[CIFS] fix for incorrect session reconnects
[safe/jmp/linux-2.6] / fs / cifs / inode.c
index b88147c..97ccc51 100644 (file)
@@ -1,7 +1,7 @@
 /*
  *   fs/cifs/inode.c
  *
- *   Copyright (C) International Business Machines  Corp., 2002,2005
+ *   Copyright (C) International Business Machines  Corp., 2002,2007
  *   Author(s): Steve French (sfrench@us.ibm.com)
  *
  *   This library is free software; you can redistribute it and/or modify
@@ -19,7 +19,6 @@
  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  */
 #include <linux/fs.h>
-#include <linux/buffer_head.h>
 #include <linux/stat.h>
 #include <linux/pagemap.h>
 #include <asm/div64.h>
@@ -58,14 +57,14 @@ int cifs_get_inode_info_unix(struct inode **pinode,
                        if (tmp_path == NULL) {
                                return -ENOMEM;
                        }
-                       /* have to skip first of the double backslash of
+                       /* have to skip first of the double backslash of
                           UNC name */
                        strncpy(tmp_path, pTcon->treeName, MAX_TREE_SIZE);
                        strncat(tmp_path, search_path, MAX_PATHCONF);
                        rc = connect_to_dfs_path(xid, pTcon->ses,
                                                 /* treename + */ tmp_path,
-                                                cifs_sb->local_nls, 
-                                                cifs_sb->mnt_cifs_flags & 
+                                                cifs_sb->local_nls,
+                                                cifs_sb->mnt_cifs_flags &
                                                    CIFS_MOUNT_MAP_SPECIAL_CHR);
                        kfree(tmp_path);
 
@@ -82,7 +81,7 @@ int cifs_get_inode_info_unix(struct inode **pinode,
                /* get new inode */
                if (*pinode == NULL) {
                        *pinode = new_inode(sb);
-                       if (*pinode == NULL) 
+                       if (*pinode == NULL)
                                return -ENOMEM;
                        /* Is an i_ino of zero legal? */
                        /* Are there sanity checks we can use to ensure that
@@ -91,6 +90,9 @@ int cifs_get_inode_info_unix(struct inode **pinode,
                                (*pinode)->i_ino =
                                        (unsigned long)findData.UniqueId;
                        } /* note ino incremented to unique num in new_inode */
+                       if (sb->s_flags & MS_NOATIME)
+                               (*pinode)->i_flags |= S_NOATIME | S_NOCMTIME;
+
                        insert_inode_hash(*pinode);
                }
 
@@ -101,7 +103,7 @@ int cifs_get_inode_info_unix(struct inode **pinode,
                cifsInfo->time = jiffies;
                cFYI(1, ("New time %ld", cifsInfo->time));
                /* this is ok to set on every inode revalidate */
-               atomic_set(&cifsInfo->inUse,1);
+               atomic_set(&cifsInfo->inUse, 1);
 
                inode->i_atime =
                    cifs_NTtimeToUnix(le64_to_cpu(findData.LastAccessTime));
@@ -112,8 +114,8 @@ int cifs_get_inode_info_unix(struct inode **pinode,
                    cifs_NTtimeToUnix(le64_to_cpu(findData.LastStatusChange));
                inode->i_mode = le64_to_cpu(findData.Permissions);
                /* since we set the inode type below we need to mask off
-                   to avoid strange results if bits set above */
-                        inode->i_mode &= ~S_IFMT;
+                  to avoid strange results if bits set above */
+                       inode->i_mode &= ~S_IFMT;
                if (type == UNIX_FILE) {
                        inode->i_mode |= S_IFREG;
                } else if (type == UNIX_SYMLINK) {
@@ -135,16 +137,25 @@ int cifs_get_inode_info_unix(struct inode **pinode,
                } else {
                        /* safest to call it a file if we do not know */
                        inode->i_mode |= S_IFREG;
-                       cFYI(1,("unknown type %d",type));
+                       cFYI(1, ("unknown type %d", type));
                }
-               inode->i_uid = le64_to_cpu(findData.Uid);
-               inode->i_gid = le64_to_cpu(findData.Gid);
+
+               if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_OVERR_UID)
+                       inode->i_uid = cifs_sb->mnt_uid;
+               else
+                       inode->i_uid = le64_to_cpu(findData.Uid);
+
+               if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_OVERR_GID)
+                       inode->i_gid = cifs_sb->mnt_gid;
+               else
+                       inode->i_gid = le64_to_cpu(findData.Gid);
+
                inode->i_nlink = le64_to_cpu(findData.Nlinks);
 
-               if (is_size_safe_to_change(cifsInfo)) {
+               spin_lock(&inode->i_lock);
+               if (is_size_safe_to_change(cifsInfo, end_of_file)) {
                /* can not safely change the file size here if the
                   client is writing to it due to potential races */
-
                        i_size_write(inode, end_of_file);
 
                /* blksize needs to be multiple of two. So safer to default to
@@ -160,6 +171,7 @@ int cifs_get_inode_info_unix(struct inode **pinode,
                /* for this calculation */
                        inode->i_blocks = (512 - 1 + num_of_bytes) >> 9;
                }
+               spin_unlock(&inode->i_lock);
 
                if (num_of_bytes < end_of_file)
                        cFYI(1, ("allocation size less than end of file"));
@@ -171,17 +183,17 @@ int cifs_get_inode_info_unix(struct inode **pinode,
                        inode->i_op = &cifs_file_inode_ops;
                        if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
                                if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NO_BRL)
-                                       inode->i_fop = 
+                                       inode->i_fop =
                                                &cifs_file_direct_nobrl_ops;
                                else
                                        inode->i_fop = &cifs_file_direct_ops;
-                       } else if(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NO_BRL)
+                       } else if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NO_BRL)
                                inode->i_fop = &cifs_file_nobrl_ops;
-                       else /* not direct, send byte range locks */ 
+                       else /* not direct, send byte range locks */
                                inode->i_fop = &cifs_file_ops;
 
                        /* check if server can support readpages */
-                       if(pTcon->ses->server->maxBuf < 
+                       if (pTcon->ses->server->maxBuf <
                            PAGE_CACHE_SIZE + MAX_CIFS_HDR_SIZE)
                                inode->i_data.a_ops = &cifs_addr_ops_smallbuf;
                        else
@@ -203,7 +215,7 @@ int cifs_get_inode_info_unix(struct inode **pinode,
        return rc;
 }
 
-static int decode_sfu_inode(struct inode * inode, __u64 size,
+static int decode_sfu_inode(struct inode *inode, __u64 size,
                            const unsigned char *path,
                            struct cifs_sb_info *cifs_sb, int xid)
 {
@@ -213,34 +225,34 @@ static int decode_sfu_inode(struct inode * inode, __u64 size,
        struct cifsTconInfo *pTcon = cifs_sb->tcon;
        char buf[24];
        unsigned int bytes_read;
-       char * pbuf;
+       char *pbuf;
 
        pbuf = buf;
 
