nfsd4: set shorter timeout
[safe/jmp/linux-2.6] / fs / cifs / file.c
index 7d6fb6f..50ca088 100644 (file)
@@ -46,13 +46,13 @@ static inline struct cifsFileInfo *cifs_init_private(
        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->fh_mutex);
        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;
+       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
@@ -75,7 +75,39 @@ static inline int cifs_convert_flags(unsigned int flags)
                return (GENERIC_READ | GENERIC_WRITE);
        }
 
-       return 0x20197;
+       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_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)
@@ -93,6 +125,80 @@ 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,
+                       struct cifsFileInfo *pCifsFile, int oplock, u16 netfid)
+{
+       struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
+/*     struct timespec temp; */   /* BB REMOVEME BB */
+
+       file->private_data = kmalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
+       if (file->private_data == NULL)
+               return -ENOMEM;
+       pCifsFile = cifs_init_private(file->private_data, inode, file, netfid);
+       write_lock(&GlobalSMBSeslock);
+       list_add(&pCifsFile->tlist, &cifs_sb->tcon->openFileList);
+
+       pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
+       if (pCifsInode == NULL) {
+               write_unlock(&GlobalSMBSeslock);
+               return -EINVAL;
+       }
+
+       /* 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);
+       }
+
+       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;
+}
+
+/* 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,
@@ -103,7 +209,7 @@ static inline int cifs_open_inode_helper(struct inode *inode, struct file *file,
 
        /* 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 prepare_write */
+          list to search for one in write_begin */
        if ((file->f_flags & O_ACCMODE) == O_WRONLY) {
                list_add_tail(&pCifsFile->flist,
                              &pCifsInode->openFileList);
@@ -130,7 +236,9 @@ static inline int cifs_open_inode_helper(struct inode *inode, struct file *file,
                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? */
-                       filemap_write_and_wait(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"));
@@ -138,20 +246,20 @@ static inline int cifs_open_inode_helper(struct inode *inode, struct file *file,
        }
 
 client_can_cache:
-       if (pTcon->ses->capabilities & CAP_UNIX)
+       if (pTcon->unix_ext)
                rc = cifs_get_inode_info_unix(&file->f_path.dentry->d_inode,
                        full_path, inode->i_sb, xid);
        else
                rc = cifs_get_inode_info(&file->f_path.dentry->d_inode,
-                       full_path, buf, inode->i_sb, xid);
+                       full_path, buf, inode->i_sb, xid, NULL);
 
        if ((*oplock & 0xF) == OPLOCK_EXCLUSIVE) {
-               pCifsInode->clientCanCacheAll = TRUE;
-               pCifsInode->clientCanCacheRead = TRUE;
+               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;
+               pCifsInode->clientCanCacheRead = true;
 
        return rc;
 }
@@ -161,7 +269,7 @@ int cifs_open(struct inode *inode, struct file *file)
        int rc = -EACCES;
        int xid, oplock;
        struct cifs_sb_info *cifs_sb;
-       struct cifsTconInfo *pTcon;
+       struct cifsTconInfo *tcon;
        struct cifsFileInfo *pCifsFile;
        struct cifsInodeInfo *pCifsInode;
        struct list_head *tmp;
@@ -174,37 +282,34 @@ 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));
+       /* 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_CREAT) && (file->f_flags & O_EXCL))
+                       cERROR(1, ("could not find file instance for "
+                                  "new file %p", file));
 
        full_path = build_path_from_dentry(file->f_path.dentry);
        if (full_path == NULL) {
@@ -212,8 +317,47 @@ int cifs_open(struct inode *inode, struct file *file)
                return -ENOMEM;
        }
 
-       cFYI(1, (" inode = 0x%p file flags are 0x%x for %s",
+       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);
+               /* can not refresh inode info since size could be stale */
+               rc = cifs_posix_open(full_path, &inode, 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 */
+
+                       cifs_posix_open_inode_helper(inode, file, pCifsInode,
+                                                    pCifsFile, 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 */
+       }
+
        desiredAccess = cifs_convert_flags(file->f_flags);
 
 /*********************************************************************
@@ -242,11 +386,6 @@ int cifs_open(struct inode *inode, struct file *file)
 
        disposition = cifs_get_disposition(file->f_flags);
 
-       if (oplockEnabled)
-               oplock = REQ_OPLOCK;
-       else
-               oplock = FALSE;
-
        /* BB pass O_SYNC flag through on file attributes .. BB */
 
