Merge tag 'nfs-rdma-for-4.1-1' of git://git.linux-nfs.org/projects/anna/nfs-rdma
authorTrond Myklebust <trond.myklebust@primarydata.com>
Thu, 23 Apr 2015 19:16:37 +0000 (15:16 -0400)
committerTrond Myklebust <trond.myklebust@primarydata.com>
Thu, 23 Apr 2015 19:16:37 +0000 (15:16 -0400)
NFS: NFSoRDMA Client Changes

This patch series creates an operation vector for each of the different
memory registration modes.  This should make it easier to one day increase
credit limit, rsize, and wsize.

Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
43 files changed:
fs/nfs/Makefile
fs/nfs/blocklayout/blocklayout.c
fs/nfs/blocklayout/dev.c
fs/nfs/callback.c
fs/nfs/client.c
fs/nfs/delegation.c
fs/nfs/dir.c
fs/nfs/direct.c
fs/nfs/file.c
fs/nfs/filelayout/filelayout.c
fs/nfs/filelayout/filelayoutdev.c
fs/nfs/flexfilelayout/flexfilelayout.c
fs/nfs/flexfilelayout/flexfilelayoutdev.c
fs/nfs/inode.c
fs/nfs/nfs42proc.c
fs/nfs/nfs42xdr.c
fs/nfs/nfs4client.c
fs/nfs/nfs4file.c
fs/nfs/nfs4idmap.c [moved from fs/nfs/idmap.c with 99% similarity]
fs/nfs/nfs4idmap.h [moved from include/linux/nfs_idmap.h with 94% similarity]
fs/nfs/nfs4proc.c
fs/nfs/nfs4state.c
fs/nfs/nfs4super.c
fs/nfs/nfs4sysctl.c
fs/nfs/nfs4xdr.c
fs/nfs/nfstrace.c
fs/nfs/objlayout/objio_osd.c
fs/nfs/pnfs.c
fs/nfs/pnfs.h
fs/nfs/pnfs_dev.c
fs/nfs/pnfs_nfs.c
fs/nfs/read.c
fs/nfs/super.c
fs/nfs/write.c
include/linux/nfs_fs.h
include/linux/nfs_xdr.h
include/linux/sunrpc/debug.h
include/uapi/linux/nfs_idmap.h
net/sunrpc/clnt.c
net/sunrpc/debugfs.c
net/sunrpc/sched.c
net/sunrpc/sunrpc_syms.c
net/sunrpc/xprt.c

index 1e987ac..8664417 100644 (file)
@@ -22,7 +22,7 @@ nfsv3-$(CONFIG_NFS_V3_ACL) += nfs3acl.o
 obj-$(CONFIG_NFS_V4) += nfsv4.o
 CFLAGS_nfs4trace.o += -I$(src)
 nfsv4-y := nfs4proc.o nfs4xdr.o nfs4state.o nfs4renewd.o nfs4super.o nfs4file.o \
-         delegation.o idmap.o callback.o callback_xdr.o callback_proc.o \
+         delegation.o nfs4idmap.o callback.o callback_xdr.o callback_proc.o \
          nfs4namespace.o nfs4getroot.o nfs4client.o nfs4session.o \
          dns_resolve.o nfs4trace.o
 nfsv4-$(CONFIG_NFS_USE_LEGACY_DNS) += cache_lib.o
index 1cac3c1..d2554fe 100644 (file)
@@ -890,6 +890,7 @@ static struct pnfs_layoutdriver_type blocklayout_type = {
        .free_deviceid_node             = bl_free_deviceid_node,
        .pg_read_ops                    = &bl_pg_read_ops,
        .pg_write_ops                   = &bl_pg_write_ops,
+       .sync                           = pnfs_generic_sync,
 };
 
 static int __init nfs4blocklayout_init(void)
index 5aed4f9..e535599 100644 (file)
@@ -33,7 +33,7 @@ bl_free_deviceid_node(struct nfs4_deviceid_node *d)
                container_of(d, struct pnfs_block_dev, node);
 
        bl_free_device(dev);
-       kfree(dev);
+       kfree_rcu(dev, node.rcu);
 }
 
 static int
index 351be92..8d129bb 100644 (file)
@@ -128,7 +128,7 @@ nfs41_callback_svc(void *vrqstp)
                if (try_to_freeze())
                        continue;
 
-               prepare_to_wait(&serv->sv_cb_waitq, &wq, TASK_UNINTERRUPTIBLE);
+               prepare_to_wait(&serv->sv_cb_waitq, &wq, TASK_INTERRUPTIBLE);
                spin_lock_bh(&serv->sv_cb_lock);
                if (!list_empty(&serv->sv_cb_list)) {
                        req = list_first_entry(&serv->sv_cb_list,
@@ -142,10 +142,10 @@ nfs41_callback_svc(void *vrqstp)
                                error);
                } else {
                        spin_unlock_bh(&serv->sv_cb_lock);
-                       /* schedule_timeout to game the hung task watchdog */
-                       schedule_timeout(60 * HZ);
+                       schedule();
                        finish_wait(&serv->sv_cb_waitq, &wq);
                }
+               flush_signals(current);
        }
        return 0;
 }
index 1987415..892aeff 100644 (file)
@@ -31,7 +31,6 @@
 #include <linux/lockd/bind.h>
 #include <linux/seq_file.h>
 #include <linux/mount.h>
-#include <linux/nfs_idmap.h>
 #include <linux/vfs.h>
 #include <linux/inet.h>
 #include <linux/in6.h>
index a6ad688..029d688 100644 (file)
@@ -378,7 +378,7 @@ int nfs_inode_set_delegation(struct inode *inode, struct rpc_cred *cred, struct
                if (freeme == NULL)
                        goto out;
        }
-       list_add_rcu(&delegation->super_list, &server->delegations);
+       list_add_tail_rcu(&delegation->super_list, &server->delegations);
        rcu_assign_pointer(nfsi->delegation, delegation);
        delegation = NULL;
 
@@ -514,7 +514,7 @@ void nfs_inode_return_delegation_noreclaim(struct inode *inode)
 
        delegation = nfs_inode_detach_delegation(inode);
        if (delegation != NULL)
-               nfs_do_return_delegation(inode, delegation, 0);
+               nfs_do_return_delegation(inode, delegation, 1);
 }
 
 /**
index c19e16f..a00ba92 100644 (file)
@@ -544,6 +544,9 @@ int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *en
        if (scratch == NULL)
                return -ENOMEM;
 
+       if (buflen == 0)
+               goto out_nopages;
+
        xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
        xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 
@@ -565,6 +568,7 @@ int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *en
                        break;
        } while (!entry->eof);
 
+out_nopages:
        if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
                array = nfs_readdir_get_array(page);
                if (!IS_ERR(array)) {
index e907c8c..eeb52b4 100644 (file)
@@ -129,22 +129,25 @@ nfs_direct_good_bytes(struct nfs_direct_req *dreq, struct nfs_pgio_header *hdr)
        int i;
        ssize_t count;
 
-       WARN_ON_ONCE(hdr->pgio_mirror_idx >= dreq->mirror_count);
-
-       count = dreq->mirrors[hdr->pgio_mirror_idx].count;
-       if (count + dreq->io_start < hdr->io_start + hdr->good_bytes) {
-               count = hdr->io_start + hdr->good_bytes - dreq->io_start;
-               dreq->mirrors[hdr->pgio_mirror_idx].count = count;
-       }
-
-       /* update the dreq->count by finding the minimum agreed count from all
-        * mirrors */
-       count = dreq->mirrors[0].count;
+       if (dreq->mirror_count == 1) {
+               dreq->mirrors[hdr->pgio_mirror_idx].count += hdr->good_bytes;
+               dreq->count += hdr->good_bytes;
+       } else {
+               /* mirrored writes */
+               count = dreq->mirrors[hdr->pgio_mirror_idx].count;
+               if (count + dreq->io_start < hdr->io_start + hdr->good_bytes) {
+                       count = hdr->io_start + hdr->good_bytes - dreq->io_start;
+                       dreq->mirrors[hdr->pgio_mirror_idx].count = count;
+               }
+               /* update the dreq->count by finding the minimum agreed count from all
+                * mirrors */
+               count = dreq->mirrors[0].count;
 