-       if(size == 0) {
+       if (size == 0) {
                inode->i_mode |= S_IFIFO;
                return 0;
        } else if (size < 8) {
                return -EINVAL;  /* EOPNOTSUPP? */
        }
-               
+
        rc = CIFSSMBOpen(xid, pTcon, path, FILE_OPEN, GENERIC_READ,
                         CREATE_NOT_DIR, &netfid, &oplock, NULL,
                         cifs_sb->local_nls,
                         cifs_sb->mnt_cifs_flags &
                                CIFS_MOUNT_MAP_SPECIAL_CHR);
-       if (rc==0) {
+       if (rc == 0) {
                int buf_type = CIFS_NO_BUFFER;
                        /* Read header */
                rc = CIFSSMBRead(xid, pTcon,
-                                netfid,
+                                netfid,
                                 24 /* length */, 0 /* offset */,
                                 &bytes_read, &pbuf, &buf_type);
-               if((rc == 0) && (bytes_read >= 8)) {
-                       if(memcmp("IntxBLK", pbuf, 8) == 0) {
-                               cFYI(1,("Block device"));
+               if ((rc == 0) && (bytes_read >= 8)) {
+                       if (memcmp("IntxBLK", pbuf, 8) == 0) {
+                               cFYI(1, ("Block device"));
                                inode->i_mode |= S_IFBLK;
-                               if(bytes_read == 24) {
+                               if (bytes_read == 24) {
                                        /* we have enough to decode dev num */
                                        __u64 mjr; /* major */
                                        __u64 mnr; /* minor */
@@ -248,37 +260,36 @@ static int decode_sfu_inode(struct inode * inode, __u64 size,
                                        mnr = le64_to_cpu(*(__le64 *)(pbuf+16));
                                        inode->i_rdev = MKDEV(mjr, mnr);
                                }
-                       } else if(memcmp("IntxCHR", pbuf, 8) == 0) {
-                               cFYI(1,("Char device"));
+                       } else if (memcmp("IntxCHR", pbuf, 8) == 0) {
+                               cFYI(1, ("Char device"));
                                inode->i_mode |= S_IFCHR;
-                               if(bytes_read == 24) {
+                               if (bytes_read == 24) {
                                        /* we have enough to decode dev num */
                                        __u64 mjr; /* major */
                                        __u64 mnr; /* minor */
                                        mjr = le64_to_cpu(*(__le64 *)(pbuf+8));
                                        mnr = le64_to_cpu(*(__le64 *)(pbuf+16));
                                        inode->i_rdev = MKDEV(mjr, mnr);
-                                }
-                       } else if(memcmp("IntxLNK", pbuf, 7) == 0) {
-                               cFYI(1,("Symlink"));
+                               }
+                       } else if (memcmp("IntxLNK", pbuf, 7) == 0) {
+                               cFYI(1, ("Symlink"));
                                inode->i_mode |= S_IFLNK;
                        } else {
                                inode->i_mode |= S_IFREG; /* file? */
-                               rc = -EOPNOTSUPP; 
+                               rc = -EOPNOTSUPP;
                        }
                } else {
                        inode->i_mode |= S_IFREG; /* then it is a file */
-                       rc = -EOPNOTSUPP; /* or some unknown SFU type */        
-               }               
+                       rc = -EOPNOTSUPP; /* or some unknown SFU type */
+               }
                CIFSSMBClose(xid, pTcon, netfid);
        }
        return rc;
-       
 }
 
 #define SFBITS_MASK (S_ISVTX | S_ISGID | S_ISUID)  /* SETFILEBITS valid bits */
 
-static int get_sfu_uid_mode(struct inode * inode,
+static int get_sfu_uid_mode(struct inode *inode,
                        const unsigned char *path,
                        struct cifs_sb_info *cifs_sb, int xid)
 {
@@ -289,15 +300,15 @@ static int get_sfu_uid_mode(struct inode * inode,
 
        rc = CIFSSMBQueryEA(xid, cifs_sb->tcon, path, "SETFILEBITS",
                        ea_value, 4 /* size of buf */, cifs_sb->local_nls,
-                        cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MAP_SPECIAL_CHR);
-       if(rc < 0)
+               cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MAP_SPECIAL_CHR);
+       if (rc < 0)
                return (int)rc;
        else if (rc > 3) {
                mode = le32_to_cpu(*((__le32 *)ea_value));
-               inode->i_mode &= ~SFBITS_MASK; 
-               cFYI(1,("special bits 0%o org mode 0%o", mode, inode->i_mode));
+               inode->i_mode &= ~SFBITS_MASK;
+               cFYI(1, ("special bits 0%o org mode 0%o", mode, inode->i_mode));
                inode->i_mode = (mode &  SFBITS_MASK) | inode->i_mode;
-               cFYI(1,("special mode bits 0%o", mode));
+               cFYI(1, ("special mode bits 0%o", mode));
                return 0;
        } else {
                return 0;
@@ -305,8 +316,6 @@ static int get_sfu_uid_mode(struct inode * inode,
 #else
        return -EOPNOTSUPP;
 #endif
-
-               
 }
 
 int cifs_get_inode_info(struct inode **pinode,
@@ -319,13 +328,14 @@ int cifs_get_inode_info(struct inode **pinode,
        struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
        char *tmp_path;
        char *buf = NULL;
+       int adjustTZ = FALSE;
 
        pTcon = cifs_sb->tcon;
-       cFYI(1,("Getting info on %s", search_path));
+       cFYI(1, ("Getting info on %s", search_path));
 
        if ((pfindData == NULL) && (*pinode != NULL)) {
                if (CIFS_I(*pinode)->clientCanCacheRead) {
-                       cFYI(1,("No need to revalidate cached inode sizes"));
+                       cFYI(1, ("No need to revalidate cached inode sizes"));
                        return rc;
                }
        }
@@ -338,18 +348,19 @@ int cifs_get_inode_info(struct inode **pinode,
                pfindData = (FILE_ALL_INFO *)buf;
                /* could do find first instead but this returns more info */
                rc = CIFSSMBQPathInfo(xid, pTcon, search_path, pfindData,
+                             0 /* not legacy */,
                              cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
                                CIFS_MOUNT_MAP_SPECIAL_CHR);
                /* BB optimize code so we do not make the above call
                when server claims no NT SMB support and the above call
                failed at least once - set flag in tcon or mount */
-               if((rc == -EOPNOTSUPP) || (rc == -EINVAL)) {
+               if ((rc == -EOPNOTSUPP) || (rc == -EINVAL)) {
                        rc = SMBQueryInformation(xid, pTcon, search_path,
-                                       pfindData, cifs_sb->local_nls, 
+                                       pfindData, cifs_sb->local_nls,
                                        cifs_sb->mnt_cifs_flags &
                                          CIFS_MOUNT_MAP_SPECIAL_CHR);
+                       adjustTZ = TRUE;
                }
-               
        }
        /* dump_mem("\nQPathInfo return data",&findData, sizeof(findData)); */
        if (rc) {
@@ -369,8 +380,8 @@ int cifs_get_inode_info(struct inode **pinode,
                        strncat(tmp_path, search_path, MAX_PATHCONF);
                        rc = connect_to_dfs_path(xid, pTcon->ses,
                                                 /* treename + */ tmp_path,
-                                                cifs_sb->local_nls, 
-                                                cifs_sb->mnt_cifs_flags & 
+                                                cifs_sb->local_nls,
+                                                cifs_sb->mnt_cifs_flags &
                                                   CIFS_MOUNT_MAP_SPECIAL_CHR);
                        kfree(tmp_path);
                        /* BB fix up inode etc. */
@@ -385,8 +396,10 @@ int cifs_get_inode_info(struct inode **pinode,
                /* get new inode */
                if (*pinode == NULL) {
                        *pinode = new_inode(sb);
-                       if (*pinode == NULL)
+                       if (*pinode == NULL) {
+                               kfree(buf);
                                return -ENOMEM;
+                       }
                        /* Is an i_ino of zero legal? Can we use that to check
                           if the server supports returning inode numbers?  Are
                           there other sanity checks we can use to ensure that
@@ -402,21 +415,23 @@ int cifs_get_inode_info(struct inode **pinode,
                           there Windows server or network appliances for which
                           IndexNumber field is not guaranteed unique? */
 
