[CIFS] New CIFS POSIX mkdir performance improvement

Signed-off-by: Steve French <sfrench@us.ibm.com>

5 files changed, 340 insertions, 15 deletions

diff --git a/fs/cifs/CHANGES b/fs/cifs/CHANGES
index f75bdc661769..713fec669135 100644
--- a/fs/cifs/CHANGES
+++ b/fs/cifs/CHANGES
@@ -12,7 +12,10 @@ when archive dos attribute not set and we are changing mode back to writeable
 on server which does not support the Unix Extensions).  Remove read only dos
 attribute on chmod when adding any write permission (ie on any of
 user/group/other (not all of user/group/other ie  0222) when
-mounted to windows.
+mounted to windows.  Add support for POSIX MkDir (slight performance
+enhancement and eliminates the network race between the mkdir and set 
+path info of the mode).
+
 
 Version 1.47
 ------------
diff --git a/fs/cifs/cifspdu.h b/fs/cifs/cifspdu.h
index 4d8948e8762c..d619ca7d1416 100644
--- a/fs/cifs/cifspdu.h
+++ b/fs/cifs/cifspdu.h
@@ -1388,7 +1388,7 @@ struct smb_t2_rsp {
 #define SMB_SET_POSIX_LOCK              0x208
 #define SMB_POSIX_OPEN                  0x209
 #define SMB_POSIX_UNLINK                0x20a
-#define SMB_SET_FILE_UNIX_INFO2
+#define SMB_SET_FILE_UNIX_INFO2         0x20b
 #define SMB_SET_FILE_BASIC_INFO2        0x3ec
 #define SMB_SET_FILE_RENAME_INFORMATION 0x3f2 /* BB check if qpathinfo too */
 #define SMB_FILE_ALL_INFO2              0x3fa
@@ -2109,22 +2109,40 @@ struct cifs_posix_acl { /* access conrol list  (ACL) */
 
 /* end of POSIX ACL definitions */
 
+/* POSIX Open Flags */
+#define SMB_O_RDONLY     0x1
+#define SMB_O_WRONLY    0x2
+#define SMB_O_RDWR      0x4
+#define SMB_O_CREAT     0x10
+#define SMB_O_EXCL      0x20
+#define SMB_O_TRUNC     0x40
+#define SMB_O_APPEND    0x80
+#define SMB_O_SYNC      0x100
+#define SMB_O_DIRECTORY 0x200
+#define SMB_O_NOFOLLOW  0x400
+#define SMB_O_DIRECT    0x800
+
 typedef struct {
-        __u32 OpenFlags; /* same as NT CreateX */
-        __u32 PosixOpenFlags;
-        __u32 Mode;
-        __u16 Level; /* reply level requested (see QPathInfo levels) */
-        __u16 Pad;  /* reserved - MBZ */
+        __le32 OpenFlags; /* same as NT CreateX */
+        __le32 PosixOpenFlags;
+        __le64 Permissions;
+        __le16 Level; /* reply level requested (see QPathInfo levels) */
 } __attribute__((packed)) OPEN_PSX_REQ; /* level 0x209 SetPathInfo data */
 
 typedef struct {
-        /* reply varies based on requested level */
+        __le16 OplockFlags;
+        __u16 Fid;
+        __le32 CreateAction;
+        __le16 ReturnedLevel;
+        __le16 Pad;
+        /* struct following varies based on requested level */
 } __attribute__((packed)) OPEN_PSX_RSP; /* level 0x209 SetPathInfo data */
 
 
 struct file_internal_info {
         __u64  UniqueId; /* inode number */
 } __attribute__((packed));      /* level 0x3ee */
+
 struct file_mode_info {
         __le32  Mode;
 } __attribute__((packed));      /* level 0x3f8 */
diff --git a/fs/cifs/cifsproto.h b/fs/cifs/cifsproto.h
index 32eb1acab630..5d163e2b6143 100644
--- a/fs/cifs/cifsproto.h
+++ b/fs/cifs/cifsproto.h
@@ -1,7 +1,7 @@
 /*
  *   fs/cifs/cifsproto.h
  *
- *   Copyright (c) International Business Machines  Corp., 2002,2006
+ *   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
@@ -244,6 +244,11 @@ extern int SMBLegacyOpen(const int xid, struct cifsTconInfo *tcon,
                         const int access_flags, const int omode,
                         __u16 * netfid, int *pOplock, FILE_ALL_INFO *,
                         const struct nls_table *nls_codepage, int remap);
+extern int CIFSPOSIXCreate(const int xid, struct cifsTconInfo *tcon, 
+                        u32 posix_flags, __u64 mode, __u16 * netfid,
+                        FILE_UNIX_BASIC_INFO *pRetData,
+                        __u32 *pOplock, const char *name,
+                        const struct nls_table *nls_codepage, int remap);                       
 extern int CIFSSMBClose(const int xid, struct cifsTconInfo *tcon,
                         const int smb_file_id);
 
