %d\n", status);
-               goto out;
-       }
-
-       if (nfs_write_need_commit(data)) {
-               struct inode *inode = page->mapping->host;
-
-               spin_lock(&inode->i_lock);
-               if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) {
-                       /* Do nothing we need to resend the writes */
-               } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) {
-                       memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
-                       dprintk(" defer commit\n");
-               } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) {
-                       set_bit(PG_NEED_RESCHED, &req->wb_flags);
-                       clear_bit(PG_NEED_COMMIT, &req->wb_flags);
-                       dprintk(" server reboot detected\n");
-               }
-               spin_unlock(&inode->i_lock);
-       } else
-               dprintk(" OK\n");
-
-out:
-       if (atomic_dec_and_test(&req->wb_complete))
-               nfs_writepage_release(req, data);
-       nfs_writedata_release(data);
-}
-
 void nfs_write_prepare(struct rpc_task *task, void *calldata)
 {
        struct nfs_write_data *data = calldata;
@@ -1190,12 +1182,6 @@ void nfs_commit_prepare(struct rpc_task *task, void *calldata)
        NFS_PROTO(data->inode)->commit_rpc_prepare(task, data);
 }
 
-static const struct rpc_call_ops nfs_write_partial_ops = {
-       .rpc_call_prepare = nfs_write_prepare,
-       .rpc_call_done = nfs_writeback_done_partial,
-       .rpc_release = nfs_writeback_release_partial,
-};
-
 /*
  * Handle a write reply that flushes a whole page.
  *
@@ -1203,60 +1189,37 @@ static const struct rpc_call_ops nfs_write_partial_ops = {
  *       writebacks since the page->count is kept > 1 for as long
  *       as the page has a write request pending.
  */
-static void nfs_writeback_done_full(struct rpc_task *task, void *calldata)
+static void nfs_writeback_done_common(struct rpc_task *task, void *calldata)
 {
        struct nfs_write_data   *data = calldata;
 
        nfs_writeback_done(task, data);
 }
 
-static void nfs_writeback_release_full(void *calldata)
+static void nfs_writeback_release_common(void *calldata)
 {
        struct nfs_write_data   *data = calldata;
        struct nfs_pgio_header *hdr = data->header;
        int status = data->task.tk_status;
+       struct nfs_page *req = hdr->req;
 
-       /* Update attributes as result of writeback. */
-       while (!list_empty(&hdr->pages)) {
-               struct nfs_page *req = nfs_list_entry(hdr->pages.next);
-               struct page *page = req->wb_page;
-
-               nfs_list_remove_request(req);
-
-               dprintk("NFS: %5u write (%s/%lld %d@%lld)",
-                       data->task.tk_pid,
-                       req->wb_context->dentry->d_inode->i_sb->s_id,
-                       (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
-                       req->wb_bytes,
-                       (long long)req_offset(req));
-
-               if (status < 0) {
-                       nfs_set_pageerror(page);
-                       nfs_context_set_write_error(req->wb_context, status);
-                       dprintk(", error = %d\n", status);
-                       goto remove_request;
-               }
-
-               if (nfs_write_need_commit(data)) {
+       if ((status >= 0) && nfs_write_need_commit(data)) {
+               spin_lock(&hdr->lock);
+               if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags))
+                       ; /* Do nothing */
+               else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags))
                        memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
-                       nfs_mark_request_commit(req, hdr->lseg);
-                       dprintk(" marked for commit\n");
-                       goto next;
-               }
-               dprintk(" OK\n");
-remove_request:
-               nfs_inode_remove_request(req);
-       next:
-               nfs_unlock_request(req);
-               nfs_end_page_writeback(page);
+               else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf)))
+                       set_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags);
+               spin_unlock(&hdr->lock);
        }
        nfs_writedata_release(data);
 }
 
-static const struct rpc_call_ops nfs_write_full_ops = {
+static const struct rpc_call_ops nfs_write_common_ops = {
        .rpc_call_prepare = nfs_write_prepare,
-       .rpc_call_done = nfs_writeback_done_full,
-       .rpc_release = nfs_writeback_release_full,
+       .rpc_call_done = nfs_writeback_done_common,
+       .rpc_release = nfs_writeback_release_common,
 };
 
 
@@ -1307,38 +1270,40 @@ void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data)
                }
        }
 #endif
-       /* Is this a short write? */
-       if (task->tk_status >= 0 && resp->count < argp->count) {
+       if (task->tk_status < 0)
+               nfs_set_pgio_error(data->header, task->tk_status, argp->offset);
+       else if (resp->count < argp->count) {
                static unsigned long    complain;
 
+               /* This a short write! */
                nfs_inc_stats(inode, NFSIOS_SHORTWRITE);
 
                /* Has the server at least made some progress? */
-               if (resp->count != 0) {
-                       /* Was this an NFSv2 write or an NFSv3 stable write? */
-                       if (resp->verf->committed != NFS_UNSTABLE) {
-                               /* Resend from where the server left off */
-                               data->mds_offset += resp->count;
-                               argp->offset += resp->count;
-                               argp->pgbase += resp->count;
-                               argp->count -= resp->count;
-                       } else {
-                               /* Resend as a stable write in order to avoid
-                                * headaches in the case of a server crash.
-                                */
-                               argp->stable = NFS_FILE_SYNC;
+               if (resp->count == 0) {
+                       if (time_before(complain, jiffies)) {
+                               printk(KERN_WARNING
+                                      "NFS: Server wrote zero bytes, expected %u.\n",
+                                      argp->count);
+                               complain = jiffies + 300 * HZ;
                        }
-                       rpc_restart_call_prepare(task);
+                       nfs_set_pgio_error(data->header, -EIO, argp->offset);
+                       task->tk_status = -EIO;
                        return;
                }
-               if (time_before(complain, jiffies)) {
-                       printk(KERN_WARNING
-                              "NFS: Server wrote zero bytes, expected %u.\n",
-                                       argp->count);
-                       complain = jiffies + 300 * HZ;
+               /* Was this an NFSv2 write or an NFSv3 stable write? */
+               if (resp->verf->committed != NFS_UNSTABLE) {
+                       /* Resend from where the server left off */
+                       data->mds_offset += resp->count;
+                       argp->offset += resp->count;
+                       argp->pgbase += resp->count;
+                       argp->count -= resp->count;
+               } else {
+                       /* Resend as a stable write in order to avoid
+                        * headaches in the case of a server crash.
+                        */
+                       argp->stable = NFS_FILE_SYNC;
                }
-               /* Can't do anything about it except throw an error. */
-               task->tk_status = -EIO;
+               rpc_restart_call_prepare(task);
        }
 }
 
index 1648621..0d17db7 100644 (file)
@@ -1192,6 +1192,8 @@ enum {
        NFS_IOHDR_ERROR = 0,
        NFS_IOHDR_EOF,
        NFS_IOHDR_REDO,
+       NFS_IOHDR_NEED_COMMIT,
+       NFS_IOHDR_NEED_RESCHED,
 };
 
 struct nfs_pgio_header {