-       for (i = 1; i < dreq->mirror_count; i++)
-               count = min(count, dreq->mirrors[i].count);
+               for (i = 1; i < dreq->mirror_count; i++)
+                       count = min(count, dreq->mirrors[i].count);
 
-       dreq->count = count;
+               dreq->count = count;
+       }
 }
 
 /*
@@ -259,18 +262,11 @@ ssize_t nfs_direct_IO(int rw, struct kiocb *iocb, struct iov_iter *iter, loff_t
        if (!IS_SWAPFILE(inode))
                return 0;
 
-#ifndef CONFIG_NFS_SWAP
-       dprintk("NFS: nfs_direct_IO (%pD) off/no(%Ld/%lu) EINVAL\n",
-                       iocb->ki_filp, (long long) pos, iter->nr_segs);
-
-       return -EINVAL;
-#else
        VM_BUG_ON(iocb->ki_nbytes != PAGE_SIZE);
 
        if (rw == READ)
                return nfs_file_direct_read(iocb, iter, pos);
        return nfs_file_direct_write(iocb, iter, pos);
-#endif /* CONFIG_NFS_SWAP */
 }
 
 static void nfs_direct_release_pages(struct page **pages, unsigned int npages)
@@ -1041,6 +1037,7 @@ ssize_t nfs_file_direct_write(struct kiocb *iocb, struct iov_iter *iter,
                        if (i_size_read(inode) < iocb->ki_pos)
                                i_size_write(inode, iocb->ki_pos);
                        spin_unlock(&inode->i_lock);
+                       generic_write_sync(file, pos, result);
                }
        }
        nfs_direct_req_release(dreq);
index e679d24..28228f3 100644 (file)
@@ -281,6 +281,7 @@ nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 
        trace_nfs_fsync_enter(inode);
 
+       nfs_inode_dio_wait(inode);
        do {
                ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
                if (ret != 0)
@@ -780,7 +781,7 @@ do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
         * Flush all pending writes before doing anything
         * with locks..
         */
-       nfs_sync_mapping(filp->f_mapping);
+       vfs_fsync(filp, 0);
 
        l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
        if (!IS_ERR(l_ctx)) {
index 91e88a7..a46bf6d 100644 (file)
@@ -258,7 +258,8 @@ filelayout_set_layoutcommit(struct nfs_pgio_header *hdr)
            hdr->res.verf->committed != NFS_DATA_SYNC)
                return;
 
-       pnfs_set_layoutcommit(hdr);
+       pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
+                       hdr->mds_offset + hdr->res.count);
        dprintk("%s inode %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino,
                (unsigned long) NFS_I(hdr->inode)->layout->plh_lwb);
 }
@@ -373,7 +374,7 @@ static int filelayout_commit_done_cb(struct rpc_task *task,
        }
 
        if (data->verf.committed == NFS_UNSTABLE)
-               pnfs_commit_set_layoutcommit(data);
+               pnfs_set_layoutcommit(data->inode, data->lseg, data->lwb);
 
        return 0;
 }
@@ -1086,7 +1087,7 @@ filelayout_alloc_deviceid_node(struct nfs_server *server,
 }
 
 static void
-filelayout_free_deveiceid_node(struct nfs4_deviceid_node *d)
+filelayout_free_deviceid_node(struct nfs4_deviceid_node *d)
 {
        nfs4_fl_free_deviceid(container_of(d, struct nfs4_file_layout_dsaddr, id_node));
 }
@@ -1137,7 +1138,8 @@ static struct pnfs_layoutdriver_type filelayout_type = {
        .read_pagelist          = filelayout_read_pagelist,
        .write_pagelist         = filelayout_write_pagelist,
        .alloc_deviceid_node    = filelayout_alloc_deviceid_node,
-       .free_deviceid_node     = filelayout_free_deveiceid_node,
+       .free_deviceid_node     = filelayout_free_deviceid_node,
+       .sync                   = pnfs_nfs_generic_sync,
 };
 
 static int __init nfs4filelayout_init(void)
index 4f372e2..4946ef4 100644 (file)
@@ -55,7 +55,7 @@ nfs4_fl_free_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
                        nfs4_pnfs_ds_put(ds);
        }
        kfree(dsaddr->stripe_indices);
-       kfree(dsaddr);
+       kfree_rcu(dsaddr, id_node.rcu);
 }
 
 /* Decode opaque device data and return the result */
index 315cc68..7d05089 100644 (file)
 #include <linux/module.h>
 
 #include <linux/sunrpc/metrics.h>
-#include <linux/nfs_idmap.h>
 
 #include "flexfilelayout.h"
 #include "../nfs4session.h"
+#include "../nfs4idmap.h"
 #include "../internal.h"
 #include "../delegation.h"
 #include "../nfs4trace.h"
@@ -891,7 +891,8 @@ static int ff_layout_read_done_cb(struct rpc_task *task,
 static void
 ff_layout_set_layoutcommit(struct nfs_pgio_header *hdr)
 {
-       pnfs_set_layoutcommit(hdr);
+       pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
+                       hdr->mds_offset + hdr->res.count);
        dprintk("%s inode %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino,
                (unsigned long) NFS_I(hdr->inode)->layout->plh_lwb);
 }
@@ -1074,7 +1075,7 @@ static int ff_layout_commit_done_cb(struct rpc_task *task,
        }
 
        if (data->verf.committed == NFS_UNSTABLE)
-               pnfs_commit_set_layoutcommit(data);
+               pnfs_set_layoutcommit(data->inode, data->lseg, data->lwb);
 
        return 0;
 }
@@ -1414,7 +1415,7 @@ ff_layout_get_ds_info(struct inode *inode)
 }
 
 static void
-ff_layout_free_deveiceid_node(struct nfs4_deviceid_node *d)
+ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
 {
        nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
                                                  id_node));
@@ -1498,7 +1499,7 @@ static struct pnfs_layoutdriver_type flexfilelayout_type = {
        .pg_read_ops            = &ff_layout_pg_read_ops,
        .pg_write_ops           = &ff_layout_pg_write_ops,
        .get_ds_info            = ff_layout_get_ds_info,
-       .free_deviceid_node     = ff_layout_free_deveiceid_node,
+       .free_deviceid_node     = ff_layout_free_deviceid_node,
        .mark_request_commit    = pnfs_layout_mark_request_commit,
        .clear_request_commit   = pnfs_generic_clear_request_commit,
        .scan_commit_lists      = pnfs_generic_scan_commit_lists,
@@ -1508,6 +1509,7 @@ static struct pnfs_layoutdriver_type flexfilelayout_type = {
        .write_pagelist         = ff_layout_write_pagelist,
        .alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
        .encode_layoutreturn    = ff_layout_encode_layoutreturn,
+       .sync                   = pnfs_nfs_generic_sync,
 };
 
 static int __init nfs4flexfilelayout_init(void)
index e2c01f2..77a2d02 100644 (file)
@@ -30,7 +30,7 @@ void nfs4_ff_layout_free_deviceid(struct nfs4_ff_layout_ds *mirror_ds)
 {
        nfs4_print_deviceid(&mirror_ds->id_node.deviceid);
        nfs4_pnfs_ds_put(mirror_ds->ds);
-       kfree(mirror_ds);
+       kfree_rcu(mirror_ds, id_node.rcu);
 }
 
 /* Decode opaque device data and construct new_ds using it */
index d42dff6..8a46468 100644 (file)
@@ -133,6 +133,13 @@ void nfs_evict_inode(struct inode *inode)
        nfs_clear_inode(inode);
 }
 
+int nfs_sync_inode(struct inode *inode)
+{
+       nfs_inode_dio_wait(inode);
+       return nfs_wb_all(inode);
+}
+EXPORT_SYMBOL_GPL(nfs_sync_inode);
+
 /**
  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
  */
@@ -192,7 +199,6 @@ void nfs_zap_caches(struct inode *inode)
        nfs_zap_caches_locked(inode);
        spin_unlock(&inode->i_lock);
 }
-EXPORT_SYMBOL_GPL(nfs_zap_caches);
 
 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
 {
@@ -525,10 +531,8 @@ nfs_setattr(struct dentry *dentry, struct iattr *attr)
        trace_nfs_setattr_enter(inode);
 