-                       if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_SERVER_INUM){
+                       if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_SERVER_INUM) {
                                int rc1 = 0;
                                __u64 inode_num;
 
-                               rc1 = CIFSGetSrvInodeNumber(xid, pTcon, 
-                                       search_path, &inode_num, 
+                               rc1 = CIFSGetSrvInodeNumber(xid, pTcon,
+                                       search_path, &inode_num,
                                        cifs_sb->local_nls,
                                        cifs_sb->mnt_cifs_flags &
                                                CIFS_MOUNT_MAP_SPECIAL_CHR);
                                if (rc1) {
-                                       cFYI(1,("GetSrvInodeNum rc %d", rc1));
+                                       cFYI(1, ("GetSrvInodeNum rc %d", rc1));
                                        /* BB EOPNOSUPP disable SERVER_INUM? */
                                } else /* do we need cast or hash to ino? */
                                        (*pinode)->i_ino = inode_num;
                        } /* else ino incremented to unique num in new_inode*/
+                       if (sb->s_flags & MS_NOATIME)
+                               (*pinode)->i_flags |= S_NOATIME | S_NOCMTIME;
                        insert_inode_hash(*pinode);
                }
                inode = *pinode;
@@ -432,21 +447,29 @@ int cifs_get_inode_info(struct inode **pinode,
                (pTcon->ses->server->maxBuf - MAX_CIFS_HDR_SIZE) & 0xFFFFFE00;*/
 
                /* Linux can not store file creation time so ignore it */
-               inode->i_atime =
-                   cifs_NTtimeToUnix(le64_to_cpu(pfindData->LastAccessTime));
+               if (pfindData->LastAccessTime)
+                       inode->i_atime = cifs_NTtimeToUnix
+                               (le64_to_cpu(pfindData->LastAccessTime));
+               else /* do not need to use current_fs_time - time not stored */
+                       inode->i_atime = CURRENT_TIME;
                inode->i_mtime =
                    cifs_NTtimeToUnix(le64_to_cpu(pfindData->LastWriteTime));
                inode->i_ctime =
                    cifs_NTtimeToUnix(le64_to_cpu(pfindData->ChangeTime));
                cFYI(0, ("Attributes came in as 0x%x", attr));
+               if (adjustTZ && (pTcon->ses) && (pTcon->ses->server)) {
+                       inode->i_ctime.tv_sec += pTcon->ses->server->timeAdj;
+                       inode->i_mtime.tv_sec += pTcon->ses->server->timeAdj;
+               }
 
                /* set default mode. will override for dirs below */
                if (atomic_read(&cifsInfo->inUse) == 0)
                        /* new inode, can safely set these fields */
                        inode->i_mode = cifs_sb->mnt_file_mode;
                else /* since we set the inode type below we need to mask off
-                    to avoid strange results if type changes and both get orred in */ 
-                       inode->i_mode &= ~S_IFMT; 
+                    to avoid strange results if type changes and both
+                    get orred in */
+                       inode->i_mode &= ~S_IFMT;
 /*             if (attr & ATTR_REPARSE)  */
                /* We no longer handle these as symlinks because we could not
                   follow them due to the absolute path with drive letter */
@@ -464,46 +487,57 @@ int cifs_get_inode_info(struct inode **pinode,
 /* BB Finish for SFU style symlinks and devices */
                } else if ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_UNX_EMUL) &&
                           (cifsInfo->cifsAttrs & ATTR_SYSTEM)) {
-                       if (decode_sfu_inode(inode, 
+                       if (decode_sfu_inode(inode,
                                         le64_to_cpu(pfindData->EndOfFile),
                                         search_path,
                                         cifs_sb, xid)) {
-                               cFYI(1,("Unrecognized sfu inode type"));
+                               cFYI(1, ("Unrecognized sfu inode type"));
                        }
-                       cFYI(1,("sfu mode 0%o",inode->i_mode));
+                       cFYI(1, ("sfu mode 0%o", inode->i_mode));
                } else {
                        inode->i_mode |= S_IFREG;
                        /* treat the dos attribute of read-only as read-only
                           mode e.g. 555 */
                        if (cifsInfo->cifsAttrs & ATTR_READONLY)
                                inode->i_mode &= ~(S_IWUGO);
+                       else if ((inode->i_mode & S_IWUGO) == 0)
+                               /* the ATTR_READONLY flag may have been */
+                               /* changed on server -- set any w bits  */
+                               /* allowed by mnt_file_mode             */
+                               inode->i_mode |= (S_IWUGO &
+                                                 cifs_sb->mnt_file_mode);
                /* BB add code here -
                   validate if device or weird share or device type? */
                }
-               if (is_size_safe_to_change(cifsInfo)) {
-                       /* can not safely change the file size here if the
+
+               spin_lock(&inode->i_lock);
+               if (is_size_safe_to_change(cifsInfo, le64_to_cpu(pfindData->EndOfFile))) {
+                       /* can not safely shrink the file size here if the
                           client is writing to it due to potential races */
-                       i_size_write(inode,le64_to_cpu(pfindData->EndOfFile));
+                       i_size_write(inode, le64_to_cpu(pfindData->EndOfFile));
 
                        /* 512 bytes (2**9) is the fake blocksize that must be
                           used for this calculation */
                        inode->i_blocks = (512 - 1 + le64_to_cpu(
                                           pfindData->AllocationSize)) >> 9;
                }
+               spin_unlock(&inode->i_lock);
 
                inode->i_nlink = le32_to_cpu(pfindData->NumberOfLinks);
 
-               /* BB fill in uid and gid here? with help from winbind? 
+               /* BB fill in uid and gid here? with help from winbind?
                   or retrieve from NTFS stream extended attribute */
-               if(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_UNX_EMUL) {
+               if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_UNX_EMUL) {
                        /* fill in uid, gid, mode from server ACL */
+                       /* BB FIXME this should also take into account the
+                        * default uid specified on mount if present */
                        get_sfu_uid_mode(inode, search_path, cifs_sb, xid);
                } else if (atomic_read(&cifsInfo->inUse) == 0) {
                        inode->i_uid = cifs_sb->mnt_uid;
                        inode->i_gid = cifs_sb->mnt_gid;
                        /* set so we do not keep refreshing these fields with
                           bad data after user has changed them in memory */
-                       atomic_set(&cifsInfo->inUse,1);
+                       atomic_set(&cifsInfo->inUse, 1);
                }
 
                if (S_ISREG(inode->i_mode)) {
@@ -515,12 +549,12 @@ int cifs_get_inode_info(struct inode **pinode,
                                                &cifs_file_direct_nobrl_ops;
                                else
                                        inode->i_fop = &cifs_file_direct_ops;
-                       } else if(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NO_BRL)
+                       } else if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NO_BRL)
                                inode->i_fop = &cifs_file_nobrl_ops;
                        else /* not direct, send byte range locks */
                                inode->i_fop = &cifs_file_ops;
 