        /* Also refresh inode by passing in file_info buf returned by SMBOpen
@@ -263,7 +402,7 @@ 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);
@@ -272,7 +411,7 @@ 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);
@@ -289,12 +428,12 @@ int cifs_open(struct inode *inode, struct file *file)
        }
        pCifsFile = cifs_init_private(file->private_data, inode, file, netfid);
        write_lock(&GlobalSMBSeslock);
-       list_add(&pCifsFile->tlist, &pTcon->openFileList);
+       list_add(&pCifsFile->tlist, &tcon->openFileList);
 
        pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
        if (pCifsInode) {
                rc = cifs_open_inode_helper(inode, file, pCifsInode,
-                                           pCifsFile, pTcon,
+                                           pCifsFile, tcon,
                                            &oplock, buf, full_path, xid);
        } else {
                write_unlock(&GlobalSMBSeslock);
@@ -303,19 +442,20 @@ int cifs_open(struct inode *inode, struct file *file)
        if (oplock & CIFS_CREATE_ACTION) {
                /* time to set mode which we can not set earlier due to
                   problems creating new read-only files */
-               if (cifs_sb->tcon->ses->capabilities & CAP_UNIX) {
-                       CIFSSMBUnixSetPerms(xid, pTcon, full_path,
-                                           inode->i_mode,
-                                           (__u64)-1, (__u64)-1, 0 /* dev */,
+               if (tcon->unix_ext) {
+                       struct cifs_unix_set_info_args args = {
+                               .mode   = inode->i_mode,
+                               .uid    = NO_CHANGE_64,
+                               .gid    = NO_CHANGE_64,
+                               .ctime  = NO_CHANGE_64,
+                               .atime  = NO_CHANGE_64,
+                               .mtime  = NO_CHANGE_64,
+                               .device = 0,
+                       };
+                       CIFSSMBUnixSetInfo(xid, tcon, full_path, &args,
                                            cifs_sb->local_nls,
                                            cifs_sb->mnt_cifs_flags &
                                                CIFS_MOUNT_MAP_SPECIAL_CHR);
-               } else {
-                       /* BB implement via Windows security descriptors eg
-                          CIFSSMBWinSetPerms(xid, pTcon, full_path, mode,
-                                             -1, -1, local_nls);
-                          in the meantime could set r/o dos attribute when
-                          perms are eg: mode & 0222 == 0 */
                }
        }
 
@@ -337,12 +477,12 @@ static int cifs_relock_file(struct cifsFileInfo *cifsFile)
        return rc;
 }
 
-static int cifs_reopen_file(struct file *file, int can_flush)
+static int cifs_reopen_file(struct file *file, bool can_flush)
 {
        int rc = -EACCES;
        int xid, oplock;
        struct cifs_sb_info *cifs_sb;
-       struct cifsTconInfo *pTcon;
+       struct cifsTconInfo *tcon;
        struct cifsFileInfo *pCifsFile;
        struct cifsInodeInfo *pCifsInode;
        struct inode *inode;
@@ -351,15 +491,15 @@ static int cifs_reopen_file(struct file *file, int can_flush)
        int disposition = FILE_OPEN;
        __u16 netfid;
 
-       if (file->private_data) {
+       if (file->private_data)
                pCifsFile = (struct cifsFileInfo *)file->private_data;
-       else
+       else
                return -EBADF;
 
        xid = GetXid();
-       down(&pCifsFile->fh_sem);
-       if (pCifsFile->invalidHandle == FALSE) {
-               up(&pCifsFile->fh_sem);
+       mutex_unlock(&pCifsFile->fh_mutex);
+       if (!pCifsFile->invalidHandle) {
+               mutex_lock(&pCifsFile->fh_mutex);
                FreeXid(xid);
                return 0;
        }
@@ -378,9 +518,9 @@ static int cifs_reopen_file(struct file *file, int can_flush)
                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
@@ -390,19 +530,36 @@ static int cifs_reopen_file(struct file *file, int can_flush)
        if (full_path == NULL) {
                rc = -ENOMEM;
 reopen_error_exit:
-               up(&pCifsFile->fh_sem);
+               mutex_lock(&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);
 