        /* Write all dirty data */
-       if (S_ISREG(inode->i_mode)) {
-               nfs_inode_dio_wait(inode);
-               nfs_wb_all(inode);
-       }
+       if (S_ISREG(inode->i_mode))
+               nfs_sync_inode(inode);
 
        fattr = nfs_alloc_fattr();
        if (fattr == NULL)
@@ -644,8 +648,9 @@ int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
        trace_nfs_getattr_enter(inode);
        /* Flush out writes to the server in order to update c/mtime.  */
        if (S_ISREG(inode->i_mode)) {
-               nfs_inode_dio_wait(inode);
-               err = filemap_write_and_wait(inode->i_mapping);
+               mutex_lock(&inode->i_mutex);
+               err = nfs_sync_inode(inode);
+               mutex_unlock(&inode->i_mutex);
                if (err)
                        goto out;
        }
@@ -1588,6 +1593,19 @@ int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fa
 }
 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
 
+
+static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
+                                   struct nfs_fattr *fattr)
+{
+       bool ret1 = true, ret2 = true;
+
+       if (fattr->valid & NFS_ATTR_FATTR_FILEID)
+               ret1 = (nfsi->fileid == fattr->fileid);
+       if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
+               ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
+       return ret1 || ret2;
+}
+
 /*
  * Many nfs protocol calls return the new file attributes after
  * an operation.  Here we update the inode to reflect the state
@@ -1614,7 +1632,7 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
                        nfs_display_fhandle_hash(NFS_FH(inode)),
                        atomic_read(&inode->i_count), fattr->valid);
 
-       if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) {
+       if (!nfs_fileid_valid(nfsi, fattr)) {
                printk(KERN_ERR "NFS: server %s error: fileid changed\n"
                        "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
                        NFS_SERVER(inode)->nfs_client->cl_hostname,
@@ -1819,7 +1837,7 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
 struct inode *nfs_alloc_inode(struct super_block *sb)
 {
        struct nfs_inode *nfsi;
-       nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
+       nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
        if (!nfsi)
                return NULL;
        nfsi->flags = 0UL;
index cb17072..3a9e752 100644 (file)
@@ -36,13 +36,16 @@ static int _nfs42_proc_fallocate(struct rpc_message *msg, struct file *filep,
                                 loff_t offset, loff_t len)
 {
        struct inode *inode = file_inode(filep);
+       struct nfs_server *server = NFS_SERVER(inode);
        struct nfs42_falloc_args args = {
                .falloc_fh      = NFS_FH(inode),
                .falloc_offset  = offset,
                .falloc_length  = len,
+               .falloc_bitmask = server->cache_consistency_bitmask,
+       };
+       struct nfs42_falloc_res res = {
+               .falloc_server  = server,
        };
-       struct nfs42_falloc_res res;
-       struct nfs_server *server = NFS_SERVER(inode);
        int status;
 
        msg->rpc_argp = &args;
@@ -52,8 +55,17 @@ static int _nfs42_proc_fallocate(struct rpc_message *msg, struct file *filep,
        if (status)
                return status;
 
-       return nfs4_call_sync(server->client, server, msg,
-                             &args.seq_args, &res.seq_res, 0);
+       res.falloc_fattr = nfs_alloc_fattr();
+       if (!res.falloc_fattr)
+               return -ENOMEM;
+
+       status = nfs4_call_sync(server->client, server, msg,
+                               &args.seq_args, &res.seq_res, 0);
+       if (status == 0)
+               status = nfs_post_op_update_inode(inode, res.falloc_fattr);
+
+       kfree(res.falloc_fattr);
+       return status;
 }
 
 static int nfs42_proc_fallocate(struct rpc_message *msg, struct file *filep,
@@ -84,9 +96,13 @@ int nfs42_proc_allocate(struct file *filep, loff_t offset, loff_t len)
        if (!nfs_server_capable(inode, NFS_CAP_ALLOCATE))
                return -EOPNOTSUPP;
 
+       mutex_lock(&inode->i_mutex);
+
        err = nfs42_proc_fallocate(&msg, filep, offset, len);
        if (err == -EOPNOTSUPP)
                NFS_SERVER(inode)->caps &= ~NFS_CAP_ALLOCATE;
+
+       mutex_unlock(&inode->i_mutex);
        return err;
 }
 
@@ -101,9 +117,16 @@ int nfs42_proc_deallocate(struct file *filep, loff_t offset, loff_t len)
        if (!nfs_server_capable(inode, NFS_CAP_DEALLOCATE))
                return -EOPNOTSUPP;
 
+       nfs_wb_all(inode);
+       mutex_lock(&inode->i_mutex);
+
        err = nfs42_proc_fallocate(&msg, filep, offset, len);
+       if (err == 0)
+               truncate_pagecache_range(inode, offset, (offset + len) -1);
        if (err == -EOPNOTSUPP)
                NFS_SERVER(inode)->caps &= ~NFS_CAP_DEALLOCATE;
+
+       mutex_unlock(&inode->i_mutex);
        return err;
 }
 
index 038a7e1..1a25b27 100644 (file)
 
 #define NFS4_enc_allocate_sz           (compound_encode_hdr_maxsz + \
                                         encode_putfh_maxsz + \
-                                        encode_allocate_maxsz)
+                                        encode_allocate_maxsz + \
+                                        encode_getattr_maxsz)
 #define NFS4_dec_allocate_sz           (compound_decode_hdr_maxsz + \
                                         decode_putfh_maxsz + \
-                                        decode_allocate_maxsz)
+                                        decode_allocate_maxsz + \
+                                        decode_getattr_maxsz)
 #define NFS4_enc_deallocate_sz         (compound_encode_hdr_maxsz + \
                                         encode_putfh_maxsz + \
-                                        encode_deallocate_maxsz)
+                                        encode_deallocate_maxsz + \
+                                        encode_getattr_maxsz)
 #define NFS4_dec_deallocate_sz         (compound_decode_hdr_maxsz + \
                                         decode_putfh_maxsz + \
-                                        decode_deallocate_maxsz)
+                                        decode_deallocate_maxsz + \
+                                        decode_getattr_maxsz)
 #define NFS4_enc_seek_sz               (compound_encode_hdr_maxsz + \
                                         encode_putfh_maxsz + \
                                         encode_seek_maxsz)
@@ -92,6 +96,7 @@ static void nfs4_xdr_enc_allocate(struct rpc_rqst *req,
        encode_sequence(xdr, &args->seq_args, &hdr);
        encode_putfh(xdr, args->falloc_fh, &hdr);
        encode_allocate(xdr, args, &hdr);
+       encode_getfattr(xdr, args->falloc_bitmask, &hdr);
        encode_nops(&hdr);
 }
 
@@ -110,6 +115,7 @@ static void nfs4_xdr_enc_deallocate(struct rpc_rqst *req,
        encode_sequence(xdr, &args->seq_args, &hdr);
        encode_putfh(xdr, args->falloc_fh, &hdr);
        encode_deallocate(xdr, args, &hdr);
+       encode_getfattr(xdr, args->falloc_bitmask, &hdr);
        encode_nops(&hdr);
 }
 
@@ -183,6 +189,9 @@ static int nfs4_xdr_dec_allocate(struct rpc_rqst *rqstp,
        if (status)
                goto out;
        status = decode_allocate(xdr, res);
+       if (status)
+               goto out;
+       decode_getfattr(xdr, res->falloc_fattr, res->falloc_server);
 out:
        return status;
 }
@@ -207,6 +216,9 @@ static int nfs4_xdr_dec_deallocate(struct rpc_rqst *rqstp,
        if (status)
                goto out;
        status = decode_deallocate(xdr, res);
+       if (status)
+               goto out;
+       decode_getfattr(xdr, res->falloc_fattr, res->falloc_server);
 out:
        return status;
 }
index 86d6214..43ca5a7 100644 (file)
@@ -4,7 +4,6 @@
  */
 #include <linux/module.h>
 #include <linux/nfs_fs.h>
-#include <linux/nfs_idmap.h>
 #include <linux/nfs_mount.h>
 #include <linux/sunrpc/addr.h>
 #include <linux/sunrpc/auth.h>
@@ -15,6 +14,7 @@
 #include "callback.h"
 #include "delegation.h"
 #include "nfs4session.h"
+#include "nfs4idmap.h"
 #include "pnfs.h"
 #include "netns.h"
 
index 8b46389..cb3c787 100644 (file)
@@ -10,6 +10,8 @@
 #include "fscache.h"
 #include "pnfs.h"
 