diff --git a/fs/cifs/cifssmb.c b/fs/cifs/cifssmb.c
index 48fc0c2ab0e5..a7d3d8e5c6c5 100644
--- a/fs/cifs/cifssmb.c
+++ b/fs/cifs/cifssmb.c
@@ -1,7 +1,7 @@
 /*
  *   fs/cifs/cifssmb.c
  *
- *   Copyright (C) International Business Machines  Corp., 2002,2006
+ *   Copyright (C) International Business Machines  Corp., 2002,2007
  *   Author(s): Steve French (sfrench@us.ibm.com)
  *
  *   Contains the routines for constructing the SMB PDUs themselves
@@ -24,8 +24,8 @@
  /* SMB/CIFS PDU handling routines here - except for leftovers in connect.c   */
  /* These are mostly routines that operate on a pathname, or on a tree id     */
  /* (mounted volume), but there are eight handle based routines which must be */
- /* treated slightly different for reconnection purposes since we never want  */
- /* to reuse a stale file handle and the caller knows the file handle */
+ /* treated slightly differently for reconnection purposes since we never     */
+ /* want to reuse a stale file handle and only the caller knows the file info */
 
 #include <linux/fs.h>
 #include <linux/kernel.h>
@@ -913,6 +913,127 @@ MkDirRetry:
         return rc;
 }
 