-                       if(pTcon->ses->server->maxBuf < 
+                       if (pTcon->ses->server->maxBuf <
                             PAGE_CACHE_SIZE + MAX_CIFS_HDR_SIZE)
                                inode->i_data.a_ops = &cifs_addr_ops_smallbuf;
                        else
@@ -549,10 +583,11 @@ void cifs_read_inode(struct inode *inode)
 
        cifs_sb = CIFS_SB(inode->i_sb);
        xid = GetXid();
-       if (cifs_sb->tcon->ses->capabilities & CAP_UNIX)
-               cifs_get_inode_info_unix(&inode, "", inode->i_sb,xid);
+
+       if (cifs_sb->tcon->unix_ext)
+               cifs_get_inode_info_unix(&inode, "", inode->i_sb, xid);
        else
-               cifs_get_inode_info(&inode, "", NULL, inode->i_sb,xid);
+               cifs_get_inode_info(&inode, "", NULL, inode->i_sb, xid);
        /* can not call macro FreeXid here since in a void func */
        _FreeXid(xid);
 }
@@ -571,7 +606,7 @@ int cifs_unlink(struct inode *inode, struct dentry *direntry)
 
        xid = GetXid();
 
-       if(inode)
+       if (inode)
                cifs_sb = CIFS_SB(inode->i_sb);
        else
                cifs_sb = CIFS_SB(direntry->d_sb);
@@ -586,12 +621,24 @@ int cifs_unlink(struct inode *inode, struct dentry *direntry)
                FreeXid(xid);
                return -ENOMEM;
        }
-       rc = CIFSSMBDelFile(xid, pTcon, full_path, cifs_sb->local_nls,
+
+       if ((pTcon->ses->capabilities & CAP_UNIX) &&
+               (CIFS_UNIX_POSIX_PATH_OPS_CAP &
+                       le64_to_cpu(pTcon->fsUnixInfo.Capability))) {
+               rc = CIFSPOSIXDelFile(xid, pTcon, full_path,
+                       SMB_POSIX_UNLINK_FILE_TARGET, cifs_sb->local_nls,
                        cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MAP_SPECIAL_CHR);
+               cFYI(1, ("posix del rc %d", rc));
+               if ((rc == 0) || (rc == -ENOENT))
+                       goto psx_del_no_retry;
+       }
 
+       rc = CIFSSMBDelFile(xid, pTcon, full_path, cifs_sb->local_nls,
+                       cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MAP_SPECIAL_CHR);
+psx_del_no_retry:
        if (!rc) {
                if (direntry->d_inode)
-                       direntry->d_inode->i_nlink--;
+                       drop_nlink(direntry->d_inode);
        } else if (rc == -ENOENT) {
                d_drop(direntry);
        } else if (rc == -ETXTBSY) {
@@ -601,16 +648,16 @@ int cifs_unlink(struct inode *inode, struct dentry *direntry)
                rc = CIFSSMBOpen(xid, pTcon, full_path, FILE_OPEN, DELETE,
                                 CREATE_NOT_DIR | CREATE_DELETE_ON_CLOSE,
                                 &netfid, &oplock, NULL, cifs_sb->local_nls,
-                                cifs_sb->mnt_cifs_flags & 
+                                cifs_sb->mnt_cifs_flags &
                                        CIFS_MOUNT_MAP_SPECIAL_CHR);
-               if (rc==0) {
+               if (rc == 0) {
                        CIFSSMBRenameOpenFile(xid, pTcon, netfid, NULL,
-                                             cifs_sb->local_nls, 
-                                             cifs_sb->mnt_cifs_flags & 
+                                             cifs_sb->local_nls,
+                                             cifs_sb->mnt_cifs_flags &
                                                CIFS_MOUNT_MAP_SPECIAL_CHR);
                        CIFSSMBClose(xid, pTcon, netfid);
                        if (direntry->d_inode)
-                               direntry->d_inode->i_nlink--;
+                               drop_nlink(direntry->d_inode);
                }
        } else if (rc == -EACCES) {
                /* try only if r/o attribute set in local lookup data? */
@@ -622,7 +669,7 @@ int cifs_unlink(struct inode *inode, struct dentry *direntry)
                                rc = CIFSSMBSetTimes(xid, pTcon, full_path,
                                                     pinfo_buf,
                                                     cifs_sb->local_nls,
-                                                    cifs_sb->mnt_cifs_flags & 
+                                                    cifs_sb->mnt_cifs_flags &
                                                        CIFS_MOUNT_MAP_SPECIAL_CHR);
                        else
                                rc = -EOPNOTSUPP;
@@ -633,7 +680,7 @@ int cifs_unlink(struct inode *inode, struct dentry *direntry)
                        /*      rc = CIFSSMBSetAttrLegacy(xid, pTcon,
                                                          full_path,
                                                          (__u16)ATTR_NORMAL,
-                                                         cifs_sb->local_nls); 
+                                                         cifs_sb->local_nls);
                           For some strange reason it seems that NT4 eats the
                           old setattr call without actually setting the
                           attributes so on to the third attempted workaround
@@ -646,9 +693,9 @@ int cifs_unlink(struct inode *inode, struct dentry *direntry)
                                                 FILE_WRITE_ATTRIBUTES, 0,
                                                 &netfid, &oplock, NULL,
                                                 cifs_sb->local_nls,
-                                                cifs_sb->mnt_cifs_flags & 
+                                                cifs_sb->mnt_cifs_flags &
                                                    CIFS_MOUNT_MAP_SPECIAL_CHR);
-                               if (rc==0) {
+                               if (rc == 0) {
                                        rc = CIFSSMBSetFileTimes(xid, pTcon,
                                                                 pinfo_buf,
                                                                 netfid);
@@ -657,14 +704,14 @@ int cifs_unlink(struct inode *inode, struct dentry *direntry)
                        }
                        kfree(pinfo_buf);
                }
-               if (rc==0) {
-                       rc = CIFSSMBDelFile(xid, pTcon, full_path, 
-                                           cifs_sb->local_nls, 
-                                           cifs_sb->mnt_cifs_flags & 
+               if (rc == 0) {
+                       rc = CIFSSMBDelFile(xid, pTcon, full_path,
+                                           cifs_sb->local_nls,
+                                           cifs_sb->mnt_cifs_flags &
                                                CIFS_MOUNT_MAP_SPECIAL_CHR);
                        if (!rc) {
                                if (direntry->d_inode)
-                                       direntry->d_inode->i_nlink--;
+                                       drop_nlink(direntry->d_inode);
                        } else if (rc == -ETXTBSY) {
                                int oplock = FALSE;
                                __u16 netfid;
@@ -674,10 +721,10 @@ int cifs_unlink(struct inode *inode, struct dentry *direntry)
                                                 CREATE_NOT_DIR |
                                                 CREATE_DELETE_ON_CLOSE,
                                                 &netfid, &oplock, NULL,
-                                                cifs_sb->local_nls, 
-                                                cifs_sb->mnt_cifs_flags & 
+                                                cifs_sb->local_nls,
+                                                cifs_sb->mnt_cifs_flags &
                                                    CIFS_MOUNT_MAP_SPECIAL_CHR);
-                               if (rc==0) {
+                               if (rc == 0) {
                                        CIFSSMBRenameOpenFile(xid, pTcon,
                                                netfid, NULL,
                                                cifs_sb->local_nls,
@@ -685,7 +732,7 @@ int cifs_unlink(struct inode *inode, struct dentry *direntry)
                                                    CIFS_MOUNT_MAP_SPECIAL_CHR);
                                        CIFSSMBClose(xid, pTcon, netfid);
                                        if (direntry->d_inode)
-                                               direntry->d_inode->i_nlink--;
+                                               drop_nlink(direntry->d_inode);
                                }
                        /* BB if rc = -ETXTBUSY goto the rename logic BB */
                        }
@@ -697,7 +744,7 @@ int cifs_unlink(struct inode *inode, struct dentry *direntry)
                                           when needed */
                direntry->d_inode->i_ctime = current_fs_time(inode->i_sb);
        }
-       if(inode) {
+       if (inode) {
                inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
                cifsInode = CIFS_I(inode);
                cifsInode->time = 0;    /* force revalidate of dir as well */
@@ -708,6 +755,137 @@ int cifs_unlink(struct inode *inode, struct dentry *direntry)
        return rc;
 }
 