        if (oplockEnabled)
                oplock = REQ_OPLOCK;
        else
-               oplock = FALSE;
+               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, 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
@@ -410,33 +567,36 @@ reopen_error_exit:
           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);
+               mutex_lock(&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);
+               pCifsFile->invalidHandle = false;
+               mutex_lock(&pCifsFile->fh_mutex);
                pCifsInode = CIFS_I(inode);
                if (pCifsInode) {
                        if (can_flush) {
-                               filemap_write_and_wait(inode->i_mapping);
+                               rc = filemap_write_and_wait(inode->i_mapping);
+                               if (rc != 0)
+                                       CIFS_I(inode)->write_behind_rc = rc;
                        /* temporarily disable caching while we
                           go to server to get inode info */
-                               pCifsInode->clientCanCacheAll = FALSE;
-                               pCifsInode->clientCanCacheRead = FALSE;
-                               if (pTcon->ses->capabilities & CAP_UNIX)
+                               pCifsInode->clientCanCacheAll = false;
+                               pCifsInode->clientCanCacheRead = false;
+                               if (tcon->unix_ext)
                                        rc = cifs_get_inode_info_unix(&inode,
                                                full_path, inode->i_sb, xid);
                                else
                                        rc = cifs_get_inode_info(&inode,
                                                full_path, NULL, inode->i_sb,
-                                               xid);
+                                               xid, NULL);
                        } /* else we are writing out data to server already
                             and could deadlock if we tried to flush data, and
                             since we do not know if we have data that would
@@ -444,21 +604,20 @@ reopen_error_exit:
                             we can not go to the server to get the new inod
                             info */
                        if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
-                               pCifsInode->clientCanCacheAll = TRUE;
-                               pCifsInode->clientCanCacheRead = TRUE;
+                               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;
-                               pCifsInode->clientCanCacheAll = FALSE;
+                               pCifsInode->clientCanCacheRead = true;
+                               pCifsInode->clientCanCacheAll = false;
                        } else {
-                               pCifsInode->clientCanCacheRead = FALSE;
-                               pCifsInode->clientCanCacheAll = FALSE;
+                               pCifsInode->clientCanCacheRead = false;
+                               pCifsInode->clientCanCacheAll = false;
                        }
                        cifs_relock_file(pCifsFile);
                }
        }
-
        kfree(full_path);
        FreeXid(xid);
        return rc;
@@ -467,7 +626,7 @@ reopen_error_exit:
 int cifs_close(struct inode *inode, struct file *file)
 {
        int rc = 0;
-       int xid;
+       int xid, timeout;
        struct cifs_sb_info *cifs_sb;
        struct cifsTconInfo *pTcon;
        struct cifsFileInfo *pSMBFile =
@@ -479,15 +638,16 @@ int cifs_close(struct inode *inode, struct file *file)
        pTcon = cifs_sb->tcon;
        if (pSMBFile) {
                struct cifsLockInfo *li, *tmp;
-
-               pSMBFile->closePend = TRUE;
+               write_lock(&GlobalSMBSeslock);
+               pSMBFile->closePend = true;
                if (pTcon) {
                        /* no sense reconnecting to close a file that is
                           already closed */
-                       if (pTcon->tidStatus != CifsNeedReconnect) {
-                               int timeout = 2;
+                       if (!pTcon->need_reconnect) {
+                               write_unlock(&GlobalSMBSeslock);
+                               timeout = 2;
                                while ((atomic_read(&pSMBFile->wrtPending) != 0)
-                                        && (timeout < 1000) ) {
+                                       && (timeout <= 2048)) {
                                        /* Give write a better chance to get to
                                        server ahead of the close.  We do not
                                        want to add a wait_q here as it would
@@ -495,18 +655,21 @@ 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 */
-#ifdef CONFIG_CIFS_DEBUG2
-                                       cFYI(1, ("close delay, write pending"));
-#endif /* DEBUG2 */
+                                       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,
+                                       cERROR(1, ("close with pending write"));
+                               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. */
@@ -521,19 +684,37 @@ int cifs_close(struct inode *inode, struct file *file)
                list_del(&pSMBFile->flist);
                list_del(&pSMBFile->tlist);
                write_unlock(&GlobalSMBSeslock);
-               kfree(pSMBFile->search_resume_name);
+               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);
                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));
                /* 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;
-               CIFS_I(inode)->clientCanCacheAll  = FALSE;
+               CIFS_I(inode)->clientCanCacheRead = false;
+               CIFS_I(inode)->clientCanCacheAll  = false;
        }
+       read_unlock(&GlobalSMBSeslock);
        if ((rc == 0) && CIFS_I(inode)->write_behind_rc)
                rc = CIFS_I(inode)->write_behind_rc;
        FreeXid(xid);
@@ -560,15 +741,18 @@ int cifs_closedir(struct inode *inode, struct file *file)
                pTcon = cifs_sb->tcon;
 
                cFYI(1, ("Freeing private data in close dir"));
-               if ((pCFileStruct->srch_inf.endOfSearch == FALSE) &&
-                  (pCFileStruct->invalidHandle == FALSE)) {
-                       pCFileStruct->invalidHandle = TRUE;
+               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));
                        /* 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"));
@@ -578,12 +762,6 @@ int cifs_closedir(struct inode *inode, struct file *file)
                        else
                                cifs_buf_release(ptmp);
                }
-               ptmp = pCFileStruct->search_resume_name;
-               if (ptmp) {
-                       cFYI(1, ("closedir free resume name"));
-                       pCFileStruct->search_resume_name = NULL;
-                       kfree(ptmp);
-               }
                kfree(file->private_data);
                file->private_data = NULL;
        }
@@ -614,12 +792,12 @@ int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
        __u32 numLock = 0;
        __u32 numUnlock = 0;
        __u64 length;
-       int wait_flag = FALSE;
+       bool wait_flag = false;
        struct cifs_sb_info *cifs_sb;
-       struct cifsTconInfo *pTcon;
+       struct cifsTconInfo *tcon;
        __u16 netfid;
        __u8 lockType = LOCKING_ANDX_LARGE_FILES;
-       int posix_locking;
+       bool posix_locking = 0;
 
