Skip to contentSkip to conversation

utilize lookup intents to open in lookup

PATCH In review Shirish Pargaonkar 2 participants 5 messages Mar 31, 2009 – Apr 9, 2009 VFS

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 utilize lookup intents to open in lookup

Hi,

Patch to expolint lookup intents in cifs lookup during open, thus reducing network traffick and speedier opens
Have also converted a semphore to mutex as per Jeff Layton's comments.

Trong, Al, your feedback is really appreciated.

Regards,

Shirish

fs/cifs/cifsglob.h +1 -1
index 9fbf4df..9a8368f 100644
@@ -350,7 +350,7 @@ struct cifsFileInfo {
 	bool invalidHandle:1;	/* file closed via session abend */
 	bool messageMode:1;	/* for pipes: message vs byte mode */
 	atomic_t wrtPending;   /* handle in use - defer close */
-	struct semaphore fh_sem; /* prevents reopen race after dead ses*/
+	struct mutex fh_mutex; /* prevents reopen race after dead ses*/
 	struct cifs_search_info srch_inf;
 };
 
fs/cifs/dir.c +88 -46

199 lines not on this page · raw message

fs/cifs/file.c +32 -33

104 lines not on this page · raw message

#1 lore raw
Quoted 6 lines · Shirish Pargaonkar

Hi,

Patch to expolint lookup intents in cifs lookup during open, thus reducing network traffick and speedier opens
Have also converted a semphore to mutex as per Jeff Layton's comments.

Quoted 6 lines

Trong, Al, your feedback is really appreciated.

Regards,

Shirish

Forgot to take out two debug lines. Did test this patch while creating regular files, directories, symbolic links,
exclusive create against older and newer version of samba and so far it works.

Regards,

Shirish

fs/cifs/cifsglob.h +1 -1
index 9fbf4df..9a8368f 100644
@@ -350,7 +350,7 @@ struct cifsFileInfo {
 	bool invalidHandle:1;	/* file closed via session abend */
 	bool messageMode:1;	/* for pipes: message vs byte mode */
 	atomic_t wrtPending;   /* handle in use - defer close */
-	struct semaphore fh_sem; /* prevents reopen race after dead ses*/
+	struct mutex fh_mutex; /* prevents reopen race after dead ses*/
 	struct cifs_search_info srch_inf;
 };
 
fs/cifs/dir.c +85 -46

196 lines not on this page · raw message

fs/cifs/file.c +32 -33

104 lines not on this page · raw message

#2 lore raw

On Wed, 1 Apr 2009 14:33:23 -0500
Shirish Pargaonkar <[email protected]> wrote:

Sorry I'm late. I know Steve has already committed this, but here are some comments. Maybe we can fix them in a later patch...

Quoted 32 lines
fs/cifs/cifsglob.h +1 -1
index 9fbf4df..9a8368f 100644
@@ -350,7 +350,7 @@ struct cifsFileInfo {
 	bool invalidHandle:1;	/* file closed via session abend */
 	bool messageMode:1;	/* for pipes: message vs byte mode */
 	atomic_t wrtPending;   /* handle in use - defer close */
-	struct semaphore fh_sem; /* prevents reopen race after dead ses*/
+	struct mutex fh_mutex; /* prevents reopen race after dead ses*/
 	struct cifs_search_info srch_inf;
 };
 
fs/cifs/dir.c +11 -0
index f9b6f68..f473c73 100644
@@ -129,12 +129,64 @@ cifs_bp_rename_retry:
 	return full_path;
 }
 
+static void
+cifs_fill_fileinfo(struct inode *newinode, __u16 fileHandle,
+			struct cifsTconInfo *tcon, bool write_only)
+{
+	int oplock = 0;
+	struct cifsFileInfo *pCifsFile;
+	struct cifsInodeInfo *pCifsInode;
+
+	pCifsFile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
+
+	if (pCifsFile == NULL)

fs/cifs/dir.c cifs_bp_rename_retry: · line 152

Quoted 10 lines
fs/cifs/dir.c +10 -0
+		return;
+
+	if (oplockEnabled)
+		oplock = REQ_OPLOCK;
+
+	pCifsFile->netfid = fileHandle;
+	pCifsFile->pid = current->tgid;
+	pCifsFile->pInode = newinode;
+	pCifsFile->invalidHandle = false;
+	pCifsFile->closePend     = false;
			^^^^^^^

None of the other '=' are lined up. Why here?