+#include "nfstrace.h"
+
 #ifdef CONFIG_NFS_V4_2
 #include "nfs42.h"
 #endif
@@ -57,7 +59,7 @@ nfs4_file_open(struct inode *inode, struct file *filp)
        if (openflags & O_TRUNC) {
                attr.ia_valid |= ATTR_SIZE;
                attr.ia_size = 0;
-               nfs_wb_all(inode);
+               nfs_sync_inode(inode);
        }
 
        inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, &attr, &opened);
@@ -100,6 +102,9 @@ nfs4_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
        int ret;
        struct inode *inode = file_inode(file);
 
+       trace_nfs_fsync_enter(inode);
+
+       nfs_inode_dio_wait(inode);
        do {
                ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
                if (ret != 0)
@@ -107,7 +112,7 @@ nfs4_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
                mutex_lock(&inode->i_mutex);
                ret = nfs_file_fsync_commit(file, start, end, datasync);
                if (!ret)
-                       ret = pnfs_layoutcommit_inode(inode, true);
+                       ret = pnfs_sync_inode(inode, !!datasync);
                mutex_unlock(&inode->i_mutex);
                /*
                 * If nfs_file_fsync_commit detected a server reboot, then
@@ -118,6 +123,7 @@ nfs4_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
                end = LLONG_MAX;
        } while (ret == -EAGAIN);
 
+       trace_nfs_fsync_exit(inode, ret);
        return ret;
 }
 
@@ -152,15 +158,9 @@ static long nfs42_fallocate(struct file *filep, int mode, loff_t offset, loff_t
        if (ret < 0)
                return ret;
 
-       mutex_lock(&inode->i_mutex);
        if (mode & FALLOC_FL_PUNCH_HOLE)
-               ret = nfs42_proc_deallocate(filep, offset, len);
-       else
-               ret = nfs42_proc_allocate(filep, offset, len);
-       mutex_unlock(&inode->i_mutex);
-
-       nfs_zap_caches(inode);
-       return ret;
+               return nfs42_proc_deallocate(filep, offset, len);
+       return nfs42_proc_allocate(filep, offset, len);
 }
 #endif /* CONFIG_NFS_V4_2 */
 
similarity index 99%
rename from fs/nfs/idmap.c
rename to fs/nfs/nfs4idmap.c
index 857e2a9..2e1737c 100644 (file)
@@ -36,7 +36,6 @@
 #include <linux/types.h>
 #include <linux/parser.h>
 #include <linux/fs.h>
-#include <linux/nfs_idmap.h>
 #include <net/net_namespace.h>
 #include <linux/sunrpc/rpc_pipe_fs.h>
 #include <linux/nfs_fs.h>
@@ -49,6 +48,7 @@
 
 #include "internal.h"
 #include "netns.h"
+#include "nfs4idmap.h"
 #include "nfs4trace.h"
 
 #define NFS_UINT_MAXLEN 11
similarity index 94%
rename from include/linux/nfs_idmap.h
rename to fs/nfs/nfs4idmap.h
index 333844e..de44d73 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * include/linux/nfs_idmap.h
+ * fs/nfs/nfs4idmap.h
  *
  *  UID and GID to name mapping for clients.
  *
@@ -46,19 +46,8 @@ struct nfs_server;
 struct nfs_fattr;
 struct nfs4_string;
 
-#if IS_ENABLED(CONFIG_NFS_V4)
 int nfs_idmap_init(void);
 void nfs_idmap_quit(void);
-#else
-static inline int nfs_idmap_init(void)
-{
-       return 0;
-}
-
-static inline void nfs_idmap_quit(void)
-{}
-#endif
-
 int nfs_idmap_new(struct nfs_client *);
 void nfs_idmap_delete(struct nfs_client *);
 
index 627f37c..addf17d 100644 (file)
@@ -51,7 +51,6 @@
 #include <linux/namei.h>
 #include <linux/mount.h>
 #include <linux/module.h>
-#include <linux/nfs_idmap.h>
 #include <linux/xattr.h>
 #include <linux/utsname.h>
 #include <linux/freezer.h>
@@ -63,6 +62,7 @@
 #include "callback.h"
 #include "pnfs.h"
 #include "netns.h"
+#include "nfs4idmap.h"
 #include "nfs4session.h"
 #include "fscache.h"
 
@@ -185,7 +185,8 @@ const u32 nfs4_fattr_bitmap[3] = {
        | FATTR4_WORD1_SPACE_USED
        | FATTR4_WORD1_TIME_ACCESS
        | FATTR4_WORD1_TIME_METADATA
-       | FATTR4_WORD1_TIME_MODIFY,
+       | FATTR4_WORD1_TIME_MODIFY
+       | FATTR4_WORD1_MOUNTED_ON_FILEID,
 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
        FATTR4_WORD2_SECURITY_LABEL
 #endif
@@ -3095,16 +3096,13 @@ int nfs4_proc_get_rootfh(struct nfs_server *server, struct nfs_fh *fhandle,
                         struct nfs_fsinfo *info,
                         bool auth_probe)
 {
-       int status;
+       int status = 0;
 
-       switch (auth_probe) {
-       case false:
+       if (!auth_probe)
                status = nfs4_lookup_root(server, fhandle, info);
-               if (status != -NFS4ERR_WRONGSEC)
-                       break;
-       default:
+
+       if (auth_probe || status == NFS4ERR_WRONGSEC)
                status = nfs4_do_find_root_sec(server, fhandle, info);
-       }
 
        if (status == 0)
                status = nfs4_server_capabilities(server, fhandle);
@@ -7944,6 +7942,8 @@ _nfs4_proc_getdeviceinfo(struct nfs_server *server,
 {
        struct nfs4_getdeviceinfo_args args = {
                .pdev = pdev,
+               .notify_types = NOTIFY_DEVICEID4_CHANGE |
+                       NOTIFY_DEVICEID4_DELETE,
        };
        struct nfs4_getdeviceinfo_res res = {
                .pdev = pdev,
@@ -7958,6 +7958,11 @@ _nfs4_proc_getdeviceinfo(struct nfs_server *server,
 
        dprintk("--> %s\n", __func__);
        status = nfs4_call_sync(server->client, server, &msg, &args.seq_args, &res.seq_res, 0);
+       if (res.notification & ~args.notify_types)
+               dprintk("%s: unsupported notification\n", __func__);
+       if (res.notification != args.notify_types)
+               pdev->nocache = 1;
+
        dprintk("<-- %s status=%d\n", __func__, status);
 
        return status;
index f95e3b5..935a6ff 100644 (file)
@@ -42,7 +42,6 @@
 #include <linux/slab.h>
 #include <linux/fs.h>
 #include <linux/nfs_fs.h>
-#include <linux/nfs_idmap.h>
 #include <linux/kthread.h>
 #include <linux/module.h>
 #include <linux/random.h>
@@ -57,6 +56,7 @@
 #include "callback.h"
 #include "delegation.h"
 #include "internal.h"
+#include "nfs4idmap.h"
 #include "nfs4session.h"
 #include "pnfs.h"
 #include "netns.h"
index 75090fe..6fb7cb6 100644 (file)
@@ -3,12 +3,12 @@
  */
 #include <linux/init.h>
 #include <linux/module.h>
-#include <linux/nfs_idmap.h>
 #include <linux/nfs4_mount.h>
 #include <linux/nfs_fs.h>
 #include "delegation.h"
 #include "internal.h"
 #include "nfs4_fs.h"
+#include "nfs4idmap.h"
 #include "dns_resolve.h"
 #include "pnfs.h"
 #include "nfs.h"
@@ -91,10 +91,11 @@ static void nfs4_evict_inode(struct inode *inode)
 {
        truncate_inode_pages_final(&inode->i_data);
        clear_inode(inode);
-       pnfs_return_layout(inode);
-       pnfs_destroy_layout(NFS_I(inode));
        /* If we are holding a delegation, return it! */
        nfs_inode_return_delegation_noreclaim(inode);
+       /* Note that above delegreturn would trigger pnfs return-on-close */
+       pnfs_return_layout(inode);
+       pnfs_destroy_layout(NFS_I(inode));
        /* First call standard NFS clear_inode() code */
        nfs_clear_inode(inode);
 }
index b6ebe7e..0fbd3ab 100644 (file)
@@ -6,10 +6,10 @@
  * Copyright (c) 2006 Trond Myklebust <Trond.Myklebust@netapp.com>
  */
 #include <linux/sysctl.h>
-#include <linux/nfs_idmap.h>
 #include <linux/nfs_fs.h>
 