+static void posix_fill_in_inode(struct inode *tmp_inode,
+       FILE_UNIX_BASIC_INFO *pData, int *pobject_type, int isNewInode)
+{
+       loff_t local_size;
+       struct timespec local_mtime;
+
+       struct cifsInodeInfo *cifsInfo = CIFS_I(tmp_inode);
+       struct cifs_sb_info *cifs_sb = CIFS_SB(tmp_inode->i_sb);
+
+       __u32 type = le32_to_cpu(pData->Type);
+       __u64 num_of_bytes = le64_to_cpu(pData->NumOfBytes);
+       __u64 end_of_file = le64_to_cpu(pData->EndOfFile);
+       cifsInfo->time = jiffies;
+       atomic_inc(&cifsInfo->inUse);
+
+       /* save mtime and size */
+       local_mtime = tmp_inode->i_mtime;
+       local_size  = tmp_inode->i_size;
+
+       tmp_inode->i_atime =
+           cifs_NTtimeToUnix(le64_to_cpu(pData->LastAccessTime));
+       tmp_inode->i_mtime =
+           cifs_NTtimeToUnix(le64_to_cpu(pData->LastModificationTime));
+       tmp_inode->i_ctime =
+           cifs_NTtimeToUnix(le64_to_cpu(pData->LastStatusChange));
+
+       tmp_inode->i_mode = le64_to_cpu(pData->Permissions);
+       /* since we set the inode type below we need to mask off type
+          to avoid strange results if bits above were corrupt */
+       tmp_inode->i_mode &= ~S_IFMT;
+       if (type == UNIX_FILE) {
+               *pobject_type = DT_REG;
+               tmp_inode->i_mode |= S_IFREG;
+       } else if (type == UNIX_SYMLINK) {
+               *pobject_type = DT_LNK;
+               tmp_inode->i_mode |= S_IFLNK;
+       } else if (type == UNIX_DIR) {
+               *pobject_type = DT_DIR;
+               tmp_inode->i_mode |= S_IFDIR;
+       } else if (type == UNIX_CHARDEV) {
+               *pobject_type = DT_CHR;
+               tmp_inode->i_mode |= S_IFCHR;
+               tmp_inode->i_rdev = MKDEV(le64_to_cpu(pData->DevMajor),
+                               le64_to_cpu(pData->DevMinor) & MINORMASK);
+       } else if (type == UNIX_BLOCKDEV) {
+               *pobject_type = DT_BLK;
+               tmp_inode->i_mode |= S_IFBLK;
+               tmp_inode->i_rdev = MKDEV(le64_to_cpu(pData->DevMajor),
+                               le64_to_cpu(pData->DevMinor) & MINORMASK);
+       } else if (type == UNIX_FIFO) {
+               *pobject_type = DT_FIFO;
+               tmp_inode->i_mode |= S_IFIFO;
+       } else if (type == UNIX_SOCKET) {
+               *pobject_type = DT_SOCK;
+               tmp_inode->i_mode |= S_IFSOCK;
+       } else {
+               /* safest to just call it a file */
+               *pobject_type = DT_REG;
+               tmp_inode->i_mode |= S_IFREG;
+               cFYI(1, ("unknown inode type %d", type));
+       }
+
+#ifdef CONFIG_CIFS_DEBUG2
+       cFYI(1, ("object type: %d", type));
+#endif
+       tmp_inode->i_uid = le64_to_cpu(pData->Uid);
+       tmp_inode->i_gid = le64_to_cpu(pData->Gid);
+       tmp_inode->i_nlink = le64_to_cpu(pData->Nlinks);
+
+       spin_lock(&tmp_inode->i_lock);
+       if (is_size_safe_to_change(cifsInfo, end_of_file)) {
+               /* can not safely change the file size here if the
+               client is writing to it due to potential races */
+               i_size_write(tmp_inode, end_of_file);
+
+       /* 512 bytes (2**9) is the fake blocksize that must be used */
+       /* for this calculation, not the real blocksize */
+               tmp_inode->i_blocks = (512 - 1 + num_of_bytes) >> 9;
+       }
+       spin_unlock(&tmp_inode->i_lock);
+
+       if (S_ISREG(tmp_inode->i_mode)) {
+               cFYI(1, ("File inode"));
+               tmp_inode->i_op = &cifs_file_inode_ops;
+
+               if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
+                       if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NO_BRL)
+                               tmp_inode->i_fop = &cifs_file_direct_nobrl_ops;
+                       else
+                               tmp_inode->i_fop = &cifs_file_direct_ops;
+
+               } else if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NO_BRL)
+                       tmp_inode->i_fop = &cifs_file_nobrl_ops;
+               else
+                       tmp_inode->i_fop = &cifs_file_ops;
+
+               if ((cifs_sb->tcon) && (cifs_sb->tcon->ses) &&
+                  (cifs_sb->tcon->ses->server->maxBuf <
+                       PAGE_CACHE_SIZE + MAX_CIFS_HDR_SIZE))
+                       tmp_inode->i_data.a_ops = &cifs_addr_ops_smallbuf;
+               else
+                       tmp_inode->i_data.a_ops = &cifs_addr_ops;
+
+               if (isNewInode)
+                       return; /* No sense invalidating pages for new inode
+                                  since we we have not started caching
+                                  readahead file data yet */
+
+               if (timespec_equal(&tmp_inode->i_mtime, &local_mtime) &&
+                       (local_size == tmp_inode->i_size)) {
+                       cFYI(1, ("inode exists but unchanged"));
+               } else {
+                       /* file may have changed on server */
+                       cFYI(1, ("invalidate inode, readdir detected change"));
+                       invalidate_remote_inode(tmp_inode);
+               }
+       } else if (S_ISDIR(tmp_inode->i_mode)) {
+               cFYI(1, ("Directory inode"));
+               tmp_inode->i_op = &cifs_dir_inode_ops;
+               tmp_inode->i_fop = &cifs_dir_ops;
+       } else if (S_ISLNK(tmp_inode->i_mode)) {
+               cFYI(1, ("Symbolic Link inode"));
+               tmp_inode->i_op = &cifs_symlink_inode_ops;
+/* tmp_inode->i_fop = *//* do not need to set to anything */
+       } else {
+               cFYI(1, ("Special inode"));
+               init_special_inode(tmp_inode, tmp_inode->i_mode,
+                                  tmp_inode->i_rdev);
+       }
+}
+
 int cifs_mkdir(struct inode *inode, struct dentry *direntry, int mode)
 {
        int rc = 0;
@@ -729,6 +907,72 @@ int cifs_mkdir(struct inode *inode, struct dentry *direntry, int mode)
                FreeXid(xid);
                return -ENOMEM;
        }
+
+       if ((pTcon->ses->capabilities & CAP_UNIX) &&
+               (CIFS_UNIX_POSIX_PATH_OPS_CAP &
+                       le64_to_cpu(pTcon->fsUnixInfo.Capability))) {
+               u32 oplock = 0;
+               FILE_UNIX_BASIC_INFO * pInfo =
+                       kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
+               if (pInfo == NULL) {
+                       rc = -ENOMEM;
+                       goto mkdir_out;
+               }
+
+               rc = CIFSPOSIXCreate(xid, pTcon, SMB_O_DIRECTORY | SMB_O_CREAT,
+                               mode, NULL /* netfid */, pInfo, &oplock,
+                               full_path, cifs_sb->local_nls,
+                               cifs_sb->mnt_cifs_flags &
+                                       CIFS_MOUNT_MAP_SPECIAL_CHR);
+               if (rc) {
+                       cFYI(1, ("posix mkdir returned 0x%x", rc));
+                       d_drop(direntry);
+               } else {
+                       int obj_type;
+                       if (pInfo->Type == -1) /* no return info - go query */
+                               goto mkdir_get_info;
+/*BB check (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_SET_UID ) to see if need
+       to set uid/gid */
+                       inc_nlink(inode);
+                       if (pTcon->nocase)
+                               direntry->d_op = &cifs_ci_dentry_ops;
+                       else
+                               direntry->d_op = &cifs_dentry_ops;
+
+                       newinode = new_inode(inode->i_sb);
+                       if (newinode == NULL)
+                               goto mkdir_get_info;
+                       /* Is an i_ino of zero legal? */
+                       /* Are there sanity checks we can use to ensure that
+                          the server is really filling in that field? */
+                       if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_SERVER_INUM) {
+                               newinode->i_ino =
+                                       (unsigned long)pInfo->UniqueId;
+                       } /* note ino incremented to unique num in new_inode */
+                       if (inode->i_sb->s_flags & MS_NOATIME)
+                               newinode->i_flags |= S_NOATIME | S_NOCMTIME;
+                       newinode->i_nlink = 2;
+
+                       insert_inode_hash(newinode);
+                       d_instantiate(direntry, newinode);
+
+                       /* we already checked in POSIXCreate whether
+                          frame was long enough */
+                       posix_fill_in_inode(direntry->d_inode,
+                                       pInfo, &obj_type, 1 /* NewInode */);
+#ifdef CONFIG_CIFS_DEBUG2
+                       cFYI(1, ("instantiated dentry %p %s to inode %p",
+                               direntry, direntry->d_name.name, newinode));
+
+                       if (newinode->i_nlink != 2)
+                               cFYI(1, ("unexpected number of links %d",
+                                       newinode->i_nlink));
+#endif
+               }
+               kfree(pInfo);
+               goto mkdir_out;
+       }
+
        /* BB add setting the equivalent of mode via CreateX w/ACLs */
        rc = CIFSSMBMkDir(xid, pTcon, full_path, cifs_sb->local_nls,
                          cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MAP_SPECIAL_CHR);
@@ -736,22 +980,26 @@ int cifs_mkdir(struct inode *inode, struct dentry *direntry, int mode)
                cFYI(1, ("cifs_mkdir returned 0x%x", rc));
                d_drop(direntry);
        } else {
-               inode->i_nlink++;
-               if (pTcon->ses->capabilities & CAP_UNIX)
+mkdir_get_info:
+               inc_nlink(inode);
+               if (pTcon->unix_ext)
                        rc = cifs_get_inode_info_unix(&newinode, full_path,
-                                                     inode->i_sb,xid);
+                                                     inode->i_sb, xid);
                else
                        rc = cifs_get_inode_info(&newinode, full_path, NULL,
-                                                inode->i_sb,xid);
+                                                inode->i_sb, xid);
 