Quoted 9 lines
fs/cifs/dir.c +9 -0
+	mutex_init(&pCifsFile->fh_mutex);
+	mutex_init(&pCifsFile->lock_mutex);
+	INIT_LIST_HEAD(&pCifsFile->llist);
+	atomic_set(&pCifsFile->wrtPending, 0);
+
+	/* set the following in open now
+			pCifsFile->pfile = file; */
+	write_lock(&GlobalSMBSeslock);
+	list_add(&pCifsFile->tlist, &tcon->openFileList);

fs/cifs/dir.c cifs_bp_rename_retry: · line 171

Quoted 10 lines
fs/cifs/dir.c +10 -0
+	pCifsInode = CIFS_I(newinode);
+	if (pCifsInode) {
+		/* if readable file instance put first in list*/
+		if (write_only) {
+			list_add_tail(&pCifsFile->flist,
+				      &pCifsInode->openFileList);
+		} else {
+			list_add(&pCifsFile->flist,
+				 &pCifsInode->openFileList);
+		}
		^^^^^^^^

Are these brackets needed?

fs/cifs/dir.c cifs_bp_rename_retry: · line 179

Quoted 8 lines
fs/cifs/dir.c +8 -0
+		if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
+			pCifsInode->clientCanCacheAll = true;
+			pCifsInode->clientCanCacheRead = true;
+			cFYI(1, ("Exclusive Oplock inode %p",
+				newinode));
+		} else if ((oplock & 0xF) == OPLOCK_READ)
+			pCifsInode->clientCanCacheRead = true;
+	}

Ditto ^^^^^^^^^

Quoted 119 lines
fs/cifs/dir.c +17 -44
+	write_unlock(&GlobalSMBSeslock);
+}
+
 int cifs_posix_open(char *full_path, struct inode **pinode,
 		    struct super_block *sb, int mode, int oflags,
 		    int *poplock, __u16 *pnetfid, int xid)
 {
 	int rc;
 	__u32 oplock;
+	bool write_only = false;
 	FILE_UNIX_BASIC_INFO *presp_data;
 	__u32 posix_flags = 0;
 	struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
@@ -172,6 +224,8 @@ int cifs_posix_open(char *full_path, struct inode **pinode,
 	if (oflags & O_DIRECT)
 		posix_flags |= SMB_O_DIRECT;
 
+	if (!(oflags & FMODE_READ))
+		write_only = true;
 
 	rc = CIFSPOSIXCreate(xid, cifs_sb->tcon, posix_flags, mode,
 			pnetfid, presp_data, &oplock, full_path,
@@ -198,6 +252,8 @@ int cifs_posix_open(char *full_path, struct inode **pinode,
 
 	posix_fill_in_inode(*pinode, presp_data, 1);
 
+	cifs_fill_fileinfo(*pinode, *pnetfid, cifs_sb->tcon, write_only);
+
 posix_open_ret:
 	kfree(presp_data);
 	return rc;
@@ -239,7 +295,6 @@ cifs_create(struct inode *inode, struct dentry *direntry, int mode,
 	char *full_path = NULL;
 	FILE_ALL_INFO *buf = NULL;
 	struct inode *newinode = NULL;
-	struct cifsInodeInfo *pCifsInode;
 	int disposition = FILE_OVERWRITE_IF;
 	bool write_only = false;
 
@@ -410,44 +465,8 @@ cifs_create_set_dentry:
 		/* mknod case - do not leave file open */
 		CIFSSMBClose(xid, tcon, fileHandle);
 	} else if (newinode) {
-		struct cifsFileInfo *pCifsFile =
-			kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
-
-		if (pCifsFile == NULL)
-			goto cifs_create_out;
-		pCifsFile->netfid = fileHandle;
-		pCifsFile->pid = current->tgid;
-		pCifsFile->pInode = newinode;
-		pCifsFile->invalidHandle = false;
-		pCifsFile->closePend     = false;
-		init_MUTEX(&pCifsFile->fh_sem);
-		mutex_init(&pCifsFile->lock_mutex);
-		INIT_LIST_HEAD(&pCifsFile->llist);
-		atomic_set(&pCifsFile->wrtPending, 0);
-
-		/* set the following in open now
-				pCifsFile->pfile = file; */
-		write_lock(&GlobalSMBSeslock);
-		list_add(&pCifsFile->tlist, &tcon->openFileList);
-		pCifsInode = CIFS_I(newinode);
-		if (pCifsInode) {
-			/* if readable file instance put first in list*/
-			if (write_only) {
-				list_add_tail(&pCifsFile->flist,
-					      &pCifsInode->openFileList);
-			} else {
-				list_add(&pCifsFile->flist,
-					 &pCifsInode->openFileList);
-			}
-			if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
-				pCifsInode->clientCanCacheAll = true;
-				pCifsInode->clientCanCacheRead = true;
-				cFYI(1, ("Exclusive Oplock inode %p",
-					newinode));
-			} else if ((oplock & 0xF) == OPLOCK_READ)
-				pCifsInode->clientCanCacheRead = true;
-		}
-		write_unlock(&GlobalSMBSeslock);
+			cifs_fill_fileinfo(newinode, fileHandle,
+					cifs_sb->tcon, write_only);
 	}
 cifs_create_out:
 	kfree(buf);
@@ -580,17 +599,21 @@ int cifs_mknod(struct inode *inode, struct dentry *direntry, int mode,
 	return rc;
 }
 