 #include "nfs4_fs.h"
+#include "nfs4idmap.h"
 #include "callback.h"
 
 static const int nfs_set_port_min = 0;
index 5c399ec..0aea978 100644 (file)
 #include <linux/nfs.h>
 #include <linux/nfs4.h>
 #include <linux/nfs_fs.h>
-#include <linux/nfs_idmap.h>
 
 #include "nfs4_fs.h"
 #include "internal.h"
+#include "nfs4idmap.h"
 #include "nfs4session.h"
 #include "pnfs.h"
 #include "netns.h"
@@ -1920,7 +1920,7 @@ encode_getdeviceinfo(struct xdr_stream *xdr,
 
        p = reserve_space(xdr, 4 + 4);
        *p++ = cpu_to_be32(1);                  /* bitmap length */
-       *p++ = cpu_to_be32(NOTIFY_DEVICEID4_CHANGE | NOTIFY_DEVICEID4_DELETE);
+       *p++ = cpu_to_be32(args->notify_types);
 }
 
 static void
@@ -5753,8 +5753,9 @@ out_overflow:
 
 #if defined(CONFIG_NFS_V4_1)
 static int decode_getdeviceinfo(struct xdr_stream *xdr,
-                               struct pnfs_device *pdev)
+                               struct nfs4_getdeviceinfo_res *res)
 {
+       struct pnfs_device *pdev = res->pdev;
        __be32 *p;
        uint32_t len, type;
        int status;
@@ -5802,12 +5803,7 @@ static int decode_getdeviceinfo(struct xdr_stream *xdr,
                if (unlikely(!p))
                        goto out_overflow;
 
-               if (be32_to_cpup(p++) &
-                   ~(NOTIFY_DEVICEID4_CHANGE | NOTIFY_DEVICEID4_DELETE)) {
-                       dprintk("%s: unsupported notification\n",
-                               __func__);
-               }
-
+               res->notification = be32_to_cpup(p++);
                for (i = 1; i < len; i++) {
                        if (be32_to_cpup(p++)) {
                                dprintk("%s: unsupported notification\n",
@@ -7061,7 +7057,7 @@ static int nfs4_xdr_dec_getdeviceinfo(struct rpc_rqst *rqstp,
        status = decode_sequence(xdr, &res->seq_res, rqstp);
        if (status != 0)
                goto out;
-       status = decode_getdeviceinfo(xdr, res->pdev);
+       status = decode_getdeviceinfo(xdr, res);
 out:
        return status;
 }
@@ -7365,6 +7361,11 @@ nfs4_stat_to_errno(int stat)
        .p_name   = #proc,                                      \
 }
 
+#define STUB(proc)             \
+[NFSPROC4_CLNT_##proc] = {     \
+       .p_name = #proc,        \
+}
+
 struct rpc_procinfo    nfs4_procedures[] = {
        PROC(READ,              enc_read,               dec_read),
        PROC(WRITE,             enc_write,              dec_write),
@@ -7417,6 +7418,7 @@ struct rpc_procinfo       nfs4_procedures[] = {
        PROC(SECINFO_NO_NAME,   enc_secinfo_no_name,    dec_secinfo_no_name),
        PROC(TEST_STATEID,      enc_test_stateid,       dec_test_stateid),
        PROC(FREE_STATEID,      enc_free_stateid,       dec_free_stateid),
+       STUB(GETDEVICELIST),
        PROC(BIND_CONN_TO_SESSION,
                        enc_bind_conn_to_session, dec_bind_conn_to_session),
        PROC(DESTROY_CLIENTID,  enc_destroy_clientid,   dec_destroy_clientid),
index 4eb0aea..c74f7af 100644 (file)
@@ -7,3 +7,6 @@
 
 #define CREATE_TRACE_POINTS
 #include "nfstrace.h"
+
+EXPORT_TRACEPOINT_SYMBOL_GPL(nfs_fsync_enter);
+EXPORT_TRACEPOINT_SYMBOL_GPL(nfs_fsync_exit);
index 24e1d74..5aaed36 100644 (file)
@@ -57,7 +57,7 @@ objio_free_deviceid_node(struct nfs4_deviceid_node *d)
 
        dprintk("%s: free od=%p\n", __func__, de->od.od);
        osduld_put_device(de->od.od);
-       kfree(de);
+       kfree_rcu(d, rcu);
 }
 
 struct objio_segment {
@@ -637,6 +637,8 @@ static struct pnfs_layoutdriver_type objlayout_type = {
        .pg_read_ops             = &objio_pg_read_ops,
        .pg_write_ops            = &objio_pg_write_ops,
 
+       .sync                    = pnfs_generic_sync,
+
        .free_deviceid_node      = objio_free_deviceid_node,
 
        .encode_layoutcommit     = objlayout_encode_layoutcommit,
index 4f802b0..2306062 100644 (file)
@@ -1090,6 +1090,7 @@ bool pnfs_roc(struct inode *ino)
        pnfs_get_layout_hdr(lo); /* matched in pnfs_roc_release */
        spin_unlock(&ino->i_lock);
        pnfs_free_lseg_list(&tmp_list);
+       pnfs_layoutcommit_inode(ino, true);
        return true;
 
 out_noroc:
@@ -1104,8 +1105,10 @@ out_noroc:
                }
        }
        spin_unlock(&ino->i_lock);
-       if (layoutreturn)
+       if (layoutreturn) {
+               pnfs_layoutcommit_inode(ino, true);
                pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY, true);
+       }
        return false;
 }
 
@@ -1841,7 +1844,8 @@ void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
 {
        trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
        if (!hdr->pnfs_error) {
-               pnfs_set_layoutcommit(hdr);
+               pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
+                               hdr->mds_offset + hdr->res.count);
                hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
        } else
                pnfs_ld_handle_write_error(hdr);
@@ -1902,7 +1906,6 @@ static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
        pnfs_put_lseg(hdr->lseg);
        nfs_pgio_header_free(hdr);
 }
-EXPORT_SYMBOL_GPL(pnfs_writehdr_free);
 
 int
 pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
@@ -2032,7 +2035,6 @@ static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
        pnfs_put_lseg(hdr->lseg);
        nfs_pgio_header_free(hdr);
 }
-EXPORT_SYMBOL_GPL(pnfs_readhdr_free);
 
 int
 pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
@@ -2099,64 +2101,34 @@ void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
 EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
 
 void
-pnfs_set_layoutcommit(struct nfs_pgio_header *hdr)
+pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
+               loff_t end_pos)
 {
-       struct inode *inode = hdr->inode;
        struct nfs_inode *nfsi = NFS_I(inode);
-       loff_t end_pos = hdr->mds_offset + hdr->res.count;
        bool mark_as_dirty = false;
 
        spin_lock(&inode->i_lock);
        if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
-               mark_as_dirty = true;
-               dprintk("%s: Set layoutcommit for inode %lu ",
-                       __func__, inode->i_ino);
-       }
-       if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &hdr->lseg->pls_flags)) {
-               /* references matched in nfs4_layoutcommit_release */
-               pnfs_get_lseg(hdr->lseg);
-       }
-       if (end_pos > nfsi->layout->plh_lwb)
                nfsi->layout->plh_lwb = end_pos;
-       spin_unlock(&inode->i_lock);
-       dprintk("%s: lseg %p end_pos %llu\n",
-               __func__, hdr->lseg, nfsi->layout->plh_lwb);
-
-       /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
-        * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
-       if (mark_as_dirty)
-               mark_inode_dirty_sync(inode);
-}
-EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
-
-void pnfs_commit_set_layoutcommit(struct nfs_commit_data *data)
-{
-       struct inode *inode = data->inode;
-       struct nfs_inode *nfsi = NFS_I(inode);
-       bool mark_as_dirty = false;
-
-       spin_lock(&inode->i_lock);
-       if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
                mark_as_dirty = true;
                dprintk("%s: Set layoutcommit for inode %lu ",
                        __func__, inode->i_ino);
-       }
-       if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &data->lseg->pls_flags)) {
+       } else if (end_pos > nfsi->layout->plh_lwb)
+               nfsi->layout->plh_lwb = end_pos;
+       if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
                /* references matched in nfs4_layoutcommit_release */
-               pnfs_get_lseg(data->lseg);
+               pnfs_get_lseg(lseg);
        }
-       if (data->lwb > nfsi->layout->plh_lwb)
-               nfsi->layout->plh_lwb = data->lwb;
        spin_unlock(&inode->i_lock);
        dprintk("%s: lseg %p end_pos %llu\n",
-               __func__, data->lseg, nfsi->layout->plh_lwb);
+               __func__, lseg, nfsi->layout->plh_lwb);
 