                if (pTcon->nocase)
                        direntry->d_op = &cifs_ci_dentry_ops;
                else
                        direntry->d_op = &cifs_dentry_ops;
                d_instantiate(direntry, newinode);
-               if (direntry->d_inode)
-                       direntry->d_inode->i_nlink = 2;
-               if (cifs_sb->tcon->ses->capabilities & CAP_UNIX)
+                /* setting nlink not necessary except in cases where we
+                 * failed to get it from the server or was set bogus */
+               if ((direntry->d_inode) && (direntry->d_inode->i_nlink < 2))
+                               direntry->d_inode->i_nlink = 2;
+               if (pTcon->unix_ext) {
+                       mode &= ~current->fs->umask;
                        if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_SET_UID) {
                                CIFSSMBUnixSetPerms(xid, pTcon, full_path,
                                                    mode,
@@ -766,26 +1014,27 @@ int cifs_mkdir(struct inode *inode, struct dentry *direntry, int mode)
                                                    mode, (__u64)-1,
                                                    (__u64)-1, 0 /* dev_t */,
                                                    cifs_sb->local_nls,
-                                                   cifs_sb->mnt_cifs_flags & 
+                                                   cifs_sb->mnt_cifs_flags &
                                                    CIFS_MOUNT_MAP_SPECIAL_CHR);
                        }
-               else {
+               else {
                        /* BB to be implemented via Windows secrty descriptors
                           eg CIFSSMBWinSetPerms(xid, pTcon, full_path, mode,
                                                 -1, -1, local_nls); */
-                       if(direntry->d_inode) {
+                       if (direntry->d_inode) {
                                direntry->d_inode->i_mode = mode;
                                direntry->d_inode->i_mode |= S_IFDIR;
-                               if(cifs_sb->mnt_cifs_flags & 
+                               if (cifs_sb->mnt_cifs_flags &
                                     CIFS_MOUNT_SET_UID) {
-                                       direntry->d_inode->i_uid = 
+                                       direntry->d_inode->i_uid =
                                                current->fsuid;
-                                       direntry->d_inode->i_gid = 
+                                       direntry->d_inode->i_gid =
                                                current->fsgid;
                                }
                        }
                }
        }
+mkdir_out:
        kfree(full_path);
        FreeXid(xid);
        return rc;
@@ -817,9 +1066,11 @@ int cifs_rmdir(struct inode *inode, struct dentry *direntry)
                          cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MAP_SPECIAL_CHR);
 
        if (!rc) {
-               inode->i_nlink--;
-               i_size_write(direntry->d_inode,0);
-               direntry->d_inode->i_nlink = 0;
+               drop_nlink(inode);
+               spin_lock(&direntry->d_inode->i_lock);
+               i_size_write(direntry->d_inode, 0);
+               clear_nlink(direntry->d_inode);
+               spin_unlock(&direntry->d_inode->i_lock);
        }
 
        cifsInode = CIFS_I(direntry->d_inode);
@@ -880,10 +1131,14 @@ int cifs_rename(struct inode *source_inode, struct dentry *source_direntry,
                        kmalloc(2 * sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
                if (info_buf_source != NULL) {
                        info_buf_target = info_buf_source + 1;
-                       rc = CIFSSMBUnixQPathInfo(xid, pTcon, fromName,
-                               info_buf_source, cifs_sb_source->local_nls, 
-                               cifs_sb_source->mnt_cifs_flags &
-                                       CIFS_MOUNT_MAP_SPECIAL_CHR);
+                       if (pTcon->unix_ext)
+                               rc = CIFSSMBUnixQPathInfo(xid, pTcon, fromName,
+                                       info_buf_source,
+                                       cifs_sb_source->local_nls,
+                                       cifs_sb_source->mnt_cifs_flags &
+                                               CIFS_MOUNT_MAP_SPECIAL_CHR);
+                       /* else rc is still EEXIST so will fall through to
+                          unlink the target and retry rename */
                        if (rc == 0) {
                                rc = CIFSSMBUnixQPathInfo(xid, pTcon, toName,
                                                info_buf_target,
@@ -928,12 +1183,12 @@ int cifs_rename(struct inode *source_inode, struct dentry *source_direntry,
                   might not right be right access to request */
                rc = CIFSSMBOpen(xid, pTcon, fromName, FILE_OPEN, GENERIC_READ,
                                 CREATE_NOT_DIR, &netfid, &oplock, NULL,
-                                cifs_sb_source->local_nls, 
-                                cifs_sb_source->mnt_cifs_flags & 
+                                cifs_sb_source->local_nls,
+                                cifs_sb_source->mnt_cifs_flags &
                                        CIFS_MOUNT_MAP_SPECIAL_CHR);
-               if (rc==0) {
-                       CIFSSMBRenameOpenFile(xid, pTcon, netfid, toName,
-                                             cifs_sb_source->local_nls, 
+               if (rc == 0) {
+                       rc = CIFSSMBRenameOpenFile(xid, pTcon, netfid, toName,
+                                             cifs_sb_source->local_nls,
                                              cifs_sb_source->mnt_cifs_flags &
                                                CIFS_MOUNT_MAP_SPECIAL_CHR);
                        CIFSSMBClose(xid, pTcon, netfid);
@@ -1004,9 +1259,9 @@ int cifs_revalidate(struct dentry *direntry)
        local_mtime = direntry->d_inode->i_mtime;
        local_size = direntry->d_inode->i_size;
 