        length = 1 + pfLock->fl_end - pfLock->fl_start;
        rc = -EACCES;
@@ -636,7 +814,7 @@ int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
                cFYI(1, ("Flock"));
        if (pfLock->fl_flags & FL_SLEEP) {
                cFYI(1, ("Blocking lock"));
-               wait_flag = TRUE;
+               wait_flag = true;
        }
        if (pfLock->fl_flags & FL_ACCESS)
                cFYI(1, ("Process suspended by mandatory locking - "
@@ -670,7 +848,7 @@ int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
                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) {
                FreeXid(xid);
@@ -678,9 +856,10 @@ int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
        }
        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 */
@@ -691,7 +870,7 @@ 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);
@@ -699,10 +878,10 @@ 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 */ );
@@ -735,11 +914,11 @@ int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
                        posix_lock_type = CIFS_RDLCK;
                else
                        posix_lock_type = CIFS_WRLCK;
-               
+
                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 {
@@ -747,7 +926,7 @@ 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);
 
@@ -766,11 +945,12 @@ int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
                        mutex_lock(&fid->lock_mutex);
                        list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
                                if (pfLock->fl_start <= li->offset &&
-                                               length >= li->length) {
-                                       stored_rc = CIFSSMBLock(xid, pTcon,
+                                               (pfLock->fl_start + length) >=
+                                               (li->offset + li->length)) {
+                                       stored_rc = CIFSSMBLock(xid, tcon,
                                                        netfid,
                                                        li->length, li->offset,
-                                                       1, 0, li->type, FALSE);
+                                                       1, 0, li->type, false);
                                        if (stored_rc)
                                                rc = stored_rc;
 
@@ -788,6 +968,40 @@ 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)
 {
@@ -798,6 +1012,7 @@ 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);
 
@@ -810,14 +1025,14 @@ ssize_t cifs_user_write(struct file *file, const char __user *write_data,
        if (file->private_data == NULL)
                return -EBADF;
        open_file = (struct cifsFileInfo *) file->private_data;
-       
-       xid = GetXid();
 
-       if (*poffset > file->f_path.dentry->d_inode->i_size)
-               long_op = 2; /* writes past end of file can take a long time */
-       else
-               long_op = 1;
+       rc = generic_write_checks(file, poffset, &write_size, 0);
+       if (rc)
+               return rc;
 
+       xid = GetXid();
+
+       long_op = cifs_write_timeout(cifsi, *poffset);
        for (total_written = 0; write_size > total_written;
             total_written += bytes_written) {
                rc = -EAGAIN;
@@ -842,7 +1057,7 @@ ssize_t cifs_user_write(struct file *file, const char __user *write_data,
                                   filemap_fdatawait from here so tell
                                   reopen_file not to flush data to server
                                   now */
-                               rc = cifs_reopen_file(file, FALSE);
+                               rc = cifs_reopen_file(file, false);
                                if (rc != 0)
                                        break;
                        }
@@ -861,9 +1076,11 @@ 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 = FALSE; /* subsequent writes fast -
+               }
+               long_op = CIFS_STD_OP; /* subsequent writes fast -
                                    15 seconds is plenty */
        }
 
@@ -889,7 +1106,7 @@ ssize_t cifs_user_write(struct file *file, const char __user *write_data,
 }
 
 static ssize_t cifs_write(struct file *file, const char *write_data,
-       size_t write_size, loff_t *poffset)
+                         size_t write_size, loff_t *poffset)
 {
        int rc = 0;
        unsigned int bytes_written = 0;
@@ -898,6 +1115,7 @@ 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);
 
@@ -909,14 +1127,10 @@ static ssize_t cifs_write(struct file *file, const char *write_data,
        if (file->private_data == NULL)
                return -EBADF;
        open_file = (struct cifsFileInfo *)file->private_data;
-       
-       xid = GetXid();
 
-       if (*poffset > file->f_path.dentry->d_inode->i_size)
-               long_op = 2; /* writes past end of file can take a long time */
-       else
-               long_op = 1;
+       xid = GetXid();
 
+       long_op = cifs_write_timeout(cifsi, *poffset);
        for (total_written = 0; write_size > total_written;
             total_written += bytes_written) {
                rc = -EAGAIN;
@@ -942,7 +1156,7 @@ static ssize_t cifs_write(struct file *file, const char *write_data,
                                   filemap_fdatawait from here so tell
                                   reopen_file not to flush data to
                                   server now */
-                               rc = cifs_reopen_file(file, FALSE);
+                               rc = cifs_reopen_file(file, false);
                                if (rc != 0)
                                        break;
                        }
@@ -979,9 +1193,11 @@ 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 = FALSE; /* subsequent writes fast -
+               }
+               long_op = CIFS_STD_OP; /* subsequent writes fast -
                                    15 seconds is plenty */
        }
 
@@ -1005,9 +1221,41 @@ static ssize_t cifs_write(struct file *file, const char *write_data,
        return total_written;
 }
 
