Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6
[safe/jmp/linux-2.6] / fs / cifs / file.c
index 1540ada..f1ff785 100644 (file)
@@ -3,7 +3,7 @@
  *
  *   vfs operations that deal with files
  *
- *   Copyright (C) International Business Machines  Corp., 2002,2007
+ *   Copyright (C) International Business Machines  Corp., 2002,2010
  *   Author(s): Steve French (sfrench@us.ibm.com)
  *              Jeremy Allison (jra@samba.org)
  *
@@ -30,6 +30,8 @@
 #include <linux/writeback.h>
 #include <linux/task_io_accounting_ops.h>
 #include <linux/delay.h>
+#include <linux/mount.h>
+#include <linux/slab.h>
 #include <asm/div64.h>
 #include "cifsfs.h"
 #include "cifspdu.h"
 #include "cifs_debug.h"
 #include "cifs_fs_sb.h"
 
-static inline struct cifsFileInfo *cifs_init_private(
-       struct cifsFileInfo *private_data, struct inode *inode,
-       struct file *file, __u16 netfid)
-{
-       memset(private_data, 0, sizeof(struct cifsFileInfo));
-       private_data->netfid = netfid;
-       private_data->pid = current->tgid;
-       init_MUTEX(&private_data->fh_sem);
-       mutex_init(&private_data->lock_mutex);
-       INIT_LIST_HEAD(&private_data->llist);
-       private_data->pfile = file; /* needed for writepage */
-       private_data->pInode = inode;
-       private_data->invalidHandle = false;
-       private_data->closePend = false;
-       /* we have to track num writers to the inode, since writepages
-       does not tell us which handle the write is for so there can
-       be a close (overlapping with write) of the filehandle that
-       cifs_writepages chose to use */
-       atomic_set(&private_data->wrtPending, 0);
-
-       return private_data;
-}
-
 static inline int cifs_convert_flags(unsigned int flags)
 {
        if ((flags & O_ACCMODE) == O_RDONLY)
@@ -78,8 +57,38 @@ static inline int cifs_convert_flags(unsigned int flags)
        return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
                FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
                FILE_READ_DATA);
+}
 
+static inline fmode_t cifs_posix_convert_flags(unsigned int flags)
+{
+       fmode_t posix_flags = 0;
 
+       if ((flags & O_ACCMODE) == O_RDONLY)
+               posix_flags = FMODE_READ;
+       else if ((flags & O_ACCMODE) == O_WRONLY)
+               posix_flags = FMODE_WRITE;
+       else if ((flags & O_ACCMODE) == O_RDWR) {
+               /* GENERIC_ALL is too much permission to request
+                  can cause unnecessary access denied on create */
+               /* return GENERIC_ALL; */
+               posix_flags = FMODE_READ | FMODE_WRITE;
+       }
+       /* can not map O_CREAT or O_EXCL or O_TRUNC flags when
+          reopening a file.  They had their effect on the original open */
+       if (flags & O_APPEND)
+               posix_flags |= (fmode_t)O_APPEND;
+       if (flags & O_DSYNC)
+               posix_flags |= (fmode_t)O_DSYNC;
+       if (flags & __O_SYNC)
+               posix_flags |= (fmode_t)__O_SYNC;
+       if (flags & O_DIRECTORY)
+               posix_flags |= (fmode_t)O_DIRECTORY;
+       if (flags & O_NOFOLLOW)
+               posix_flags |= (fmode_t)O_NOFOLLOW;
+       if (flags & O_DIRECT)
+               posix_flags |= (fmode_t)O_DIRECT;
+
+       return posix_flags;
 }
 
 static inline int cifs_get_disposition(unsigned int flags)
@@ -97,6 +106,95 @@ static inline int cifs_get_disposition(unsigned int flags)
 }
 
 /* all arguments to this function must be checked for validity in caller */
+static inline int
+cifs_posix_open_inode_helper(struct inode *inode, struct file *file,
+                            struct cifsInodeInfo *pCifsInode, __u32 oplock,
+                            u16 netfid)
+{
+
+       write_lock(&GlobalSMBSeslock);
+
+       pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
+       if (pCifsInode == NULL) {
+               write_unlock(&GlobalSMBSeslock);
+               return -EINVAL;
+       }
+
+       if (pCifsInode->clientCanCacheRead) {
+               /* we have the inode open somewhere else
+                  no need to discard cache data */
+               goto psx_client_can_cache;
+       }
+
+       /* BB FIXME need to fix this check to move it earlier into posix_open
+          BB  fIX following section BB FIXME */
+
+       /* if not oplocked, invalidate inode pages if mtime or file
+          size changed */
+/*     temp = cifs_NTtimeToUnix(le64_to_cpu(buf->LastWriteTime));
+       if (timespec_equal(&file->f_path.dentry->d_inode->i_mtime, &temp) &&
+                          (file->f_path.dentry->d_inode->i_size ==
+                           (loff_t)le64_to_cpu(buf->EndOfFile))) {
+               cFYI(1, "inode unchanged on server");
+       } else {
+               if (file->f_path.dentry->d_inode->i_mapping) {
+                       rc = filemap_write_and_wait(file->f_path.dentry->d_inode->i_mapping);
+                       if (rc != 0)
+                               CIFS_I(file->f_path.dentry->d_inode)->write_behind_rc = rc;
+               }
+               cFYI(1, "invalidating remote inode since open detected it "
+                        "changed");
+               invalidate_remote_inode(file->f_path.dentry->d_inode);
+       } */
+
+psx_client_can_cache:
+       if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
+               pCifsInode->clientCanCacheAll = true;
+               pCifsInode->clientCanCacheRead = true;
+               cFYI(1, "Exclusive Oplock granted on inode %p",
+                        file->f_path.dentry->d_inode);
+       } else if ((oplock & 0xF) == OPLOCK_READ)
+               pCifsInode->clientCanCacheRead = true;
+
+       /* will have to change the unlock if we reenable the
+          filemap_fdatawrite (which does not seem necessary */
+       write_unlock(&GlobalSMBSeslock);
+       return 0;
+}
+
+static struct cifsFileInfo *
+cifs_fill_filedata(struct file *file)
+{
+       struct list_head *tmp;
+       struct cifsFileInfo *pCifsFile = NULL;
+       struct cifsInodeInfo *pCifsInode = NULL;
+
+       /* search inode for this file and fill in file->private_data */
+       pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
+       read_lock(&GlobalSMBSeslock);
+       list_for_each(tmp, &pCifsInode->openFileList) {
+               pCifsFile = list_entry(tmp, struct cifsFileInfo, flist);
+               if ((pCifsFile->pfile == NULL) &&
+                   (pCifsFile->pid == current->tgid)) {
+                       /* mode set in cifs_create */
+
+                       /* needed for writepage */
+                       pCifsFile->pfile = file;
+                       file->private_data = pCifsFile;
+                       break;
+               }
+       }
+       read_unlock(&GlobalSMBSeslock);
+
+       if (file->private_data != NULL) {
+               return pCifsFile;
+       } else if ((file->f_flags & O_CREAT) && (file->f_flags & O_EXCL))
+                       cERROR(1, "could not find file instance for "
+                                  "new file %p", file);
+       return NULL;
+}
+
+/* all arguments to this function must be checked for validity in caller */
 static inline int cifs_open_inode_helper(struct inode *inode, struct file *file,
        struct cifsInodeInfo *pCifsInode, struct cifsFileInfo *pCifsFile,
        struct cifsTconInfo *pTcon, int *oplock, FILE_ALL_INFO *buf,
@@ -105,17 +203,6 @@ static inline int cifs_open_inode_helper(struct inode *inode, struct file *file,
        struct timespec temp;
        int rc;
 