-       if (cifs_sb->tcon->ses->capabilities & CAP_UNIX) {
+       if (cifs_sb->tcon->unix_ext) {
                rc = cifs_get_inode_info_unix(&direntry->d_inode, full_path,
-                                             direntry->d_sb,xid);
+                                             direntry->d_sb, xid);
                if (rc) {
                        cFYI(1, ("error on getting revalidate info %d", rc));
 /*                     if (rc != -ENOENT)
@@ -1015,7 +1270,7 @@ int cifs_revalidate(struct dentry *direntry)
                }
        } else {
                rc = cifs_get_inode_info(&direntry->d_inode, full_path, NULL,
-                                        direntry->d_sb,xid);
+                                        direntry->d_sb, xid);
                if (rc) {
                        cFYI(1, ("error on getting revalidate info %d", rc));
 /*                     if (rc != -ENOENT)
@@ -1028,7 +1283,7 @@ int cifs_revalidate(struct dentry *direntry)
        /* if not oplocked, we invalidate inode pages if mtime or file size
           had changed on server */
 
-       if (timespec_equal(&local_mtime,&direntry->d_inode->i_mtime) && 
+       if (timespec_equal(&local_mtime, &direntry->d_inode->i_mtime) &&
            (local_size == direntry->d_inode->i_size)) {
                cFYI(1, ("cifs_revalidate - inode unchanged"));
        } else {
@@ -1055,7 +1310,7 @@ int cifs_revalidate(struct dentry *direntry)
        if (invalidate_inode) {
        /* shrink_dcache not necessary now that cifs dentry ops
        are exported for negative dentries */
-/*             if(S_ISDIR(direntry->d_inode->i_mode)) 
+/*             if (S_ISDIR(direntry->d_inode->i_mode))
                        shrink_dcache_parent(direntry); */
                if (S_ISREG(direntry->d_inode->i_mode)) {
                        if (direntry->d_inode->i_mapping)
@@ -1070,7 +1325,7 @@ int cifs_revalidate(struct dentry *direntry)
                }
        }
 /*     mutex_unlock(&direntry->d_inode->i_mutex); */
-       
+
        kfree(full_path);
        FreeXid(xid);
        return rc;
@@ -1080,8 +1335,10 @@ int cifs_getattr(struct vfsmount *mnt, struct dentry *dentry,
        struct kstat *stat)
 {
        int err = cifs_revalidate(dentry);
-       if (!err)
+       if (!err) {
                generic_fillattr(dentry->d_inode, stat);
+               stat->blksize = CIFS_MAX_MSGSIZE;
+       }
        return err;
 }
 
@@ -1090,22 +1347,73 @@ static int cifs_truncate_page(struct address_space *mapping, loff_t from)
        pgoff_t index = from >> PAGE_CACHE_SHIFT;
        unsigned offset = from & (PAGE_CACHE_SIZE - 1);
        struct page *page;
-       char *kaddr;
        int rc = 0;
 
        page = grab_cache_page(mapping, index);
        if (!page)
                return -ENOMEM;
 
-       kaddr = kmap_atomic(page, KM_USER0);
-       memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
-       flush_dcache_page(page);
-       kunmap_atomic(kaddr, KM_USER0);
+       zero_user_page(page, offset, PAGE_CACHE_SIZE - offset, KM_USER0);
        unlock_page(page);
        page_cache_release(page);
        return rc;
 }
 
+static int cifs_vmtruncate(struct inode *inode, loff_t offset)
+{
+       struct address_space *mapping = inode->i_mapping;
+       unsigned long limit;
+
+       spin_lock(&inode->i_lock);
+       if (inode->i_size < offset)
+               goto do_expand;
+       /*
+        * truncation of in-use swapfiles is disallowed - it would cause
+        * subsequent swapout to scribble on the now-freed blocks.
+        */
+       if (IS_SWAPFILE(inode)) {
+               spin_unlock(&inode->i_lock);
+               goto out_busy;
+       }
+       i_size_write(inode, offset);
+       spin_unlock(&inode->i_lock);
+       /*
+        * unmap_mapping_range is called twice, first simply for efficiency
+        * so that truncate_inode_pages does fewer single-page unmaps. However
+        * after this first call, and before truncate_inode_pages finishes,
+        * it is possible for private pages to be COWed, which remain after
+        * truncate_inode_pages finishes, hence the second unmap_mapping_range
+        * call must be made for correctness.
+        */
+       unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
+       truncate_inode_pages(mapping, offset);
+       unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
+       goto out_truncate;
+
+do_expand:
+       limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
+       if (limit != RLIM_INFINITY && offset > limit) {
+               spin_unlock(&inode->i_lock);
+               goto out_sig;
+       }
+       if (offset > inode->i_sb->s_maxbytes) {
+               spin_unlock(&inode->i_lock);
+               goto out_big;
+       }
+       i_size_write(inode, offset);
+       spin_unlock(&inode->i_lock);
+out_truncate:
+       if (inode->i_op && inode->i_op->truncate)
+               inode->i_op->truncate(inode);
+       return 0;
+out_sig:
+       send_sig(SIGXFSZ, current, 0);
+out_big:
+       return -EFBIG;
+out_busy:
+       return -ETXTBSY;
+}
+
 int cifs_setattr(struct dentry *direntry, struct iattr *attrs)
 {
        int xid;
@@ -1116,6 +1424,7 @@ int cifs_setattr(struct dentry *direntry, struct iattr *attrs)
        struct cifsFileInfo *open_file = NULL;
        FILE_BASIC_INFO time_buf;
        int set_time = FALSE;
+       int set_dosattr = FALSE;
        __u64 mode = 0xFFFFFFFFFFFFFFFFULL;
        __u64 uid = 0xFFFFFFFFFFFFFFFFULL;
        __u64 gid = 0xFFFFFFFFFFFFFFFFULL;
@@ -1132,13 +1441,13 @@ int cifs_setattr(struct dentry *direntry, struct iattr *attrs)
        if ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NO_PERM) == 0) {
                /* check if we have permission to change attrs */
                rc = inode_change_ok(direntry->d_inode, attrs);
-               if(rc < 0) {
+               if (rc < 0) {
                        FreeXid(xid);
                        return rc;
                } else
                        rc = 0;
        }
-               
+
        full_path = build_path_from_dentry(direntry);
        if (full_path == NULL) {
                FreeXid(xid);
@@ -1167,16 +1476,16 @@ int cifs_setattr(struct dentry *direntry, struct iattr *attrs)
                        rc = CIFSSMBSetFileSize(xid, pTcon, attrs->ia_size,
                                                nfid, npid, FALSE);
                        atomic_dec(&open_file->wrtPending);
-                       cFYI(1,("SetFSize for attrs rc = %d", rc));
-                       if((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
+                       cFYI(1, ("SetFSize for attrs rc = %d", rc));
+                       if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
                                int bytes_written;
                                rc = CIFSSMBWrite(xid, pTcon,
                                                  nfid, 0, attrs->ia_size,
                                                  &bytes_written, NULL, NULL,
                                                  1 /* 45 seconds */);
-                               cFYI(1,("Wrt seteof rc %d", rc));
+                               cFYI(1, ("Wrt seteof rc %d", rc));
                        }
-               } else 
+               } else
                        rc = -EINVAL;
 