+#ifdef CONFIG_CIFS_EXPERIMENTAL
+struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode)
+{
+       struct cifsFileInfo *open_file = NULL;
+
+       read_lock(&GlobalSMBSeslock);
+       /* we could simply get the first_list_entry since write-only entries
+          are always at the end of the list but since the first entry might
+          have a close pending, we go through the whole list */
+       list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
+               if (open_file->closePend)
+                       continue;
+               if (open_file->pfile && ((open_file->pfile->f_flags & O_RDWR) ||
+                   (open_file->pfile->f_flags & O_RDONLY))) {
+                       if (!open_file->invalidHandle) {
+                               /* found a good file */
+                               /* lock it so it will not be closed on us */
+                               atomic_inc(&open_file->wrtPending);
+                               read_unlock(&GlobalSMBSeslock);
+                               return open_file;
+                       } /* else might as well continue, and look for
+                            another, or simply have the caller reopen it
+                            again rather than trying to fix this handle */
+               } else /* write only file */
+                       break; /* write only files are last so must be done */
+       }
+       read_unlock(&GlobalSMBSeslock);
+       return NULL;
+}
+#endif
+
 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode)
 {
        struct cifsFileInfo *open_file;
+       bool any_available = false;
        int rc;
 
        /* Having a null inode here (because mapping->host was set to zero by
@@ -1021,32 +1269,66 @@ struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode)
        }
 
        read_lock(&GlobalSMBSeslock);
+refind_writable:
        list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
-               if (open_file->closePend)
+               if (open_file->closePend ||
+                   (!any_available && open_file->pid != current->tgid))
                        continue;
+
                if (open_file->pfile &&
                    ((open_file->pfile->f_flags & O_RDWR) ||
                     (open_file->pfile->f_flags & O_WRONLY))) {
                        atomic_inc(&open_file->wrtPending);
+
+                       if (!open_file->invalidHandle) {
+                               /* found a good writable file */
+                               read_unlock(&GlobalSMBSeslock);
+                               return open_file;
+                       }
+
                        read_unlock(&GlobalSMBSeslock);
-                       if ((open_file->invalidHandle) &&
-                          (!open_file->closePend) /* BB fixme -since the second clause can not be true remove it BB */) {
-                               rc = cifs_reopen_file(open_file->pfile, FALSE);
-                               /* if it fails, try another handle - might be */
-                               /* dangerous to hold up writepages with retry */
-                               if (rc) {
-                                       cFYI(1, ("failed on reopen file in wp"));
+                       /* Had to unlock since following call can block */
+                       rc = cifs_reopen_file(open_file->pfile, false);
+                       if (!rc) {
+                               if (!open_file->closePend)
+                                       return open_file;
+                               else { /* start over in case this was deleted */
+                                      /* since the list could be modified */
                                        read_lock(&GlobalSMBSeslock);
-                                       /* can not use this handle, no write
-                                       pending on this one after all */
-                                       atomic_dec
-                                            (&open_file->wrtPending);
-                                       continue;
+                                       atomic_dec(&open_file->wrtPending);
+                                       goto refind_writable;
                                }
                        }
-                       return open_file;
+
+                       /* if it fails, try another handle if possible -
+                       (we can not do this if closePending since
+                       loop could be modified - in which case we
+                       have to start at the beginning of the list
+                       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"));
+                       read_lock(&GlobalSMBSeslock);
+                       /* can not use this handle, no write
+                          pending on this one after all */
+                       atomic_dec(&open_file->wrtPending);
+
+                       if (open_file->closePend) /* list could have changed */
+                               goto refind_writable;
+                       /* else we simply continue to the next entry. Thus
+                          we do not loop on reopen errors.  If we
+                          can not reopen the file, for example if we
+                          reconnected to a server with another client
+                          racing to delete or lock the file we would not
+                          make progress if we restarted before the beginning
+                          of the loop here. */
                }
        }
+       /* couldn't find useable FH with same pid, try any available */
+       if (!any_available) {
+               any_available = true;
+               goto refind_writable;
+       }
        read_unlock(&GlobalSMBSeslock);
        return NULL;
 }
@@ -1096,12 +1378,10 @@ static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
                atomic_dec(&open_file->wrtPending);
                /* Does mm or vfs already set times? */
                inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
-               if ((bytes_written > 0) && (offset)) {
+               if ((bytes_written > 0) && (offset))
                        rc = 0;
-               } else if (bytes_written < 0) {
-                       if (rc != -EBADF)
-                               rc = bytes_written;
-               }
+               else if (bytes_written < 0)
+                       rc = bytes_written;
        } else {
                cFYI(1, ("No writeable filehandles for inode"));
                rc = -EIO;
@@ -1129,14 +1409,15 @@ 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);
-       
+
        /*
         * If wsize is smaller that the page cache size, default to writing
         * one page at a time via cifs_writepage
@@ -1203,7 +1484,7 @@ retry:
 
                        if (first < 0)
                                lock_page(page);
-                       else if (TestSetPageLocked(page))
+                       else if (!trylock_page(page))
                                break;
 