        /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
         * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
        if (mark_as_dirty)
                mark_inode_dirty_sync(inode);
 }
-EXPORT_SYMBOL_GPL(pnfs_commit_set_layoutcommit);
+EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
 
 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
 {
@@ -2216,7 +2188,6 @@ pnfs_layoutcommit_inode(struct inode *inode, bool sync)
        pnfs_list_write_lseg(inode, &data->lseg_list);
 
        end_pos = nfsi->layout->plh_lwb;
-       nfsi->layout->plh_lwb = 0;
 
        nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
        spin_unlock(&inode->i_lock);
@@ -2233,11 +2204,11 @@ pnfs_layoutcommit_inode(struct inode *inode, bool sync)
                status = ld->prepare_layoutcommit(&data->args);
                if (status) {
                        spin_lock(&inode->i_lock);
-                       if (end_pos < nfsi->layout->plh_lwb)
+                       set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
+                       if (end_pos > nfsi->layout->plh_lwb)
                                nfsi->layout->plh_lwb = end_pos;
                        spin_unlock(&inode->i_lock);
                        put_rpccred(data->cred);
-                       set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
                        goto clear_layoutcommitting;
                }
        }
@@ -2258,6 +2229,13 @@ clear_layoutcommitting:
 }
 EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
 
+int
+pnfs_generic_sync(struct inode *inode, bool datasync)
+{
+       return pnfs_layoutcommit_inode(inode, true);
+}
+EXPORT_SYMBOL_GPL(pnfs_generic_sync);
+
 struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
 {
        struct nfs4_threshold *thp;
index 635f086..231eb23 100644 (file)
@@ -155,6 +155,8 @@ struct pnfs_layoutdriver_type {
                               int how,
                               struct nfs_commit_info *cinfo);
 
+       int (*sync)(struct inode *inode, bool datasync);
+
        /*
         * Return PNFS_ATTEMPTED to indicate the layout code has attempted
         * I/O, else return PNFS_NOT_ATTEMPTED to fall back to normal NFS
@@ -203,6 +205,7 @@ struct pnfs_device {
        struct page **pages;
        unsigned int  pgbase;
        unsigned int  pglen;    /* reply buffer length */
+       unsigned char nocache : 1;/* May not be cached */
 };
 
 #define NFS4_PNFS_GETDEVLIST_MAXNUM 16
@@ -263,10 +266,11 @@ bool pnfs_roc(struct inode *ino);
 void pnfs_roc_release(struct inode *ino);
 void pnfs_roc_set_barrier(struct inode *ino, u32 barrier);
 bool pnfs_roc_drain(struct inode *ino, u32 *barrier, struct rpc_task *task);
-void pnfs_set_layoutcommit(struct nfs_pgio_header *);
-void pnfs_commit_set_layoutcommit(struct nfs_commit_data *data);
+void pnfs_set_layoutcommit(struct inode *, struct pnfs_layout_segment *, loff_t);
 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data);
 int pnfs_layoutcommit_inode(struct inode *inode, bool sync);
+int pnfs_generic_sync(struct inode *inode, bool datasync);
+int pnfs_nfs_generic_sync(struct inode *inode, bool datasync);
 int _pnfs_return_layout(struct inode *);
 int pnfs_commit_and_return_layout(struct inode *);
 void pnfs_ld_write_done(struct nfs_pgio_header *);
@@ -291,6 +295,7 @@ void pnfs_error_mark_layout_for_return(struct inode *inode,
 enum {
        NFS_DEVICEID_INVALID = 0,       /* set when MDS clientid recalled */
        NFS_DEVICEID_UNAVAILABLE,       /* device temporarily unavailable */
+       NFS_DEVICEID_NOCACHE,           /* device may not be cached */
 };
 
 /* pnfs_dev.c */
@@ -302,6 +307,7 @@ struct nfs4_deviceid_node {
        unsigned long                   flags;
        unsigned long                   timestamp_unavailable;
        struct nfs4_deviceid            deviceid;
+       struct rcu_head                 rcu;
        atomic_t                        ref;
 };
 
@@ -486,6 +492,14 @@ pnfs_ld_read_whole_page(struct inode *inode)
        return NFS_SERVER(inode)->pnfs_curr_ld->flags & PNFS_READ_WHOLE_PAGE;
 }
 
+static inline int
+pnfs_sync_inode(struct inode *inode, bool datasync)
+{
+       if (!pnfs_enabled_sb(NFS_SERVER(inode)))
+               return 0;
+       return NFS_SERVER(inode)->pnfs_curr_ld->sync(inode, datasync);
+}
+
 static inline bool
 pnfs_layoutcommit_outstanding(struct inode *inode)
 {
@@ -568,6 +582,12 @@ pnfs_ld_read_whole_page(struct inode *inode)
        return false;
 }
 
+static inline int
+pnfs_sync_inode(struct inode *inode, bool datasync)
+{
+       return 0;
+}
+
 static inline bool
 pnfs_roc(struct inode *ino)
 {
index aa2ec00..2961fcd 100644 (file)
@@ -149,6 +149,8 @@ nfs4_get_device_info(struct nfs_server *server,
         */
        d = server->pnfs_curr_ld->alloc_deviceid_node(server, pdev,
                        gfp_flags);
+       if (d && pdev->nocache)
+               set_bit(NFS_DEVICEID_NOCACHE, &d->flags);
 
 out_free_pages:
        for (i = 0; i < max_pages; i++)
@@ -175,8 +177,8 @@ __nfs4_find_get_deviceid(struct nfs_server *server,
        rcu_read_lock();
        d = _lookup_deviceid(server->pnfs_curr_ld, server->nfs_client, id,
                        hash);
-       if (d != NULL)
-               atomic_inc(&d->ref);
+       if (d != NULL && !atomic_inc_not_zero(&d->ref))
+               d = NULL;
        rcu_read_unlock();
        return d;
 }
@@ -235,12 +237,11 @@ nfs4_delete_deviceid(const struct pnfs_layoutdriver_type *ld,
                return;
        }
        hlist_del_init_rcu(&d->node);
+       clear_bit(NFS_DEVICEID_NOCACHE, &d->flags);
        spin_unlock(&nfs4_deviceid_lock);
-       synchronize_rcu();
 
        /* balance the initial ref set in pnfs_insert_deviceid */
-       if (atomic_dec_and_test(&d->ref))
-               d->ld->free_deviceid_node(d);
+       nfs4_put_deviceid_node(d);
 }
 EXPORT_SYMBOL_GPL(nfs4_delete_deviceid);
 
@@ -271,6 +272,11 @@ EXPORT_SYMBOL_GPL(nfs4_init_deviceid_node);
 bool
 nfs4_put_deviceid_node(struct nfs4_deviceid_node *d)
 {
+       if (test_bit(NFS_DEVICEID_NOCACHE, &d->flags)) {
+               if (atomic_add_unless(&d->ref, -1, 2))
+                       return false;
+               nfs4_delete_deviceid(d->ld, d->nfs_client, &d->deviceid);
+       }
        if (!atomic_dec_and_test(&d->ref))
                return false;
        d->ld->free_deviceid_node(d);
@@ -314,6 +320,7 @@ _deviceid_purge_client(const struct nfs_client *clp, long hash)
                if (d->nfs_client == clp && atomic_read(&d->ref)) {
                        hlist_del_init_rcu(&d->node);
                        hlist_add_head(&d->tmpnode, &tmp);
+                       clear_bit(NFS_DEVICEID_NOCACHE, &d->flags);
                }
        rcu_read_unlock();
        spin_unlock(&nfs4_deviceid_lock);
@@ -321,12 +328,10 @@ _deviceid_purge_client(const struct nfs_client *clp, long hash)
        if (hlist_empty(&tmp))
                return;
 