-       /* want handles we can use to read with first
-          in the list so we do not have to walk the
-          list to search for one in write_begin */
-       if ((file->f_flags & O_ACCMODE) == O_WRONLY) {
-               list_add_tail(&pCifsFile->flist,
-                             &pCifsInode->openFileList);
-       } else {
-               list_add(&pCifsFile->flist,
-                        &pCifsInode->openFileList);
-       }
-       write_unlock(&GlobalSMBSeslock);
        if (pCifsInode->clientCanCacheRead) {
                /* we have the inode open somewhere else
                   no need to discard cache data */
@@ -125,21 +212,21 @@ static inline int cifs_open_inode_helper(struct inode *inode, struct file *file,
        /* BB need same check in cifs_create too? */
        /* if not oplocked, invalidate inode pages if mtime or file
           size changed */
-       temp = cifs_NTtimeToUnix(le64_to_cpu(buf->LastWriteTime));
+       temp = cifs_NTtimeToUnix(buf->LastWriteTime);
        if (timespec_equal(&file->f_path.dentry->d_inode->i_mtime, &temp) &&
                           (file->f_path.dentry->d_inode->i_size ==
                            (loff_t)le64_to_cpu(buf->EndOfFile))) {
-               cFYI(1, ("inode unchanged on server"));
+               cFYI(1, "inode unchanged on server");
        } else {
                if (file->f_path.dentry->d_inode->i_mapping) {
-               /* BB no need to lock inode until after invalidate
-                  since namei code should already have it locked? */
+                       /* BB no need to lock inode until after invalidate
+                       since namei code should already have it locked? */
                        rc = filemap_write_and_wait(file->f_path.dentry->d_inode->i_mapping);
                        if (rc != 0)
                                CIFS_I(file->f_path.dentry->d_inode)->write_behind_rc = rc;
                }
-               cFYI(1, ("invalidating remote inode since open detected it "
-                        "changed"));
+               cFYI(1, "invalidating remote inode since open detected it "
+                        "changed");
                invalidate_remote_inode(file->f_path.dentry->d_inode);
        }
 
@@ -154,8 +241,8 @@ client_can_cache:
        if ((*oplock & 0xF) == OPLOCK_EXCLUSIVE) {
                pCifsInode->clientCanCacheAll = true;
                pCifsInode->clientCanCacheRead = true;
-               cFYI(1, ("Exclusive Oplock granted on inode %p",
-                        file->f_path.dentry->d_inode));
+               cFYI(1, "Exclusive Oplock granted on inode %p",
+                        file->f_path.dentry->d_inode);
        } else if ((*oplock & 0xF) == OPLOCK_READ)
                pCifsInode->clientCanCacheRead = true;
 
@@ -165,12 +252,12 @@ client_can_cache:
 int cifs_open(struct inode *inode, struct file *file)
 {
        int rc = -EACCES;
-       int xid, oplock;
+       int xid;
+       __u32 oplock;
        struct cifs_sb_info *cifs_sb;
-       struct cifsTconInfo *pTcon;
+       struct cifsTconInfo *tcon;
        struct cifsFileInfo *pCifsFile;
        struct cifsInodeInfo *pCifsInode;
-       struct list_head *tmp;
        char *full_path = NULL;
        int desiredAccess;
        int disposition;
@@ -180,46 +267,67 @@ int cifs_open(struct inode *inode, struct file *file)
        xid = GetXid();
 
        cifs_sb = CIFS_SB(inode->i_sb);
-       pTcon = cifs_sb->tcon;
+       tcon = cifs_sb->tcon;
 
-       if (file->f_flags & O_CREAT) {
-               /* search inode for this file and fill in file->private_data */
-               pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
-               read_lock(&GlobalSMBSeslock);
-               list_for_each(tmp, &pCifsInode->openFileList) {
-                       pCifsFile = list_entry(tmp, struct cifsFileInfo,
-                                              flist);
-                       if ((pCifsFile->pfile == NULL) &&
-                           (pCifsFile->pid == current->tgid)) {
-                               /* mode set in cifs_create */
-
-                               /* needed for writepage */
-                               pCifsFile->pfile = file;
-
-                               file->private_data = pCifsFile;
-                               break;
-                       }
-               }
-               read_unlock(&GlobalSMBSeslock);
-               if (file->private_data != NULL) {
-                       rc = 0;
-                       FreeXid(xid);
-                       return rc;
-               } else {
-                       if (file->f_flags & O_EXCL)
-                               cERROR(1, ("could not find file instance for "
-                                          "new file %p", file));
-               }
+       pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
+       pCifsFile = cifs_fill_filedata(file);
+       if (pCifsFile) {
+               rc = 0;
+               FreeXid(xid);
+               return rc;
        }
 
        full_path = build_path_from_dentry(file->f_path.dentry);
        if (full_path == NULL) {
+               rc = -ENOMEM;
                FreeXid(xid);
-               return -ENOMEM;
+               return rc;
+       }
+
+       cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
+                inode, file->f_flags, full_path);
+
+       if (oplockEnabled)
+               oplock = REQ_OPLOCK;
+       else
+               oplock = 0;
+
+       if (!tcon->broken_posix_open && tcon->unix_ext &&
+           (tcon->ses->capabilities & CAP_UNIX) &&
+           (CIFS_UNIX_POSIX_PATH_OPS_CAP &
+                       le64_to_cpu(tcon->fsUnixInfo.Capability))) {
+               int oflags = (int) cifs_posix_convert_flags(file->f_flags);
+               oflags |= SMB_O_CREAT;
+               /* can not refresh inode info since size could be stale */
+               rc = cifs_posix_open(full_path, &inode, file->f_path.mnt,
+                               inode->i_sb,
+                               cifs_sb->mnt_file_mode /* ignored */,
+                               oflags, &oplock, &netfid, xid);
+               if (rc == 0) {
+                       cFYI(1, "posix open succeeded");
+                       /* no need for special case handling of setting mode
+                          on read only files needed here */
+
+                       pCifsFile = cifs_fill_filedata(file);
+                       cifs_posix_open_inode_helper(inode, file, pCifsInode,
+                                                    oplock, netfid);
+                       goto out;
+               } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
+                       if (tcon->ses->serverNOS)
+                               cERROR(1, "server %s of type %s returned"
+                                          " unexpected error on SMB posix open"
+                                          ", disabling posix open support."
+                                          " Check if server update available.",
+                                          tcon->ses->serverName,
+                                          tcon->ses->serverNOS);
+                       tcon->broken_posix_open = true;
+               } else if ((rc != -EIO) && (rc != -EREMOTE) &&
+                        (rc != -EOPNOTSUPP)) /* path not found or net err */
+                       goto out;
+               /* else fallthrough to retry open the old way on network i/o
+                  or DFS errors */
        }
 
-       cFYI(1, ("inode = 0x%p file flags are 0x%x for %s",
-                inode, file->f_flags, full_path));
        desiredAccess = cifs_convert_flags(file->f_flags);
 
 /*********************************************************************
@@ -248,11 +356,6 @@ int cifs_open(struct inode *inode, struct file *file)
 
        disposition = cifs_get_disposition(file->f_flags);
 
-       if (oplockEnabled)
-               oplock = REQ_OPLOCK;
-       else
-               oplock = 0;
-
        /* BB pass O_SYNC flag through on file attributes .. BB */
 