                        if (unlikely(page->mapping != mapping)) {
@@ -1277,18 +1558,24 @@ retry:
                                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,
-                                                  1);
+                                                  long_op);
                                atomic_dec(&open_file->wrtPending);
+                               cifs_update_eof(cifsi, offset, bytes_written);
+
                                if (rc || bytes_written < bytes_to_write) {
-                                       cERROR(1,("Write2 ret %d, written = %d",
+                                       cERROR(1, ("Write2 ret %d, wrote %d",
                                                  rc, bytes_written));
                                        /* BB what if continued retry is
                                           requested via mount flags? */
-                                       set_bit(AS_EIO, &mapping->flags);
+                                       if (rc == -ENOSPC)
+                                               set_bit(AS_ENOSPC, &mapping->flags);
+                                       else
+                                               set_bit(AS_EIO, &mapping->flags);
                                } else {
                                        cifs_stats_bytes_written(cifs_sb->tcon,
                                                                 bytes_written);
@@ -1311,7 +1598,10 @@ 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) {
@@ -1339,9 +1629,8 @@ static int cifs_writepage(struct page *page, struct writeback_control *wbc)
        xid = GetXid();
 /* BB add check for wbc flags */
        page_cache_get(page);
-       if (!PageUptodate(page)) {
+       if (!PageUptodate(page))
                cFYI(1, ("ppw - page not up to date"));
-       }
 
        /*
         * Set the "writeback" flag, and clear "dirty" in the radix tree.
@@ -1363,49 +1652,56 @@ static int cifs_writepage(struct page *page, struct writeback_control *wbc)
        return rc;
 }
 
-static int cifs_commit_write(struct file *file, struct page *page,
-       unsigned offset, unsigned to)
+static int cifs_write_end(struct file *file, struct address_space *mapping,
+                       loff_t pos, unsigned len, unsigned copied,
+                       struct page *page, void *fsdata)
 {
-       int xid;
-       int rc = 0;
-       struct inode *inode = page->mapping->host;
-       loff_t position = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
-       char *page_data;
+       int rc;
+       struct inode *inode = mapping->host;
+
+       cFYI(1, ("write_end for page %p from pos %lld with %d bytes",
+                page, pos, copied));
+
+       if (PageChecked(page)) {
+               if (copied == len)
+                       SetPageUptodate(page);
+               ClearPageChecked(page);
+       } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
+               SetPageUptodate(page);
 
-       xid = GetXid();
-       cFYI(1, ("commit write for page %p up to position %lld for %d",
-                page, position, to));
-       spin_lock(&inode->i_lock);
-       if (position > inode->i_size) {
-               i_size_write(inode, position);
-       }
-       spin_unlock(&inode->i_lock);
        if (!PageUptodate(page)) {
-               position =  ((loff_t)page->index << PAGE_CACHE_SHIFT) + offset;
-               /* can not rely on (or let) writepage write this data */
-               if (to < offset) {
-                       cFYI(1, ("Illegal offsets, can not copy from %d to %d",
-                               offset, to));
-                       FreeXid(xid);
-                       return rc;
-               }
+               char *page_data;
+               unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
+               int xid;
+
+               xid = GetXid();
                /* this is probably better than directly calling
                   partialpage_write since in this function the file handle is
                   known which we might as well leverage */
                /* BB check if anything else missing out of ppw
                   such as updating last write time */
                page_data = kmap(page);
-               rc = cifs_write(file, page_data + offset, to-offset,
-                               &position);
-               if (rc > 0)
-                       rc = 0;
-               /* else if (rc < 0) should we set writebehind rc? */
+               rc = cifs_write(file, page_data + offset, copied, &pos);
+               /* if (rc < 0) should we set writebehind rc? */
                kunmap(page);
+
+               FreeXid(xid);
        } else {
+               rc = copied;
+               pos += copied;
                set_page_dirty(page);
        }
 
-       FreeXid(xid);
+       if (rc > 0) {
+               spin_lock(&inode->i_lock);
+               if (pos > inode->i_size)
+                       i_size_write(inode, pos);
+               spin_unlock(&inode->i_lock);
+       }
+
+       unlock_page(page);
+       page_cache_release(page);
+
        return rc;
 }
 
@@ -1413,16 +1709,26 @@ int cifs_fsync(struct file *file, struct dentry *dentry, 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));
-       
-       rc = filemap_fdatawrite(inode->i_mapping);
-       if (rc == 0)
+
+       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;
 }
@@ -1474,9 +1780,12 @@ int cifs_flush(struct file *file, fl_owner_t id)
           filemapfdatawrite appears easier for the time being */
 
        rc = filemap_fdatawrite(inode->i_mapping);
-       if (!rc) /* reset wb rc if we were able to write out dirty pages */
+       /* reset wb rc if we were able to write out dirty pages */
+       if (!rc) {
+               rc = CIFS_I(inode)->write_behind_rc;
                CIFS_I(inode)->write_behind_rc = 0;
-               
+       }
+
        cFYI(1, ("Flush inode %p file %p rc %d", inode, file, rc));
 
        return rc;
@@ -1507,9 +1816,9 @@ ssize_t cifs_user_read(struct file *file, char __user *read_data,
        }
        open_file = (struct cifsFileInfo *)file->private_data;
 