-       synchronize_rcu();
        while (!hlist_empty(&tmp)) {
                d = hlist_entry(tmp.first, struct nfs4_deviceid_node, tmpnode);
                hlist_del(&d->tmpnode);
-               if (atomic_dec_and_test(&d->ref))
-                       d->ld->free_deviceid_node(d);
+               nfs4_put_deviceid_node(d);
        }
 }
 
index 54e36b3..f37e25b 100644 (file)
@@ -561,7 +561,7 @@ static bool load_v3_ds_connect(void)
        return(get_v3_ds_connect != NULL);
 }
 
-void __exit nfs4_pnfs_v3_ds_connect_unload(void)
+void nfs4_pnfs_v3_ds_connect_unload(void)
 {
        if (get_v3_ds_connect) {
                symbol_put(nfs3_set_ds_client);
@@ -868,3 +868,13 @@ pnfs_layout_mark_request_commit(struct nfs_page *req,
        nfs_request_add_commit_list(req, list, cinfo);
 }
 EXPORT_SYMBOL_GPL(pnfs_layout_mark_request_commit);
+
+int
+pnfs_nfs_generic_sync(struct inode *inode, bool datasync)
+{
+       if (datasync)
+               return 0;
+       return pnfs_layoutcommit_inode(inode, true);
+}
+EXPORT_SYMBOL_GPL(pnfs_nfs_generic_sync);
+
index 568ecf0..848d8b1 100644 (file)
@@ -284,7 +284,7 @@ int nfs_readpage(struct file *file, struct page *page)
        dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
                page, PAGE_CACHE_SIZE, page_file_index(page));
        nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
-       nfs_inc_stats(inode, NFSIOS_READPAGES);
+       nfs_add_stats(inode, NFSIOS_READPAGES, 1);
 
        /*
         * Try to flush any pending writes to the file..
index 322b2de..e862093 100644 (file)
@@ -43,7 +43,6 @@
 #include <linux/seq_file.h>
 #include <linux/mount.h>
 #include <linux/namei.h>
-#include <linux/nfs_idmap.h>
 #include <linux/vfs.h>
 #include <linux/inet.h>
 #include <linux/in6.h>
@@ -2193,7 +2192,7 @@ nfs_compare_remount_data(struct nfs_server *nfss,
            data->version != nfss->nfs_client->rpc_ops->version ||
            data->minorversion != nfss->nfs_client->cl_minorversion ||
            data->retrans != nfss->client->cl_timeout->to_retries ||
-           data->selected_flavor != nfss->client->cl_auth->au_flavor ||
+           !nfs_auth_info_match(&data->auth_info, nfss->client->cl_auth->au_flavor) ||
            data->acregmin != nfss->acregmin / HZ ||
            data->acregmax != nfss->acregmax / HZ ||
            data->acdirmin != nfss->acdirmin / HZ ||
@@ -2241,7 +2240,6 @@ nfs_remount(struct super_block *sb, int *flags, char *raw_data)
        data->wsize = nfss->wsize;
        data->retrans = nfss->client->cl_timeout->to_retries;
        data->selected_flavor = nfss->client->cl_auth->au_flavor;
-       data->auth_info = nfss->auth_info;
        data->acregmin = nfss->acregmin / HZ;
        data->acregmax = nfss->acregmax / HZ;
        data->acdirmin = nfss->acdirmin / HZ;
index 849ed78..36d0c0a 100644 (file)
@@ -580,7 +580,7 @@ static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, st
        int ret;
 
        nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
-       nfs_inc_stats(inode, NFSIOS_WRITEPAGES);
+       nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1);
 
        nfs_pageio_cond_complete(pgio, page_file_index(page));
        ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE);
@@ -1840,17 +1840,16 @@ EXPORT_SYMBOL_GPL(nfs_write_inode);
  */
 int nfs_wb_all(struct inode *inode)
 {
-       struct writeback_control wbc = {
-               .sync_mode = WB_SYNC_ALL,
-               .nr_to_write = LONG_MAX,
-               .range_start = 0,
-               .range_end = LLONG_MAX,
-       };
        int ret;
 
        trace_nfs_writeback_inode_enter(inode);
 
-       ret = sync_inode(inode, &wbc);
+       ret = filemap_write_and_wait(inode->i_mapping);
+       if (!ret) {
+               ret = nfs_commit_inode(inode, FLUSH_SYNC);
+               if (!ret)
+                       pnfs_sync_inode(inode, true);
+       }
 
        trace_nfs_writeback_inode_exit(inode, ret);
        return ret;
index b01ccf3..b638eb6 100644 (file)
@@ -512,6 +512,7 @@ extern int  nfs_updatepage(struct file *, struct page *, unsigned int, unsigned
  * Try to write back everything synchronously (but check the
  * return value!)
  */
+extern int nfs_sync_inode(struct inode *inode);
 extern int nfs_wb_all(struct inode *inode);
 extern int nfs_wb_page(struct inode *inode, struct page* page);
 extern int nfs_wb_page_cancel(struct inode *inode, struct page* page);
index 4cb3eaa..93ab607 100644 (file)
@@ -255,11 +255,13 @@ struct nfs4_layoutget {
 struct nfs4_getdeviceinfo_args {
        struct nfs4_sequence_args seq_args;
        struct pnfs_device *pdev;
+       __u32 notify_types;
 };
 
 struct nfs4_getdeviceinfo_res {
        struct nfs4_sequence_res seq_res;
        struct pnfs_device *pdev;
+       __u32 notification;
 };
 
 struct nfs4_layoutcommit_args {
@@ -1271,11 +1273,15 @@ struct nfs42_falloc_args {
        nfs4_stateid                     falloc_stateid;
        u64                              falloc_offset;
        u64                              falloc_length;
+       const u32                       *falloc_bitmask;
 };
 
 struct nfs42_falloc_res {
        struct nfs4_sequence_res        seq_res;
        unsigned int                    status;
+
+       struct nfs_fattr                *falloc_fattr;
+       const struct nfs_server         *falloc_server;
 };
 
 struct nfs42_seek_args {
index c57d8ea..59a7889 100644 (file)
@@ -60,17 +60,17 @@ struct rpc_xprt;
 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 void           rpc_register_sysctl(void);
 void           rpc_unregister_sysctl(void);
-int            sunrpc_debugfs_init(void);
+void           sunrpc_debugfs_init(void);
 void           sunrpc_debugfs_exit(void);
-int            rpc_clnt_debugfs_register(struct rpc_clnt *);
+void           rpc_clnt_debugfs_register(struct rpc_clnt *);
 void           rpc_clnt_debugfs_unregister(struct rpc_clnt *);
-int            rpc_xprt_debugfs_register(struct rpc_xprt *);
+void           rpc_xprt_debugfs_register(struct rpc_xprt *);
 void           rpc_xprt_debugfs_unregister(struct rpc_xprt *);
 #else
-static inline int
+static inline void
 sunrpc_debugfs_init(void)
 {
-       return 0;
+       return;
 }
 
 static inline void
@@ -79,10 +79,10 @@ sunrpc_debugfs_exit(void)
        return;
 }
 
-static inline int
+static inline void
 rpc_clnt_debugfs_register(struct rpc_clnt *clnt)
 {
-       return 0;
+       return;
 }
 
 static inline void
@@ -91,10 +91,10 @@ rpc_clnt_debugfs_unregister(struct rpc_clnt *clnt)
        return;
 }
 
-static inline int
+static inline void
 rpc_xprt_debugfs_register(struct rpc_xprt *xprt)
 {
-       return 0;
+       return;
 }
 
 static inline void
index 8d4b1c7..038e36c 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * include/linux/nfs_idmap.h
+ * include/uapi/linux/nfs_idmap.h
  *
  *  UID and GID to name mapping for clients.
  *
index 612aa73..e6ce151 100644 (file)
@@ -303,9 +303,7 @@ static int rpc_client_register(struct rpc_clnt *clnt,
        struct super_block *pipefs_sb;
        int err;
 