                if (rc != 0) {
@@ -1186,11 +1495,11 @@ int cifs_setattr(struct dentry *direntry, struct iattr *attrs)
                           it by handle */
                        rc = CIFSSMBSetEOF(xid, pTcon, full_path,
                                           attrs->ia_size, FALSE,
-                                          cifs_sb->local_nls, 
+                                          cifs_sb->local_nls,
                                           cifs_sb->mnt_cifs_flags &
                                                CIFS_MOUNT_MAP_SPECIAL_CHR);
                        cFYI(1, ("SetEOF by path (setattrs) rc = %d", rc));
-                       if((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
+                       if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
                                __u16 netfid;
                                int oplock = FALSE;
 
@@ -1201,14 +1510,14 @@ int cifs_setattr(struct dentry *direntry, struct iattr *attrs)
                                        NULL, cifs_sb->local_nls,
                                        cifs_sb->mnt_cifs_flags &
                                                CIFS_MOUNT_MAP_SPECIAL_CHR);
-                               if (rc==0) {
+                               if (rc == 0) {
                                        int bytes_written;
                                        rc = CIFSSMBWrite(xid, pTcon,
                                                        netfid, 0,
                                                        attrs->ia_size,
                                                        &bytes_written, NULL,
                                                        NULL, 1 /* 45 sec */);
-                                       cFYI(1,("wrt seteof rc %d",rc));
+                                       cFYI(1, ("wrt seteof rc %d", rc));
                                        CIFSSMBClose(xid, pTcon, netfid);
                                }
 
@@ -1222,10 +1531,10 @@ int cifs_setattr(struct dentry *direntry, struct iattr *attrs)
                   */
 
                if (rc == 0) {
-                       rc = vmtruncate(direntry->d_inode, attrs->ia_size);
+                       rc = cifs_vmtruncate(direntry->d_inode, attrs->ia_size);
                        cifs_truncate_page(direntry->d_inode->i_mapping,
                                           direntry->d_inode->i_size);
-               } else 
+               } else
                        goto cifs_setattr_exit;
        }
        if (attrs->ia_valid & ATTR_UID) {
@@ -1243,24 +1552,32 @@ int cifs_setattr(struct dentry *direntry, struct iattr *attrs)
                mode = attrs->ia_mode;
        }
 
-       if ((cifs_sb->tcon->ses->capabilities & CAP_UNIX)
+       if ((pTcon->unix_ext)
            && (attrs->ia_valid & (ATTR_MODE | ATTR_GID | ATTR_UID)))
                rc = CIFSSMBUnixSetPerms(xid, pTcon, full_path, mode, uid, gid,
                                         0 /* dev_t */, cifs_sb->local_nls,
-                                        cifs_sb->mnt_cifs_flags & 
+                                        cifs_sb->mnt_cifs_flags &
                                                CIFS_MOUNT_MAP_SPECIAL_CHR);
        else if (attrs->ia_valid & ATTR_MODE) {
                rc = 0;
                if ((mode & S_IWUGO) == 0) /* not writeable */ {
-                       if ((cifsInode->cifsAttrs & ATTR_READONLY) == 0)
+                       if ((cifsInode->cifsAttrs & ATTR_READONLY) == 0) {
+                               set_dosattr = TRUE;
                                time_buf.Attributes =
                                        cpu_to_le32(cifsInode->cifsAttrs |
                                                    ATTR_READONLY);
-               } else if ((mode & S_IWUGO) == S_IWUGO) {
-                       if (cifsInode->cifsAttrs & ATTR_READONLY)
-                               time_buf.Attributes =
-                                       cpu_to_le32(cifsInode->cifsAttrs &
-                                                   (~ATTR_READONLY));
+                       }
+               } else if (cifsInode->cifsAttrs & ATTR_READONLY) {
+                       /* If file is readonly on server, we would
+                       not be able to write to it - so if any write
+                       bit is enabled for user or group or other we
+                       need to at least try to remove r/o dos attr */
+                       set_dosattr = TRUE;
+                       time_buf.Attributes = cpu_to_le32(cifsInode->cifsAttrs &
+                                           (~ATTR_READONLY));
+                       /* Windows ignores set to zero */
+                       if (time_buf.Attributes == 0)
+                               time_buf.Attributes |= cpu_to_le32(ATTR_NORMAL);
                }
                /* BB to be implemented -
                   via Windows security descriptors or streams */
@@ -1285,7 +1602,7 @@ int cifs_setattr(struct dentry *direntry, struct iattr *attrs)
           stamps are changed explicitly (i.e. by utime()
           since we would then have a mix of client and
           server times */
-          
+
        if (set_time && (attrs->ia_valid & ATTR_CTIME)) {
                set_time = TRUE;
                /* Although Samba throws this field away
@@ -1298,7 +1615,7 @@ int cifs_setattr(struct dentry *direntry, struct iattr *attrs)
        } else
                time_buf.ChangeTime = 0;
 
-       if (set_time || time_buf.Attributes) {
+       if (set_time || set_dosattr) {
                time_buf.CreationTime = 0;      /* do not change */
                /* In the future we should experiment - try setting timestamps
                   via Handle (SetFileInfo) instead of by path */
@@ -1324,7 +1641,7 @@ int cifs_setattr(struct dentry *direntry, struct iattr *attrs)
                                         NULL, cifs_sb->local_nls,
                                         cifs_sb->mnt_cifs_flags &
                                                CIFS_MOUNT_MAP_SPECIAL_CHR);
-                       if (rc==0) {
+                       if (rc == 0) {
                                rc = CIFSSMBSetFileTimes(xid, pTcon, &time_buf,
                                                         netfid);
                                CIFSSMBClose(xid, pTcon, netfid);
@@ -1334,7 +1651,7 @@ int cifs_setattr(struct dentry *direntry, struct iattr *attrs)
                           granularity */
 
                        /* rc = CIFSSMBSetTimesLegacy(xid, pTcon, full_path,
-                                       &time_buf, cifs_sb->local_nls); */
+                                       &time_buf, cifs_sb->local_nls); */
                        }
                }
                /* Even if error on time set, no sense failing the call if
@@ -1342,7 +1659,7 @@ int cifs_setattr(struct dentry *direntry, struct iattr *attrs)
                and this check ensures that we are not being called from
                sys_utimes in which case we ought to fail the call back to
                the user when the server rejects the call */
-               if((rc) && (attrs->ia_valid &&
+               if ((rc) && (attrs->ia_valid &
                         (ATTR_MODE | ATTR_GID | ATTR_UID | ATTR_SIZE)))
                        rc = 0;
        }
@@ -1357,9 +1674,11 @@ cifs_setattr_exit:
        return rc;
 }
 
+#if 0
 void cifs_delete_inode(struct inode *inode)
 {
        cFYI(1, ("In cifs_delete_inode, inode = 0x%p", inode));
        /* may have to add back in if and when safe distributed caching of
           directories added e.g. via FindNotify */
 }
+#endif