        /* Also refresh inode by passing in file_info buf returned by SMBOpen
@@ -269,7 +372,7 @@ int cifs_open(struct inode *inode, struct file *file)
        }
 
        if (cifs_sb->tcon->ses->capabilities & CAP_NT_SMBS)
-               rc = CIFSSMBOpen(xid, pTcon, full_path, disposition,
+               rc = CIFSSMBOpen(xid, tcon, full_path, disposition,
                         desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
                         cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
                                 & CIFS_MOUNT_MAP_SPECIAL_CHR);
@@ -278,38 +381,31 @@ int cifs_open(struct inode *inode, struct file *file)
 
        if (rc == -EIO) {
                /* Old server, try legacy style OpenX */
-               rc = SMBLegacyOpen(xid, pTcon, full_path, disposition,
+               rc = SMBLegacyOpen(xid, tcon, full_path, disposition,
                        desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
                        cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
                                & CIFS_MOUNT_MAP_SPECIAL_CHR);
        }
        if (rc) {
-               cFYI(1, ("cifs_open returned 0x%x", rc));
+               cFYI(1, "cifs_open returned 0x%x", rc);
                goto out;
        }
-       file->private_data =
-               kmalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
+
+       pCifsFile = cifs_new_fileinfo(inode, netfid, file, file->f_path.mnt,
+                                       file->f_flags);
+       file->private_data = pCifsFile;
        if (file->private_data == NULL) {
                rc = -ENOMEM;
                goto out;
        }
-       pCifsFile = cifs_init_private(file->private_data, inode, file, netfid);
-       write_lock(&GlobalSMBSeslock);
-       list_add(&pCifsFile->tlist, &pTcon->openFileList);
 
-       pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
-       if (pCifsInode) {
-               rc = cifs_open_inode_helper(inode, file, pCifsInode,
-                                           pCifsFile, pTcon,
-                                           &oplock, buf, full_path, xid);
-       } else {
-               write_unlock(&GlobalSMBSeslock);
-       }
+       rc = cifs_open_inode_helper(inode, file, pCifsInode, pCifsFile, tcon,
+                                   &oplock, buf, full_path, xid);
 
        if (oplock & CIFS_CREATE_ACTION) {
                /* time to set mode which we can not set earlier due to
                   problems creating new read-only files */
-               if (pTcon->unix_ext) {
+               if (tcon->unix_ext) {
                        struct cifs_unix_set_info_args args = {
                                .mode   = inode->i_mode,
                                .uid    = NO_CHANGE_64,
@@ -319,9 +415,9 @@ int cifs_open(struct inode *inode, struct file *file)
                                .mtime  = NO_CHANGE_64,
                                .device = 0,
                        };
-                       CIFSSMBUnixSetInfo(xid, pTcon, full_path, &args,
-                                           cifs_sb->local_nls,
-                                           cifs_sb->mnt_cifs_flags &
+                       CIFSSMBUnixSetPathInfo(xid, tcon, full_path, &args,
+                                              cifs_sb->local_nls,
+                                              cifs_sb->mnt_cifs_flags &
                                                CIFS_MOUNT_MAP_SPECIAL_CHR);
                }
        }
@@ -347,9 +443,10 @@ static int cifs_relock_file(struct cifsFileInfo *cifsFile)
 static int cifs_reopen_file(struct file *file, bool can_flush)
 {
        int rc = -EACCES;
-       int xid, oplock;
+       int xid;
+       __u32 oplock;
        struct cifs_sb_info *cifs_sb;
-       struct cifsTconInfo *pTcon;
+       struct cifsTconInfo *tcon;
        struct cifsFileInfo *pCifsFile;
        struct cifsInodeInfo *pCifsInode;
        struct inode *inode;
@@ -364,15 +461,16 @@ static int cifs_reopen_file(struct file *file, bool can_flush)
                return -EBADF;
 
        xid = GetXid();
-       down(&pCifsFile->fh_sem);
+       mutex_lock(&pCifsFile->fh_mutex);
        if (!pCifsFile->invalidHandle) {
-               up(&pCifsFile->fh_sem);
+               mutex_unlock(&pCifsFile->fh_mutex);
+               rc = 0;
                FreeXid(xid);
-               return 0;
+               return rc;
        }
 
        if (file->f_path.dentry == NULL) {
-               cERROR(1, ("no valid name if dentry freed"));
+               cERROR(1, "no valid name if dentry freed");
                dump_stack();
                rc = -EBADF;
                goto reopen_error_exit;
@@ -380,14 +478,14 @@ static int cifs_reopen_file(struct file *file, bool can_flush)
 
        inode = file->f_path.dentry->d_inode;
        if (inode == NULL) {
-               cERROR(1, ("inode not valid"));
+               cERROR(1, "inode not valid");
                dump_stack();
                rc = -EBADF;
                goto reopen_error_exit;
        }
 
        cifs_sb = CIFS_SB(inode->i_sb);
-       pTcon = cifs_sb->tcon;
+       tcon = cifs_sb->tcon;
 
 /* can not grab rename sem here because various ops, including
    those that already have the rename sem can end up causing writepage
@@ -397,38 +495,57 @@ static int cifs_reopen_file(struct file *file, bool can_flush)
        if (full_path == NULL) {
                rc = -ENOMEM;
 reopen_error_exit:
-               up(&pCifsFile->fh_sem);
+               mutex_unlock(&pCifsFile->fh_mutex);
                FreeXid(xid);
                return rc;
        }
 
-       cFYI(1, ("inode = 0x%p file flags 0x%x for %s",
-                inode, file->f_flags, full_path));
-       desiredAccess = cifs_convert_flags(file->f_flags);
+       cFYI(1, "inode = 0x%p file flags 0x%x for %s",
+                inode, file->f_flags, full_path);
 
        if (oplockEnabled)
                oplock = REQ_OPLOCK;
        else
                oplock = 0;
 
+       if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
+           (CIFS_UNIX_POSIX_PATH_OPS_CAP &
+                       le64_to_cpu(tcon->fsUnixInfo.Capability))) {
+               int oflags = (int) cifs_posix_convert_flags(file->f_flags);
+               /* can not refresh inode info since size could be stale */
+               rc = cifs_posix_open(full_path, NULL, file->f_path.mnt,
+                               inode->i_sb,
+                               cifs_sb->mnt_file_mode /* ignored */,
+                               oflags, &oplock, &netfid, xid);
+               if (rc == 0) {
+                       cFYI(1, "posix reopen succeeded");
+                       goto reopen_success;
+               }
+               /* fallthrough to retry open the old way on errors, especially
+                  in the reconnect path it is important to retry hard */
+       }
+
+       desiredAccess = cifs_convert_flags(file->f_flags);
+
        /* Can not refresh inode by passing in file_info buf to be returned
           by SMBOpen and then calling get_inode_info with returned buf
           since file might have write behind data that needs to be flushed
           and server version of file size can be stale. If we knew for sure
           that inode was not dirty locally we could do this */
 