-       if ((file->f_flags & O_ACCMODE) == O_WRONLY) {
+       if ((file->f_flags & O_ACCMODE) == O_WRONLY)
                cFYI(1, ("attempting read on write only file instance"));
-       }
+
        for (total_read = 0, current_offset = read_data;
             read_size > total_read;
             total_read += bytes_read, current_offset += bytes_read) {
@@ -1521,7 +1830,7 @@ ssize_t cifs_user_read(struct file *file, char __user *read_data,
                        int buf_type = CIFS_NO_BUFFER;
                        if ((open_file->invalidHandle) &&
                            (!open_file->closePend)) {
-                               rc = cifs_reopen_file(file, TRUE);
+                               rc = cifs_reopen_file(file, true);
                                if (rc != 0)
                                        break;
                        }
@@ -1536,9 +1845,8 @@ ssize_t cifs_user_read(struct file *file, char __user *read_data,
                                                smb_read_data +
                                                4 /* RFC1001 length field */ +
                                                le16_to_cpu(pSMBr->DataOffset),
-                                               bytes_read)) {
+                                               bytes_read))
                                        rc = -EFAULT;
-                               }
 
                                if (buf_type == CIFS_SMALL_BUFFER)
                                        cifs_small_buf_release(smb_read_data);
@@ -1607,7 +1915,7 @@ static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
                while (rc == -EAGAIN) {
                        if ((open_file->invalidHandle) &&
                            (!open_file->closePend)) {
-                               rc = cifs_reopen_file(file, TRUE);
+                               rc = cifs_reopen_file(file, true);
                                if (rc != 0)
                                        break;
                        }
@@ -1692,7 +2000,7 @@ static void cifs_copy_cache_pages(struct address_space *mapping,
                SetPageUptodate(page);
                unlock_page(page);
                if (!pagevec_add(plru_pvec, page))
-                       __pagevec_lru_add(plru_pvec);
+                       __pagevec_lru_add_file(plru_pvec);
                data += PAGE_CACHE_SIZE;
        }
        return;
@@ -1707,7 +2015,7 @@ static int cifs_readpages(struct file *file, struct address_space *mapping,
        struct page *page;
        struct cifs_sb_info *cifs_sb;
        struct cifsTconInfo *pTcon;
-       int bytes_read = 0;
+       unsigned int bytes_read = 0;
        unsigned int read_size, i;
        char *smb_read_data = NULL;
        struct smb_com_read_rsp *pSMBr;
@@ -1725,9 +2033,7 @@ static int cifs_readpages(struct file *file, struct address_space *mapping,
        pTcon = cifs_sb->tcon;
 
        pagevec_init(&lru_pvec, 0);
-#ifdef CONFIG_CIFS_DEBUG2
-               cFYI(1, ("rpages: num pages %d", num_pages));
-#endif
+       cFYI(DBG2, ("rpages: num pages %d", num_pages));
        for (i = 0; i < num_pages; ) {
                unsigned contig_pages;
                struct page *tmp_page;
@@ -1760,15 +2066,13 @@ 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);
-#ifdef CONFIG_CIFS_DEBUG2
-               cFYI(1, ("rpages: read size 0x%x  contiguous pages %d",
+               cFYI(DBG2, ("rpages: read size 0x%x  contiguous pages %d",
                                read_size, contig_pages));
-#endif
                rc = -EAGAIN;
                while (rc == -EAGAIN) {
                        if ((open_file->invalidHandle) &&
                            (!open_file->closePend)) {
-                               rc = cifs_reopen_file(file, TRUE);
+                               rc = cifs_reopen_file(file, true);
                                if (rc != 0)
                                        break;
                        }
@@ -1801,7 +2105,7 @@ static int cifs_readpages(struct file *file, struct address_space *mapping,
 
                        i +=  bytes_read >> PAGE_CACHE_SHIFT;
                        cifs_stats_bytes_read(pTcon, bytes_read);
-                       if ((int)(bytes_read & PAGE_CACHE_MASK) != bytes_read) {
+                       if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
                                i++; /* account for partial page */
 
                                /* server copy of file can have smaller size
@@ -1830,7 +2134,7 @@ static int cifs_readpages(struct file *file, struct address_space *mapping,
                bytes_read = 0;
        }
 
-       pagevec_lru_add(&lru_pvec);
+       pagevec_lru_add_file(&lru_pvec);
 
 /* need to free smb_read_data buf before exit */
        if (smb_read_data) {
@@ -1902,93 +2206,125 @@ static int cifs_readpage(struct file *file, struct page *page)
        return rc;
 }
 