-       err = rpc_clnt_debugfs_register(clnt);
-       if (err)
-               return err;
+       rpc_clnt_debugfs_register(clnt);
 
        pipefs_sb = rpc_get_sb_net(net);
        if (pipefs_sb) {
index e811f39..82962f7 100644 (file)
@@ -129,48 +129,52 @@ static const struct file_operations tasks_fops = {
        .release        = tasks_release,
 };
 
-int
+void
 rpc_clnt_debugfs_register(struct rpc_clnt *clnt)
 {
-       int len, err;
+       int len;
        char name[24]; /* enough for "../../rpc_xprt/ + 8 hex digits + NULL */
+       struct rpc_xprt *xprt;
 
        /* Already registered? */
-       if (clnt->cl_debugfs)
-               return 0;
+       if (clnt->cl_debugfs || !rpc_clnt_dir)
+               return;
 
        len = snprintf(name, sizeof(name), "%x", clnt->cl_clid);
        if (len >= sizeof(name))
-               return -EINVAL;
+               return;
 
        /* make the per-client dir */
        clnt->cl_debugfs = debugfs_create_dir(name, rpc_clnt_dir);
        if (!clnt->cl_debugfs)
-               return -ENOMEM;
+               return;
 
        /* make tasks file */
-       err = -ENOMEM;
        if (!debugfs_create_file("tasks", S_IFREG | S_IRUSR, clnt->cl_debugfs,
                                 clnt, &tasks_fops))
                goto out_err;
 
-       err = -EINVAL;
        rcu_read_lock();
+       xprt = rcu_dereference(clnt->cl_xprt);
+       /* no "debugfs" dentry? Don't bother with the symlink. */
+       if (!xprt->debugfs) {
+               rcu_read_unlock();
+               return;
+       }
        len = snprintf(name, sizeof(name), "../../rpc_xprt/%s",
-                       rcu_dereference(clnt->cl_xprt)->debugfs->d_name.name);
+                       xprt->debugfs->d_name.name);
        rcu_read_unlock();
+
        if (len >= sizeof(name))
                goto out_err;
 
-       err = -ENOMEM;
        if (!debugfs_create_symlink("xprt", clnt->cl_debugfs, name))
                goto out_err;
 
-       return 0;
+       return;
 out_err:
        debugfs_remove_recursive(clnt->cl_debugfs);
        clnt->cl_debugfs = NULL;
-       return err;
 }
 
 void
@@ -226,33 +230,33 @@ static const struct file_operations xprt_info_fops = {
        .release        = xprt_info_release,
 };
 
-int
+void
 rpc_xprt_debugfs_register(struct rpc_xprt *xprt)
 {
        int len, id;
        static atomic_t cur_id;
        char            name[9]; /* 8 hex digits + NULL term */
 
+       if (!rpc_xprt_dir)
+               return;
+
        id = (unsigned int)atomic_inc_return(&cur_id);
 
        len = snprintf(name, sizeof(name), "%x", id);
        if (len >= sizeof(name))
-               return -EINVAL;
+               return;
 
        /* make the per-client dir */
        xprt->debugfs = debugfs_create_dir(name, rpc_xprt_dir);
        if (!xprt->debugfs)
-               return -ENOMEM;
+               return;
 
        /* make tasks file */
        if (!debugfs_create_file("info", S_IFREG | S_IRUSR, xprt->debugfs,
                                 xprt, &xprt_info_fops)) {
                debugfs_remove_recursive(xprt->debugfs);
                xprt->debugfs = NULL;
-               return -ENOMEM;
        }
-
-       return 0;
 }
 
 void
@@ -266,14 +270,17 @@ void __exit
 sunrpc_debugfs_exit(void)
 {
        debugfs_remove_recursive(topdir);
+       topdir = NULL;
+       rpc_clnt_dir = NULL;
+       rpc_xprt_dir = NULL;
 }
 
-int __init
+void __init
 sunrpc_debugfs_init(void)
 {
        topdir = debugfs_create_dir("sunrpc", NULL);
        if (!topdir)
-               goto out;
+               return;
 
        rpc_clnt_dir = debugfs_create_dir("rpc_clnt", topdir);
        if (!rpc_clnt_dir)
@@ -283,10 +290,9 @@ sunrpc_debugfs_init(void)
        if (!rpc_xprt_dir)
                goto out_remove;
 
-       return 0;
+       return;
 out_remove:
        debugfs_remove_recursive(topdir);
        topdir = NULL;
-out:
-       return -ENOMEM;
+       rpc_clnt_dir = NULL;
 }
index b91fd9c..337ca85 100644 (file)
@@ -89,8 +89,8 @@ __rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
        if (!task->tk_timeout)
                return;
 
-       dprintk("RPC: %5u setting alarm for %lu ms\n",
-                       task->tk_pid, task->tk_timeout * 1000 / HZ);
+       dprintk("RPC: %5u setting alarm for %u ms\n",
+               task->tk_pid, jiffies_to_msecs(task->tk_timeout));
 
        task->u.tk_wait.expires = jiffies + task->tk_timeout;
        if (list_empty(&queue->timer_list.list) || time_before(task->u.tk_wait.expires, queue->timer_list.expires))
index e37fbed..ee5d3d2 100644 (file)
@@ -98,10 +98,7 @@ init_sunrpc(void)
        if (err)
                goto out4;
 
-       err = sunrpc_debugfs_init();
-       if (err)
-               goto out5;
-
+       sunrpc_debugfs_init();
 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
        rpc_register_sysctl();
 #endif
@@ -109,8 +106,6 @@ init_sunrpc(void)
        init_socket_xprt();     /* clnt sock transport */
        return 0;
 
-out5:
-       unregister_rpc_pipefs();
 out4:
        unregister_pernet_subsys(&sunrpc_net_ops);
 out3:
index e3015ae..1d4fe24 100644 (file)
@@ -326,6 +326,15 @@ out_unlock:
        xprt_clear_locked(xprt);
 }
 
+static void xprt_task_clear_bytes_sent(struct rpc_task *task)
+{
+       if (task != NULL) {
+               struct rpc_rqst *req = task->tk_rqstp;
+               if (req != NULL)
+                       req->rq_bytes_sent = 0;
+       }
+}
+
 /**
  * xprt_release_xprt - allow other requests to use a transport
  * @xprt: transport with other tasks potentially waiting
@@ -336,11 +345,7 @@ out_unlock:
 void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 {
        if (xprt->snd_task == task) {
-               if (task != NULL) {
-                       struct rpc_rqst *req = task->tk_rqstp;
-                       if (req != NULL)
-                               req->rq_bytes_sent = 0;
-               }
+               xprt_task_clear_bytes_sent(task);
                xprt_clear_locked(xprt);
                __xprt_lock_write_next(xprt);
        }
@@ -358,11 +363,7 @@ EXPORT_SYMBOL_GPL(xprt_release_xprt);
 void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 {
        if (xprt->snd_task == task) {
-               if (task != NULL) {
-                       struct rpc_rqst *req = task->tk_rqstp;
-                       if (req != NULL)
-                               req->rq_bytes_sent = 0;
-               }
+               xprt_task_clear_bytes_sent(task);
                xprt_clear_locked(xprt);
                __xprt_lock_write_next_cong(xprt);
        }
@@ -700,6 +701,7 @@ bool xprt_lock_connect(struct rpc_xprt *xprt,
                goto out;
        if (xprt->snd_task != task)
                goto out;
+       xprt_task_clear_bytes_sent(task);
        xprt->snd_task = cookie;
        ret = true;
 out:
@@ -1331,7 +1333,6 @@ static void xprt_init(struct rpc_xprt *xprt, struct net *net)
  */
 struct rpc_xprt *xprt_create_transport(struct xprt_create *args)
 {
-       int err;
        struct rpc_xprt *xprt;
        struct xprt_class *t;
 
@@ -1372,11 +1373,7 @@ found:
                return ERR_PTR(-ENOMEM);
        }
 
-       err = rpc_xprt_debugfs_register(xprt);
-       if (err) {
-               xprt_destroy(xprt);
-               return ERR_PTR(err);
-       }
+       rpc_xprt_debugfs_register(xprt);
 
        dprintk("RPC:       created transport %p with %u slots\n", xprt,
                        xprt->max_reqs);