-       rc = CIFSSMBOpen(xid, pTcon, full_path, disposition, desiredAccess,
+       rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
                         CREATE_NOT_DIR, &netfid, &oplock, NULL,
                         cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
                                CIFS_MOUNT_MAP_SPECIAL_CHR);
        if (rc) {
-               up(&pCifsFile->fh_sem);
-               cFYI(1, ("cifs_open returned 0x%x", rc));
-               cFYI(1, ("oplock: %d", oplock));
+               mutex_unlock(&pCifsFile->fh_mutex);
+               cFYI(1, "cifs_open returned 0x%x", rc);
+               cFYI(1, "oplock: %d", oplock);
        } else {
+reopen_success:
                pCifsFile->netfid = netfid;
                pCifsFile->invalidHandle = false;
-               up(&pCifsFile->fh_sem);
+               mutex_unlock(&pCifsFile->fh_mutex);
                pCifsInode = CIFS_I(inode);
                if (pCifsInode) {
                        if (can_flush) {
@@ -439,7 +556,7 @@ reopen_error_exit:
                           go to server to get inode info */
                                pCifsInode->clientCanCacheAll = false;
                                pCifsInode->clientCanCacheRead = false;
-                               if (pTcon->unix_ext)
+                               if (tcon->unix_ext)
                                        rc = cifs_get_inode_info_unix(&inode,
                                                full_path, inode->i_sb, xid);
                                else
@@ -455,8 +572,8 @@ reopen_error_exit:
                        if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
                                pCifsInode->clientCanCacheAll = true;
                                pCifsInode->clientCanCacheRead = true;
-                               cFYI(1, ("Exclusive Oplock granted on inode %p",
-                                        file->f_path.dentry->d_inode));
+                               cFYI(1, "Exclusive Oplock granted on inode %p",
+                                        file->f_path.dentry->d_inode);
                        } else if ((oplock & 0xF) == OPLOCK_READ) {
                                pCifsInode->clientCanCacheRead = true;
                                pCifsInode->clientCanCacheAll = false;
@@ -467,7 +584,6 @@ reopen_error_exit:
                        cifs_relock_file(pCifsFile);
                }
        }
-
        kfree(full_path);
        FreeXid(xid);
        return rc;
@@ -488,14 +604,15 @@ int cifs_close(struct inode *inode, struct file *file)
        pTcon = cifs_sb->tcon;
        if (pSMBFile) {
                struct cifsLockInfo *li, *tmp;
-
+               write_lock(&GlobalSMBSeslock);
                pSMBFile->closePend = true;
                if (pTcon) {
                        /* no sense reconnecting to close a file that is
                           already closed */
                        if (!pTcon->need_reconnect) {
+                               write_unlock(&GlobalSMBSeslock);
                                timeout = 2;
-                               while ((atomic_read(&pSMBFile->wrtPending) != 0)
+                               while ((atomic_read(&pSMBFile->count) != 1)
                                        && (timeout <= 2048)) {
                                        /* Give write a better chance to get to
                                        server ahead of the close.  We do not
@@ -504,18 +621,18 @@ int cifs_close(struct inode *inode, struct file *file)
                                        the struct would be in each open file,
                                        but this should give enough time to
                                        clear the socket */
-                                       cFYI(DBG2,
-                                               ("close delay, write pending"));
+                                       cFYI(DBG2, "close delay, write pending");
                                        msleep(timeout);
                                        timeout *= 4;
                                }
-                               if (atomic_read(&pSMBFile->wrtPending))
-                                       cERROR(1,
-                                               ("close with pending writes"));
-                               rc = CIFSSMBClose(xid, pTcon,
+                               if (!pTcon->need_reconnect &&
+                                   !pSMBFile->invalidHandle)
+                                       rc = CIFSSMBClose(xid, pTcon,
                                                  pSMBFile->netfid);
-                       }
-               }
+                       } else
+                               write_unlock(&GlobalSMBSeslock);
+               } else
+                       write_unlock(&GlobalSMBSeslock);
 
                /* Delete any outstanding lock records.
                   We'll lose them when the file is closed anyway. */
@@ -530,31 +647,14 @@ int cifs_close(struct inode *inode, struct file *file)
                list_del(&pSMBFile->flist);
                list_del(&pSMBFile->tlist);
                write_unlock(&GlobalSMBSeslock);
-               timeout = 10;
-               /* We waited above to give the SMBWrite a chance to issue
-                  on the wire (so we do not get SMBWrite returning EBADF
-                  if writepages is racing with close.  Note that writepages
-                  does not specify a file handle, so it is possible for a file
-                  to be opened twice, and the application close the "wrong"
-                  file handle - in these cases we delay long enough to allow
-                  the SMBWrite to get on the wire before the SMB Close.
-                  We allow total wait here over 45 seconds, more than
-                  oplock break time, and more than enough to allow any write
-                  to complete on the server, or to time out on the client */
-               while ((atomic_read(&pSMBFile->wrtPending) != 0)
-                               && (timeout <= 50000)) {
-                       cERROR(1, ("writes pending, delay free of handle"));
-                       msleep(timeout);
-                       timeout *= 8;
-               }
-               kfree(file->private_data);
+               cifsFileInfo_put(file->private_data);
                file->private_data = NULL;
        } else
                rc = -EBADF;
 
        read_lock(&GlobalSMBSeslock);
        if (list_empty(&(CIFS_I(inode)->openFileList))) {
-               cFYI(1, ("closing last open instance for inode %p", inode));
+               cFYI(1, "closing last open instance for inode %p", inode);
                /* if the file is not open we do not know if we can cache info
                   on this inode, much less write behind and read ahead */
                CIFS_I(inode)->clientCanCacheRead = false;
@@ -575,7 +675,7 @@ int cifs_closedir(struct inode *inode, struct file *file)
            (struct cifsFileInfo *)file->private_data;
        char *ptmp;
 
-       cFYI(1, ("Closedir inode = 0x%p", inode));
+       cFYI(1, "Closedir inode = 0x%p", inode);
 
        xid = GetXid();
 
@@ -586,19 +686,22 @@ int cifs_closedir(struct inode *inode, struct file *file)
 
                pTcon = cifs_sb->tcon;
 
-               cFYI(1, ("Freeing private data in close dir"));
+               cFYI(1, "Freeing private data in close dir");
+               write_lock(&GlobalSMBSeslock);
                if (!pCFileStruct->srch_inf.endOfSearch &&
                    !pCFileStruct->invalidHandle) {
                        pCFileStruct->invalidHandle = true;
+                       write_unlock(&GlobalSMBSeslock);
                        rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
-                       cFYI(1, ("Closing uncompleted readdir with rc %d",
-                                rc));
+                       cFYI(1, "Closing uncompleted readdir with rc %d",
+                                rc);
                        /* not much we can do if it fails anyway, ignore rc */
                        rc = 0;
-               }
+               } else
+                       write_unlock(&GlobalSMBSeslock);
                ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
                if (ptmp) {
-                       cFYI(1, ("closedir free smb buf in srch struct"));
+                       cFYI(1, "closedir free smb buf in srch struct");
                        pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
                        if (pCFileStruct->srch_inf.smallBuf)
                                cifs_small_buf_release(ptmp);
@@ -637,71 +740,73 @@ int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
        __u64 length;
        bool wait_flag = false;
        struct cifs_sb_info *cifs_sb;
-       struct cifsTconInfo *pTcon;
+       struct cifsTconInfo *tcon;
        __u16 netfid;
        __u8 lockType = LOCKING_ANDX_LARGE_FILES;
-       bool posix_locking;
+       bool posix_locking = 0;
 
        length = 1 + pfLock->fl_end - pfLock->fl_start;
        rc = -EACCES;
        xid = GetXid();
 
-       cFYI(1, ("Lock parm: 0x%x flockflags: "
+       cFYI(1, "Lock parm: 0x%x flockflags: "
                 "0x%x flocktype: 0x%x start: %lld end: %lld",
                cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
-               pfLock->fl_end));
+               pfLock->fl_end);
 
        if (pfLock->fl_flags & FL_POSIX)
-               cFYI(1, ("Posix"));
+               cFYI(1, "Posix");
        if (pfLock->fl_flags & FL_FLOCK)
-               cFYI(1, ("Flock"));
+               cFYI(1, "Flock");
        if (pfLock->fl_flags & FL_SLEEP) {
-               cFYI(1, ("Blocking lock"));
+               cFYI(1, "Blocking lock");
                wait_flag = true;
        }
        if (pfLock->fl_flags & FL_ACCESS)
-               cFYI(1, ("Process suspended by mandatory locking - "
-                        "not implemented yet"));
+               cFYI(1, "Process suspended by mandatory locking - "
+                        "not implemented yet");
        if (pfLock->fl_flags & FL_LEASE)
-               cFYI(1, ("Lease on file - not implemented yet"));
+               cFYI(1, "Lease on file - not implemented yet");
        if (pfLock->fl_flags &
            (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
-               cFYI(1, ("Unknown lock flags 0x%x", pfLock->fl_flags));
+               cFYI(1, "Unknown lock flags 0x%x", pfLock->fl_flags);
 