+static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
+{
+       struct cifsFileInfo *open_file;
+
+       read_lock(&GlobalSMBSeslock);
+       list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
+               if (open_file->closePend)
+                       continue;
+               if (open_file->pfile &&
+                   ((open_file->pfile->f_flags & O_RDWR) ||
+                    (open_file->pfile->f_flags & O_WRONLY))) {
+                       read_unlock(&GlobalSMBSeslock);
+                       return 1;
+               }
+       }
+       read_unlock(&GlobalSMBSeslock);
+       return 0;
+}
+
 /* We do not want to update the file size from server for inodes
    open for write - to avoid races with writepage extending
    the file - in the future we could consider allowing
    refreshing the inode only on increases in the file size
    but this is tricky to do without racing with writebehind
    page caching in the current Linux kernel design */
-int is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
+bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
 {
-       struct cifsFileInfo *open_file = NULL;
+       if (!cifsInode)
+               return true;
 
-       if (cifsInode)
-               open_file =  find_writable_file(cifsInode);
-       if (open_file) {
+       if (is_inode_writable(cifsInode)) {
+               /* This inode is open for write at least once */
                struct cifs_sb_info *cifs_sb;
 
-               /* there is not actually a write pending so let
-               this handle go free and allow it to
-               be closable if needed */
-               atomic_dec(&open_file->wrtPending);
-
                cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
-               if ( cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO ) {
+               if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
                        /* since no page cache to corrupt on directio
                        we can change size safely */
-                       return 1;
+                       return true;
                }
 
                if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
-                       return 1;
+                       return true;
 
-               return 0;
+               return false;
        } else
-               return 1;
+               return true;
 }
 
-static int cifs_prepare_write(struct file *file, struct page *page,
-       unsigned from, unsigned to)
+static int cifs_write_begin(struct file *file, struct address_space *mapping,
+                       loff_t pos, unsigned len, unsigned flags,
+                       struct page **pagep, void **fsdata)
 {
-       int rc = 0;
+       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;
-       loff_t offset;
+       struct page *page;
+       int rc = 0;
 
-       cFYI(1, ("prepare write for page %p from %d to %d", page, from, to));
-       if (PageUptodate(page))
-               return 0;
+       cFYI(1, ("write_begin from %lld len %d", (long long)pos, len));
 
-       /* If we are writing a full page it will be up to date,
-          no need to read from the server */
-       if ((to == PAGE_CACHE_SIZE) && (from == 0)) {
-               SetPageUptodate(page);
-               return 0;
+       page = grab_cache_page_write_begin(mapping, index, flags);
+       if (!page) {
+               rc = -ENOMEM;
+               goto out;
        }
 
-       offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
-       i_size = i_size_read(page->mapping->host);
+       if (PageUptodate(page))
+               goto out;
+
+       /*
+        * 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;
+
+       /*
+        * 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;
+               }
+       }
 
-       if ((offset >= i_size) ||
-           ((from == 0) && (offset + to) >= i_size)) {
+       if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
                /*
-                * We don't need to read data beyond the end of the file.
-                * zero it, and set the page uptodate
+                * 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.
                 */
-               void *kaddr = kmap_atomic(page, KM_USER0);
-
-               if (from)
-                       memset(kaddr, 0, from);
-               if (to < PAGE_CACHE_SIZE)
-                       memset(kaddr + to, 0, PAGE_CACHE_SIZE - to);
-               flush_dcache_page(page);
-               kunmap_atomic(kaddr, KM_USER0);
-               SetPageUptodate(page);
-       } else if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
-               /* might as well read a page, it is fast enough */
-               rc = cifs_readpage_worker(file, page, &offset);
+               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 commit_write so is fine */
+                  this will be written out by write_end so is fine */
        }
-
-       /* we do not need to pass errors back
-          e.g. if we do not have read access to the file
-          because cifs_commit_write will do the right thing.  -- shaggy */
-
-       return 0;
+out:
+       *pagep = page;
+       return rc;
 }
 
 const struct address_space_operations cifs_addr_ops = {
@@ -1996,8 +2332,8 @@ const struct address_space_operations cifs_addr_ops = {
        .readpages = cifs_readpages,
        .writepage = cifs_writepage,
        .writepages = cifs_writepages,
-       .prepare_write = cifs_prepare_write,
-       .commit_write = cifs_commit_write,
+       .write_begin = cifs_write_begin,
+       .write_end = cifs_write_end,
        .set_page_dirty = __set_page_dirty_nobuffers,
        /* .sync_page = cifs_sync_page, */
        /* .direct_IO = */
@@ -2012,8 +2348,8 @@ const struct address_space_operations cifs_addr_ops_smallbuf = {
        .readpage = cifs_readpage,
        .writepage = cifs_writepage,
        .writepages = cifs_writepages,
-       .prepare_write = cifs_prepare_write,
-       .commit_write = cifs_commit_write,
+       .write_begin = cifs_write_begin,
+       .write_end = cifs_write_end,
        .set_page_dirty = __set_page_dirty_nobuffers,
        /* .sync_page = cifs_sync_page, */
        /* .direct_IO = */