+int
+CIFSPOSIXCreate(const int xid, struct cifsTconInfo *tcon, __u32 posix_flags,
+                __u64 mode, __u16 * netfid, FILE_UNIX_BASIC_INFO *pRetData,
+                __u32 *pOplock, const char *name, 
+                const struct nls_table *nls_codepage, int remap)
+{
+        TRANSACTION2_SPI_REQ *pSMB = NULL;
+        TRANSACTION2_SPI_RSP *pSMBr = NULL;
+        int name_len;
+        int rc = 0;
+        int bytes_returned = 0;
+        char *data_offset;
+        __u16 params, param_offset, offset, byte_count, count;
+        OPEN_PSX_REQ * pdata;
+        OPEN_PSX_RSP * psx_rsp;
+
+        cFYI(1, ("In POSIX Create"));
+PsxCreat:
+        rc = smb_init(SMB_COM_TRANSACTION2, 15, tcon, (void **) &pSMB,
+                      (void **) &pSMBr);
+        if (rc)
+                return rc;
+
+        if (pSMB->hdr.Flags2 & SMBFLG2_UNICODE) {
+                name_len =
+                    cifsConvertToUCS((__le16 *) pSMB->FileName, name,
+                                     PATH_MAX, nls_codepage, remap);
+                name_len++;     /* trailing null */
+                name_len *= 2;
+        } else {        /* BB improve the check for buffer overruns BB */
+                name_len = strnlen(name, PATH_MAX);
+                name_len++;     /* trailing null */
+                strncpy(pSMB->FileName, name, name_len);
+        }
+
+        params = 6 + name_len;
+        count = sizeof(OPEN_PSX_REQ);
+        pSMB->MaxParameterCount = cpu_to_le16(2);
+        pSMB->MaxDataCount = cpu_to_le16(1000); /* large enough */
+        pSMB->MaxSetupCount = 0;
+        pSMB->Reserved = 0;
+        pSMB->Flags = 0;
+        pSMB->Timeout = 0;
+        pSMB->Reserved2 = 0;
+        param_offset = offsetof(struct smb_com_transaction2_spi_req,
+                                     InformationLevel) - 4;
+        offset = param_offset + params;
+        data_offset = (char *) (&pSMB->hdr.Protocol) + offset;
+        pdata = (OPEN_PSX_REQ *)(((char *)&pSMB->hdr.Protocol) + offset);
+        pdata->Level = SMB_QUERY_FILE_UNIX_BASIC;
+        pdata->Permissions = cpu_to_le64(mode);
+        pdata->PosixOpenFlags = cpu_to_le32(posix_flags); 
+        pdata->OpenFlags =  cpu_to_le32(*pOplock);
+        pSMB->ParameterOffset = cpu_to_le16(param_offset);
+        pSMB->DataOffset = cpu_to_le16(offset);
+        pSMB->SetupCount = 1;
+        pSMB->Reserved3 = 0;
+        pSMB->SubCommand = cpu_to_le16(TRANS2_SET_PATH_INFORMATION);
+        byte_count = 3 /* pad */  + params + count;
+
+        pSMB->DataCount = cpu_to_le16(count);
+        pSMB->ParameterCount = cpu_to_le16(params);
+        pSMB->TotalDataCount = pSMB->DataCount;
+        pSMB->TotalParameterCount = pSMB->ParameterCount;
+        pSMB->InformationLevel = cpu_to_le16(SMB_POSIX_OPEN);
+        pSMB->Reserved4 = 0;
+        pSMB->hdr.smb_buf_length += byte_count; 
+        pSMB->ByteCount = cpu_to_le16(byte_count);
+        rc = SendReceive(xid, tcon->ses, (struct smb_hdr *) pSMB,
+                         (struct smb_hdr *) pSMBr, &bytes_returned, 0);
+        if (rc) {
+                cFYI(1, ("Posix create returned %d", rc));
+                goto psx_create_err;
+        }
+
+        cFYI(1,("copying inode info"));
+        rc = validate_t2((struct smb_t2_rsp *)pSMBr);
+
+        if (rc || (pSMBr->ByteCount < sizeof(OPEN_PSX_RSP))) {
+                rc = -EIO;      /* bad smb */
+                goto psx_create_err;
+        }
+
+        /* copy return information to pRetData */
+        psx_rsp = (OPEN_PSX_RSP *)((char *) &pSMBr->hdr.Protocol 
+                        + le16_to_cpu(pSMBr->t2.DataOffset));
+                
+        *pOplock = le16_to_cpu(psx_rsp->OplockFlags);
+        if(netfid)
+                *netfid = psx_rsp->Fid;   /* cifs fid stays in le */
+        /* Let caller know file was created so we can set the mode. */
+        /* Do we care about the CreateAction in any other cases? */
+        if(cpu_to_le32(FILE_CREATE) == psx_rsp->CreateAction)
+                *pOplock |= CIFS_CREATE_ACTION;
+        /* check to make sure response data is there */
+        if(psx_rsp->ReturnedLevel != SMB_QUERY_FILE_UNIX_BASIC)
+                pRetData->Type = -1; /* unknown */              
+        else {
+                if(pSMBr->ByteCount < sizeof(OPEN_PSX_RSP) 
+                                        + sizeof(FILE_UNIX_BASIC_INFO)) {
+                        cERROR(1,("Open response data too small"));
+                        pRetData->Type = -1;
+                        goto psx_create_err;
+                }
+                memcpy((char *) pRetData, 
+                        (char *)&psx_rsp + sizeof(OPEN_PSX_RSP),
+                        sizeof (FILE_UNIX_BASIC_INFO));
+        }
+                        
+
+psx_create_err:
+        cifs_buf_release(pSMB);
+
+        cifs_stats_inc(&tcon->num_mkdirs);
+
+        if (rc == -EAGAIN)
+                goto PsxCreat;
+
+        return rc;      
+}
+
 static __u16 convert_disposition(int disposition)
 {
         __u16 ofun = 0;
diff --git a/fs/cifs/inode.c b/fs/cifs/inode.c
index eeea33752e68..9b7e0dbdd826 100644
--- a/fs/cifs/inode.c
+++ b/fs/cifs/inode.c
@@ -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
@@ -734,6 +734,133 @@ 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)); 
+        }
+
+        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
+                                           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;
@@ -755,6 +882,53 @@ 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 {
+                        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;
+                        d_instantiate(direntry, newinode);
+                        if (direntry->d_inode) {
+                                int obj_type;
+                                direntry->d_inode->i_nlink = 2;
+                                /* already checked in POSIXCreate whether
+                                frame was long enough */
+                                posix_fill_in_inode(direntry->d_inode,
+                                        pInfo, &obj_type, 1 /* NewInode */);
+                                /* could double check that we actually
+                                 * created what we thought we did ie
+                                 * a directory
+                                 */     
+                        }
+                }
+                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);
@@ -762,6 +936,7 @@ int cifs_mkdir(struct inode *inode, struct dentry *direntry, int mode)
                 cFYI(1, ("cifs_mkdir returned 0x%x", rc));
                 d_drop(direntry);
         } else {
+mkdir_get_info:         
                 inc_nlink(inode);
                 if (pTcon->ses->capabilities & CAP_UNIX)
                         rc = cifs_get_inode_info_unix(&newinode, full_path,
@@ -775,8 +950,10 @@ int cifs_mkdir(struct inode *inode, struct dentry *direntry, int mode)
                 else
                         direntry->d_op = &cifs_dentry_ops;
                 d_instantiate(direntry, newinode);
-                if (direntry->d_inode)
-                        direntry->d_inode->i_nlink = 2;
+                 /* 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 (cifs_sb->tcon->ses->capabilities & CAP_UNIX)
                         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_SET_UID) {
                                 CIFSSMBUnixSetPerms(xid, pTcon, full_path,
@@ -812,6 +989,7 @@ int cifs_mkdir(struct inode *inode, struct dentry *direntry, int mode)
                         }
                 }
         }
+mkdir_out:      
         kfree(full_path);
         FreeXid(xid);
         return rc;