        if (pfLock->fl_type == F_WRLCK) {
-               cFYI(1, ("F_WRLCK "));
+               cFYI(1, "F_WRLCK ");
                numLock = 1;
        } else if (pfLock->fl_type == F_UNLCK) {
-               cFYI(1, ("F_UNLCK"));
+               cFYI(1, "F_UNLCK");
                numUnlock = 1;
                /* Check if unlock includes more than
                one lock range */
        } else if (pfLock->fl_type == F_RDLCK) {
-               cFYI(1, ("F_RDLCK"));
+               cFYI(1, "F_RDLCK");
                lockType |= LOCKING_ANDX_SHARED_LOCK;
                numLock = 1;
        } else if (pfLock->fl_type == F_EXLCK) {
-               cFYI(1, ("F_EXLCK"));
+               cFYI(1, "F_EXLCK");
                numLock = 1;
        } else if (pfLock->fl_type == F_SHLCK) {
-               cFYI(1, ("F_SHLCK"));
+               cFYI(1, "F_SHLCK");
                lockType |= LOCKING_ANDX_SHARED_LOCK;
                numLock = 1;
        } else
-               cFYI(1, ("Unknown type of lock"));
+               cFYI(1, "Unknown type of lock");
 
        cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
-       pTcon = cifs_sb->tcon;
+       tcon = cifs_sb->tcon;
 
        if (file->private_data == NULL) {
+               rc = -EBADF;
                FreeXid(xid);
-               return -EBADF;
+               return rc;
        }
        netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
 
-       posix_locking = (cifs_sb->tcon->ses->capabilities & CAP_UNIX) &&
-                       (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(cifs_sb->tcon->fsUnixInfo.Capability));
-
+       if ((tcon->ses->capabilities & CAP_UNIX) &&
+           (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
+           ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
+               posix_locking = 1;
        /* BB add code here to normalize offset and length to
        account for negative length which we can not accept over the
        wire */
@@ -712,7 +817,7 @@ int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
                                posix_lock_type = CIFS_RDLCK;
                        else
                                posix_lock_type = CIFS_WRLCK;
-                       rc = CIFSSMBPosixLock(xid, pTcon, netfid, 1 /* get */,
+                       rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
                                        length, pfLock,
                                        posix_lock_type, wait_flag);
                        FreeXid(xid);
@@ -720,23 +825,47 @@ int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
                }
 
                /* BB we could chain these into one lock request BB */
-               rc = CIFSSMBLock(xid, pTcon, netfid, length, pfLock->fl_start,
+               rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
                                 0, 1, lockType, 0 /* wait flag */ );
                if (rc == 0) {
-                       rc = CIFSSMBLock(xid, pTcon, netfid, length,
+                       rc = CIFSSMBLock(xid, tcon, netfid, length,
                                         pfLock->fl_start, 1 /* numUnlock */ ,
                                         0 /* numLock */ , lockType,
                                         0 /* wait flag */ );
                        pfLock->fl_type = F_UNLCK;
                        if (rc != 0)
-                               cERROR(1, ("Error unlocking previously locked "
-                                          "range %d during test of lock", rc));
+                               cERROR(1, "Error unlocking previously locked "
+                                          "range %d during test of lock", rc);
                        rc = 0;
 
                } else {
                        /* if rc == ERR_SHARING_VIOLATION ? */
-                       rc = 0; /* do not change lock type to unlock
-                                  since range in use */
+                       rc = 0;
+
+                       if (lockType & LOCKING_ANDX_SHARED_LOCK) {
+                               pfLock->fl_type = F_WRLCK;
+                       } else {
+                               rc = CIFSSMBLock(xid, tcon, netfid, length,
+                                       pfLock->fl_start, 0, 1,
+                                       lockType | LOCKING_ANDX_SHARED_LOCK,
+                                       0 /* wait flag */);
+                               if (rc == 0) {
+                                       rc = CIFSSMBLock(xid, tcon, netfid,
+                                               length, pfLock->fl_start, 1, 0,
+                                               lockType |
+                                               LOCKING_ANDX_SHARED_LOCK,
+                                               0 /* wait flag */);
+                                       pfLock->fl_type = F_RDLCK;
+                                       if (rc != 0)
+                                               cERROR(1, "Error unlocking "
+                                               "previously locked range %d "
+                                               "during test of lock", rc);
+                                       rc = 0;
+                               } else {
+                                       pfLock->fl_type = F_WRLCK;
+                                       rc = 0;
+                               }
+                       }
                }
 
                FreeXid(xid);
@@ -760,7 +889,7 @@ int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
                if (numUnlock == 1)
                        posix_lock_type = CIFS_UNLCK;
 
-               rc = CIFSSMBPosixLock(xid, pTcon, netfid, 0 /* set */,
+               rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
                                      length, pfLock,
                                      posix_lock_type, wait_flag);
        } else {
@@ -768,7 +897,7 @@ int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
                        (struct cifsFileInfo *)file->private_data;
 
                if (numLock) {
-                       rc = CIFSSMBLock(xid, pTcon, netfid, length,
+                       rc = CIFSSMBLock(xid, tcon, netfid, length,
                                        pfLock->fl_start,
                                        0, numLock, lockType, wait_flag);
 
@@ -789,15 +918,16 @@ int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
                                if (pfLock->fl_start <= li->offset &&
                                                (pfLock->fl_start + length) >=
                                                (li->offset + li->length)) {
-                                       stored_rc = CIFSSMBLock(xid, pTcon,
+                                       stored_rc = CIFSSMBLock(xid, tcon,
                                                        netfid,
                                                        li->length, li->offset,
                                                        1, 0, li->type, false);
                                        if (stored_rc)
                                                rc = stored_rc;
-
-                                       list_del(&li->llist);
-                                       kfree(li);
+                                       else {
+                                               list_del(&li->llist);
+                                               kfree(li);
+                                       }
                                }
                        }
                        mutex_unlock(&fid->lock_mutex);
@@ -810,6 +940,40 @@ int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
        return rc;
 }
 
+/*
+ * Set the timeout on write requests past EOF. For some servers (Windows)
+ * these calls can be very long.
+ *
+ * If we're writing >10M past the EOF we give a 180s timeout. Anything less
+ * than that gets a 45s timeout. Writes not past EOF get 15s timeouts.
+ * The 10M cutoff is totally arbitrary. A better scheme for this would be
+ * welcome if someone wants to suggest one.
+ *
+ * We may be able to do a better job with this if there were some way to
+ * declare that a file should be sparse.
+ */
+static int
+cifs_write_timeout(struct cifsInodeInfo *cifsi, loff_t offset)
+{
+       if (offset <= cifsi->server_eof)
+               return CIFS_STD_OP;
+       else if (offset > (cifsi->server_eof + (10 * 1024 * 1024)))
+               return CIFS_VLONG_OP;
+       else
+               return CIFS_LONG_OP;
+}
+
+/* update the file size (if needed) after a write */
+static void
+cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
+                     unsigned int bytes_written)
+{
+       loff_t end_of_write = offset + bytes_written;
+
+       if (end_of_write > cifsi->server_eof)
+               cifsi->server_eof = end_of_write;
+}
+
 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
        size_t write_size, loff_t *poffset)
 {
@@ -820,14 +984,14 @@ ssize_t cifs_user_write(struct file *file, const char __user *write_data,
        struct cifsTconInfo *pTcon;
        int xid, long_op;
        struct cifsFileInfo *open_file;
+       struct cifsInodeInfo *cifsi = CIFS_I(file->f_path.dentry->d_inode);
 
        cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
 
        pTcon = cifs_sb->tcon;
 
-       /* cFYI(1,
-          (" write %d bytes to offset %lld of %s", write_size,
-          *poffset, file->f_path.dentry->d_name.name)); */
+       /* cFYI(1, " write %d bytes to offset %lld of %s", write_size,
+          *poffset, file->f_path.dentry->d_name.name); */
 
        if (file->private_data == NULL)
                return -EBADF;
@@ -839,11 +1003,7 @@ ssize_t cifs_user_write(struct file *file, const char __user *write_data,
 
        xid = GetXid();
 
-       if (*poffset > file->f_path.dentry->d_inode->i_size)
-               long_op = CIFS_VLONG_OP; /* writes past EOF take long time */
-       else
-               long_op = CIFS_LONG_OP;
-
+       long_op = cifs_write_timeout(cifsi, *poffset);
        for (total_written = 0; write_size > total_written;
             total_written += bytes_written) {
                rc = -EAGAIN;
@@ -887,8 +1047,10 @@ ssize_t cifs_user_write(struct file *file, const char __user *write_data,
                                FreeXid(xid);
                                return rc;
                        }
-               } else
+               } else {
+                       cifs_update_eof(cifsi, *poffset, bytes_written);
                        *poffset += bytes_written;
+               }
                long_op = CIFS_STD_OP; /* subsequent writes fast -
                                    15 seconds is plenty */
        }
@@ -924,13 +1086,14 @@ static ssize_t cifs_write(struct file *file, const char *write_data,
        struct cifsTconInfo *pTcon;
        int xid, long_op;
        struct cifsFileInfo *open_file;
+       struct cifsInodeInfo *cifsi = CIFS_I(file->f_path.dentry->d_inode);
 
        cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
 
        pTcon = cifs_sb->tcon;
 
-       cFYI(1, ("write %zd bytes to offset %lld of %s", write_size,
-          *poffset, file->f_path.dentry->d_name.name));
+       cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
+          *poffset, file->f_path.dentry->d_name.name);
 
        if (file->private_data == NULL)
                return -EBADF;
@@ -938,11 +1101,7 @@ static ssize_t cifs_write(struct file *file, const char *write_data,
 
        xid = GetXid();
 
-       if (*poffset > file->f_path.dentry->d_inode->i_size)
-               long_op = CIFS_VLONG_OP; /* writes past EOF can be slow */
-       else
-               long_op = CIFS_LONG_OP;
-
+       long_op = cifs_write_timeout(cifsi, *poffset);
        for (total_written = 0; write_size > total_written;
             total_written += bytes_written) {
                rc = -EAGAIN;
@@ -1005,8 +1164,10 @@ static ssize_t cifs_write(struct file *file, const char *write_data,
                                FreeXid(xid);
                                return rc;
                        }
-               } else
+               } else {
+                       cifs_update_eof(cifsi, *poffset, bytes_written);
                        *poffset += bytes_written;
+               }
                long_op = CIFS_STD_OP; /* subsequent writes fast -
                                    15 seconds is plenty */
        }
@@ -1048,7 +1209,7 @@ struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode)
                        if (!open_file->invalidHandle) {
                                /* found a good file */
                                /* lock it so it will not be closed on us */
-                               atomic_inc(&open_file->wrtPending);
+                               cifsFileInfo_get(open_file);
                                read_unlock(&GlobalSMBSeslock);
                                return open_file;
                        } /* else might as well continue, and look for
@@ -1073,7 +1234,7 @@ struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode)
        it being zero) during stress testcases so we need to check for it */
 
        if (cifs_inode == NULL) {
-               cERROR(1, ("Null inode passed to cifs_writeable_file"));
+               cERROR(1, "Null inode passed to cifs_writeable_file");
                dump_stack();
                return NULL;
        }
@@ -1088,7 +1249,7 @@ refind_writable:
                if (open_file->pfile &&
                    ((open_file->pfile->f_flags & O_RDWR) ||
                     (open_file->pfile->f_flags & O_WRONLY))) {
-                       atomic_inc(&open_file->wrtPending);
+                       cifsFileInfo_get(open_file);
 
                        if (!open_file->invalidHandle) {
                                /* found a good writable file */
@@ -1105,7 +1266,7 @@ refind_writable:
                                else { /* start over in case this was deleted */
                                       /* since the list could be modified */
                                        read_lock(&GlobalSMBSeslock);
-                                       atomic_dec(&open_file->wrtPending);
+                                       cifsFileInfo_put(open_file);
                                        goto refind_writable;
                                }
                        }
@@ -1117,11 +1278,11 @@ refind_writable:
                        again. Note that it would be bad
                        to hold up writepages here (rather than
                        in caller) with continuous retries */
-                       cFYI(1, ("wp failed on reopen file"));
+                       cFYI(1, "wp failed on reopen file");
                        read_lock(&GlobalSMBSeslock);
                        /* can not use this handle, no write
                           pending on this one after all */
-                       atomic_dec(&open_file->wrtPending);
+                       cifsFileInfo_put(open_file);
 
                        if (open_file->closePend) /* list could have changed */
                                goto refind_writable;
@@ -1185,7 +1346,7 @@ static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
        if (open_file) {
                bytes_written = cifs_write(open_file->pfile, write_data,
                                           to-from, &offset);
-               atomic_dec(&open_file->wrtPending);
+               cifsFileInfo_put(open_file);
                /* Does mm or vfs already set times? */
                inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
                if ((bytes_written > 0) && (offset))
@@ -1193,7 +1354,7 @@ static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
                else if (bytes_written < 0)
                        rc = bytes_written;
        } else {
-               cFYI(1, ("No writeable filehandles for inode"));
+               cFYI(1, "No writeable filehandles for inode");
                rc = -EIO;
        }
 
@@ -1219,11 +1380,12 @@ static int cifs_writepages(struct address_space *mapping,
        int nr_pages;
        __u64 offset = 0;
        struct cifsFileInfo *open_file;
+       struct cifsInodeInfo *cifsi = CIFS_I(mapping->host);
        struct page *page;
        struct pagevec pvec;
        int rc = 0;
        int scanned = 0;
-       int xid;
+       int xid, long_op;
 
        cifs_sb = CIFS_SB(mapping->host->i_sb);
 
@@ -1364,18 +1526,21 @@ retry:
                         */
                        open_file = find_writable_file(CIFS_I(mapping->host));
                        if (!open_file) {
-                               cERROR(1, ("No writable handles for inode"));
+                               cERROR(1, "No writable handles for inode");
                                rc = -EBADF;
                        } else {
+                               long_op = cifs_write_timeout(cifsi, offset);
                                rc = CIFSSMBWrite2(xid, cifs_sb->tcon,
                                                   open_file->netfid,
                                                   bytes_to_write, offset,
                                                   &bytes_written, iov, n_iov,
-                                                  CIFS_LONG_OP);
-                               atomic_dec(&open_file->wrtPending);
+                                                  long_op);
+                               cifsFileInfo_put(open_file);
+                               cifs_update_eof(cifsi, offset, bytes_written);
+
                                if (rc || bytes_written < bytes_to_write) {
-                                       cERROR(1, ("Write2 ret %d, wrote %d",
-                                                 rc, bytes_written));
+                                       cERROR(1, "Write2 ret %d, wrote %d",
+                                                 rc, bytes_written);
                                        /* BB what if continued retry is
                                           requested via mount flags? */
                                        if (rc == -ENOSPC)
@@ -1404,7 +1569,10 @@ retry:
                        if ((wbc->nr_to_write -= n_iov) <= 0)
                                done = 1;
                        index = next;
-               }
+               } else
+                       /* Need to re-find the pages we skipped */
+                       index = pvec.pages[0]->index + 1;
+
                pagevec_release(&pvec);
        }
        if (!scanned && !done) {
@@ -1433,7 +1601,7 @@ static int cifs_writepage(struct page *page, struct writeback_control *wbc)
 /* BB add check for wbc flags */
        page_cache_get(page);
        if (!PageUptodate(page))
-               cFYI(1, ("ppw - page not up to date"));
+               cFYI(1, "ppw - page not up to date");
 
        /*
         * Set the "writeback" flag, and clear "dirty" in the radix tree.
@@ -1462,10 +1630,14 @@ static int cifs_write_end(struct file *file, struct address_space *mapping,
        int rc;
        struct inode *inode = mapping->host;
 
-       cFYI(1, ("write_end for page %p from pos %lld with %d bytes",
-                page, pos, copied));
+       cFYI(1, "write_end for page %p from pos %lld with %d bytes",
+                page, pos, copied);
 
-       if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
+       if (PageChecked(page)) {
+               if (copied == len)
+                       SetPageUptodate(page);
+               ClearPageChecked(page);
+       } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
                SetPageUptodate(page);
 
        if (!PageUptodate(page)) {
@@ -1504,22 +1676,30 @@ static int cifs_write_end(struct file *file, struct address_space *mapping,
        return rc;
 }
 
-int cifs_fsync(struct file *file, struct dentry *dentry, int datasync)
+int cifs_fsync(struct file *file, int datasync)
 {
        int xid;
        int rc = 0;
+       struct cifsTconInfo *tcon;
+       struct cifsFileInfo *smbfile =
+               (struct cifsFileInfo *)file->private_data;
        struct inode *inode = file->f_path.dentry->d_inode;
 
        xid = GetXid();
 
-       cFYI(1, ("Sync file - name: %s datasync: 0x%x",
-               dentry->d_name.name, datasync));
+       cFYI(1, "Sync file - name: %s datasync: 0x%x",
+               file->f_path.dentry->d_name.name, datasync);
 
        rc = filemap_write_and_wait(inode->i_mapping);
        if (rc == 0) {
                rc = CIFS_I(inode)->write_behind_rc;
                CIFS_I(inode)->write_behind_rc = 0;
+               tcon = CIFS_SB(inode->i_sb)->tcon;
+               if (!rc && tcon && smbfile &&
+                  !(CIFS_SB(inode->i_sb)->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
+                       rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
        }
+
        FreeXid(xid);
        return rc;
 }
@@ -1532,7 +1712,7 @@ int cifs_fsync(struct file *file, struct dentry *dentry, int datasync)
        unsigned int rpages = 0;
        int rc = 0;
 
-       cFYI(1, ("sync page %p",page));
+       cFYI(1, "sync page %p", page);
        mapping = page->mapping;
        if (!mapping)
                return 0;
@@ -1543,7 +1723,7 @@ int cifs_fsync(struct file *file, struct dentry *dentry, int datasync)
 /*     fill in rpages then
        result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
 
-/*     cFYI(1, ("rpages is %d for sync page of Index %ld", rpages, index));
+/*     cFYI(1, "rpages is %d for sync page of Index %ld", rpages, index);
 
 #if 0
        if (rc < 0)
@@ -1577,7 +1757,7 @@ int cifs_flush(struct file *file, fl_owner_t id)
                CIFS_I(inode)->write_behind_rc = 0;
        }
 
-       cFYI(1, ("Flush inode %p file %p rc %d", inode, file, rc));
+       cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
 
        return rc;
 }
@@ -1602,13 +1782,14 @@ ssize_t cifs_user_read(struct file *file, char __user *read_data,
        pTcon = cifs_sb->tcon;
 
        if (file->private_data == NULL) {
+               rc = -EBADF;
                FreeXid(xid);
-               return -EBADF;
+               return rc;
        }
        open_file = (struct cifsFileInfo *)file->private_data;
 
        if ((file->f_flags & O_ACCMODE) == O_WRONLY)
-               cFYI(1, ("attempting read on write only file instance"));
+               cFYI(1, "attempting read on write only file instance");
 
        for (total_read = 0, current_offset = read_data;
             read_size > total_read;
@@ -1682,13 +1863,14 @@ static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
        pTcon = cifs_sb->tcon;
 
        if (file->private_data == NULL) {
+               rc = -EBADF;
                FreeXid(xid);
-               return -EBADF;
+               return rc;
        }
        open_file = (struct cifsFileInfo *)file->private_data;
 
        if ((file->f_flags & O_ACCMODE) == O_WRONLY)
-               cFYI(1, ("attempting read on write only file instance"));
+               cFYI(1, "attempting read on write only file instance");
 
        for (total_read = 0, current_offset = read_data;
             read_size > total_read;
@@ -1734,13 +1916,12 @@ static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
 
 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
 {
-       struct dentry *dentry = file->f_path.dentry;
        int rc, xid;
 
        xid = GetXid();
-       rc = cifs_revalidate(dentry);
+       rc = cifs_revalidate_file(file);
        if (rc) {
-               cFYI(1, ("Validation prior to mmap failed, error=%d", rc));
+               cFYI(1, "Validation prior to mmap failed, error=%d", rc);
                FreeXid(xid);
                return rc;
        }
@@ -1751,8 +1932,7 @@ int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
 
 
 static void cifs_copy_cache_pages(struct address_space *mapping,
-       struct list_head *pages, int bytes_read, char *data,
-       struct pagevec *plru_pvec)
+       struct list_head *pages, int bytes_read, char *data)
 {
        struct page *page;
        char *target;
@@ -1764,10 +1944,10 @@ static void cifs_copy_cache_pages(struct address_space *mapping,
                page = list_entry(pages->prev, struct page, lru);
                list_del(&page->lru);
 
-               if (add_to_page_cache(page, mapping, page->index,
+               if (add_to_page_cache_lru(page, mapping, page->index,
                                      GFP_KERNEL)) {
                        page_cache_release(page);
-                       cFYI(1, ("Add page cache failed"));
+                       cFYI(1, "Add page cache failed");
                        data += PAGE_CACHE_SIZE;
                        bytes_read -= PAGE_CACHE_SIZE;
                        continue;
@@ -1790,8 +1970,6 @@ static void cifs_copy_cache_pages(struct address_space *mapping,
                flush_dcache_page(page);
                SetPageUptodate(page);
                unlock_page(page);
-               if (!pagevec_add(plru_pvec, page))
-                       __pagevec_lru_add_file(plru_pvec);
                data += PAGE_CACHE_SIZE;
        }
        return;
@@ -1810,21 +1988,20 @@ static int cifs_readpages(struct file *file, struct address_space *mapping,
        unsigned int read_size, i;
        char *smb_read_data = NULL;
        struct smb_com_read_rsp *pSMBr;
-       struct pagevec lru_pvec;
        struct cifsFileInfo *open_file;
        int buf_type = CIFS_NO_BUFFER;
 
        xid = GetXid();
        if (file->private_data == NULL) {
+               rc = -EBADF;
                FreeXid(xid);
-               return -EBADF;
+               return rc;
        }
        open_file = (struct cifsFileInfo *)file->private_data;
        cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
        pTcon = cifs_sb->tcon;
 
-       pagevec_init(&lru_pvec, 0);
-       cFYI(DBG2, ("rpages: num pages %d", num_pages));
+       cFYI(DBG2, "rpages: num pages %d", num_pages);
        for (i = 0; i < num_pages; ) {
                unsigned contig_pages;
                struct page *tmp_page;
@@ -1857,8 +2034,8 @@ static int cifs_readpages(struct file *file, struct address_space *mapping,
                /* Read size needs to be in multiples of one page */
                read_size = min_t(const unsigned int, read_size,
                                  cifs_sb->rsize & PAGE_CACHE_MASK);
-               cFYI(DBG2, ("rpages: read size 0x%x  contiguous pages %d",
-                               read_size, contig_pages));
+               cFYI(DBG2, "rpages: read size 0x%x  contiguous pages %d",
+                               read_size, contig_pages);
                rc = -EAGAIN;
                while (rc == -EAGAIN) {
                        if ((open_file->invalidHandle) &&
@@ -1885,14 +2062,14 @@ static int cifs_readpages(struct file *file, struct address_space *mapping,
                        }
                }
                if ((rc < 0) || (smb_read_data == NULL)) {
-                       cFYI(1, ("Read error in readpages: %d", rc));
+                       cFYI(1, "Read error in readpages: %d", rc);
                        break;
                } else if (bytes_read > 0) {
                        task_io_account_read(bytes_read);
                        pSMBr = (struct smb_com_read_rsp *)smb_read_data;
                        cifs_copy_cache_pages(mapping, page_list, bytes_read,
                                smb_read_data + 4 /* RFC1001 hdr */ +
-                               le16_to_cpu(pSMBr->DataOffset), &lru_pvec);
+                               le16_to_cpu(pSMBr->DataOffset));
 
                        i +=  bytes_read >> PAGE_CACHE_SHIFT;
                        cifs_stats_bytes_read(pTcon, bytes_read);
@@ -1908,9 +2085,9 @@ static int cifs_readpages(struct file *file, struct address_space *mapping,
                                /* break; */
                        }
                } else {
-                       cFYI(1, ("No bytes read (%d) at offset %lld . "
-                                "Cleaning remaining pages from readahead list",
-                                bytes_read, offset));
+                       cFYI(1, "No bytes read (%d) at offset %lld . "
+                               "Cleaning remaining pages from readahead list",
+                               bytes_read, offset);
                        /* BB turn off caching and do new lookup on
                           file size at server? */
                        break;
@@ -1925,8 +2102,6 @@ static int cifs_readpages(struct file *file, struct address_space *mapping,
                bytes_read = 0;
        }
 
-       pagevec_lru_add_file(&lru_pvec);
-
 /* need to free smb_read_data buf before exit */
        if (smb_read_data) {
                if (buf_type == CIFS_SMALL_BUFFER)
@@ -1955,7 +2130,7 @@ static int cifs_readpage_worker(struct file *file, struct page *page,
        if (rc < 0)
                goto io_error;
        else
-               cFYI(1, ("Bytes read %d", rc));
+               cFYI(1, "Bytes read %d", rc);
 
        file->f_path.dentry->d_inode->i_atime =
                current_fs_time(file->f_path.dentry->d_inode->i_sb);
@@ -1982,12 +2157,13 @@ static int cifs_readpage(struct file *file, struct page *page)
        xid = GetXid();
 
        if (file->private_data == NULL) {
+               rc = -EBADF;
                FreeXid(xid);
-               return -EBADF;
+               return rc;
        }
 
-       cFYI(1, ("readpage %p at offset %d 0x%x\n",
-                page, (int)offset, (int)offset));
+       cFYI(1, "readpage %p at offset %d 0x%x\n",
+                page, (int)offset, (int)offset);
 
        rc = cifs_readpage_worker(file, page, &offset);
 
@@ -2052,41 +2228,137 @@ static int cifs_write_begin(struct file *file, struct address_space *mapping,
 {
        pgoff_t index = pos >> PAGE_CACHE_SHIFT;
        loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
+       loff_t page_start = pos & PAGE_MASK;
+       loff_t i_size;
+       struct page *page;
+       int rc = 0;
 
-       cFYI(1, ("write_begin from %lld len %d", (long long)pos, len));
-
-       *pagep = __grab_cache_page(mapping, index);
-       if (!*pagep)
-               return -ENOMEM;
+       cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
 
-       if (PageUptodate(*pagep))
-               return 0;
+       page = grab_cache_page_write_begin(mapping, index, flags);
+       if (!page) {
+               rc = -ENOMEM;
+               goto out;
+       }
 
-       /* If we are writing a full page it will be up to date,
-          no need to read from the server */
-       if (len == PAGE_CACHE_SIZE && flags & AOP_FLAG_UNINTERRUPTIBLE)
-               return 0;
+       if (PageUptodate(page))
+               goto out;
 
-       if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
-               int rc;
+       /*
+        * If we write a full page it will be up to date, no need to read from
+        * the server. If the write is short, we'll end up doing a sync write
+        * instead.
+        */
+       if (len == PAGE_CACHE_SIZE)
+               goto out;
 
-               /* might as well read a page, it is fast enough */
-               rc = cifs_readpage_worker(file, *pagep, &offset);
+       /*
+        * optimize away the read when we have an oplock, and we're not
+        * expecting to use any of the data we'd be reading in. That
+        * is, when the page lies beyond the EOF, or straddles the EOF
+        * and the write will cover all of the existing data.
+        */
+       if (CIFS_I(mapping->host)->clientCanCacheRead) {
+               i_size = i_size_read(mapping->host);
+               if (page_start >= i_size ||
+                   (offset == 0 && (pos + len) >= i_size)) {
+                       zero_user_segments(page, 0, offset,
+                                          offset + len,
+                                          PAGE_CACHE_SIZE);
+                       /*
+                        * PageChecked means that the parts of the page
+                        * to which we're not writing are considered up
+                        * to date. Once the data is copied to the
+                        * page, it can be set uptodate.
+                        */
+                       SetPageChecked(page);
+                       goto out;
+               }
+       }
 
-               /* we do not need to pass errors back
-                  e.g. if we do not have read access to the file
-                  because cifs_write_end will attempt synchronous writes
-                  -- shaggy */
+       if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
+               /*
+                * might as well read a page, it is fast enough. If we get
+                * an error, we don't need to return it. cifs_write_end will
+                * do a sync write instead since PG_uptodate isn't set.
+                */
+               cifs_readpage_worker(file, page, &page_start);
        } else {
                /* we could try using another file handle if there is one -
                   but how would we lock it to prevent close of that handle
                   racing with this read? In any case
                   this will be written out by write_end so is fine */
        }
+out:
+       *pagep = page;
+       return rc;
+}
+
+static void
+cifs_oplock_break(struct slow_work *work)
+{
+       struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
+                                                 oplock_break);
+       struct inode *inode = cfile->pInode;
+       struct cifsInodeInfo *cinode = CIFS_I(inode);
+       struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->mnt->mnt_sb);
+       int rc, waitrc = 0;
+
+       if (inode && S_ISREG(inode->i_mode)) {
+               if (cinode->clientCanCacheRead)
+                       break_lease(inode, O_RDONLY);
+               else
+                       break_lease(inode, O_WRONLY);
+               rc = filemap_fdatawrite(inode->i_mapping);
+               if (cinode->clientCanCacheRead == 0) {
+                       waitrc = filemap_fdatawait(inode->i_mapping);
+                       invalidate_remote_inode(inode);
+               }
+               if (!rc)
+                       rc = waitrc;
+               if (rc)
+                       cinode->write_behind_rc = rc;
+               cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
+       }
+
+       /*
+        * releasing stale oplock after recent reconnect of smb session using
+        * a now incorrect file handle is not a data integrity issue but do
+        * not bother sending an oplock release if session to server still is
+        * disconnected since oplock already released by the server
+        */
+       if (!cfile->closePend && !cfile->oplock_break_cancelled) {
+               rc = CIFSSMBLock(0, cifs_sb->tcon, cfile->netfid, 0, 0, 0, 0,
+                                LOCKING_ANDX_OPLOCK_RELEASE, false);
+               cFYI(1, "Oplock release rc = %d", rc);
+       }
+}
 
+static int
+cifs_oplock_break_get(struct slow_work *work)
+{
+       struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
+                                                 oplock_break);
+       mntget(cfile->mnt);
+       cifsFileInfo_get(cfile);
        return 0;
 }
 
+static void
+cifs_oplock_break_put(struct slow_work *work)
+{
+       struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
+                                                 oplock_break);
+       mntput(cfile->mnt);
+       cifsFileInfo_put(cfile);
+}
+
+const struct slow_work_ops cifs_oplock_break_ops = {
+       .get_ref        = cifs_oplock_break_get,
+       .put_ref        = cifs_oplock_break_put,
+       .execute        = cifs_oplock_break,
+};
+
 const struct address_space_operations cifs_addr_ops = {
        .readpage = cifs_readpage,
        .readpages = cifs_readpages,