-
 struct dentry *
 cifs_lookup(struct inode *parent_dir_inode, struct dentry *direntry,
 	    struct nameidata *nd)
 {
 	int xid;
 	int rc = 0; /* to get around spurious gcc warning, set to zero here */
+	int oplock = 0;
+	int mode;
+	__u16 fileHandle = 0;
+	bool posix_open = false;
 	struct cifs_sb_info *cifs_sb;
 	struct cifsTconInfo *pTcon;
 	struct inode *newInode = NULL;
 	char *full_path = NULL;
+	struct file *filp;
 
 	xid = GetXid();
 
@@ -632,12 +655,27 @@ cifs_lookup(struct inode *parent_dir_inode, struct dentry *direntry,
 	}
 	cFYI(1, ("Full path: %s inode = 0x%p", full_path, direntry->d_inode));
 
-	if (pTcon->unix_ext)
-		rc = cifs_get_inode_info_unix(&newInode, full_path,
-					      parent_dir_inode->i_sb, xid);
-	else
+	if (pTcon->unix_ext) {
+		if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY)) &&

fs/cifs/dir.c cifs_lookup(struct inode *parent_dir_inode, struct dentry *direntry, · line 669

Quoted 10 lines
fs/cifs/dir.c +10 -0
+				(nd->flags & LOOKUP_OPEN)) {
+			if (!((nd->intent.open.flags & O_CREAT) &&
+					(nd->intent.open.flags & O_EXCL))) {
+				mode = nd->intent.open.create_mode &
+						~current->fs->umask;
+				rc = cifs_posix_open(full_path, &newInode,
+					parent_dir_inode->i_sb, mode,
+					nd->intent.open.flags, &oplock,
+					&fileHandle, xid);
+				if ((rc != -EINVAL) && (rc != -EOPNOTSUPP))
					^^^^^^^^^^^^^^^^^^^^^^^^^^^^

This looks wrong to me. It's possible that you'd get a different return code here on an unsuccessful open call? Why would we want to instantiate the filp on any error from cifs_posix_open? For instance, that function can return -ENOMEM. I don't think we want to instantiate the filp in that case.

Maybe that should just be: "if (!rc)" ?

Quoted 82 lines
fs/cifs/dir.c +10 -2
+					posix_open = true;
+			}
+		}
+		if (!posix_open)
+			rc = cifs_get_inode_info_unix(&newInode, full_path,
+						parent_dir_inode->i_sb, xid);
+	} else
 		rc = cifs_get_inode_info(&newInode, full_path, NULL,
-					 parent_dir_inode->i_sb, xid, NULL);
+				parent_dir_inode->i_sb, xid, NULL);
 
 	if ((rc == 0) && (newInode != NULL)) {
 		if (pTcon->nocase)
@@ -645,7 +683,8 @@ cifs_lookup(struct inode *parent_dir_inode, struct dentry *direntry,
 		else
 			direntry->d_op = &cifs_dentry_ops;
 		d_add(direntry, newInode);
-
+		if (posix_open)
+			filp = lookup_instantiate_filp(nd, direntry, NULL);
 		/* since paths are not looked up by component - the parent
 		   directories are presumed to be good here */
 		renew_parental_timestamps(direntry);
fs/cifs/file.c +16 -28
index 81747ac..c3f51de 100644
@@ -46,7 +46,7 @@ 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 */
@@ -284,35 +284,34 @@ int cifs_open(struct inode *inode, struct file *file)
 	cifs_sb = CIFS_SB(inode->i_sb);
 	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;

fs/cifs/file.c int cifs_open(struct inode *inode, struct file *file) · line 311

Quoted 10 lines
fs/cifs/file.c +8 -0
 		}
 	}
+	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))

^^^^^ how about "else if" here and reduce the indentation below?

Quoted 44 lines
fs/cifs/file.c +8 -5
+			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) {
@@ -500,9 +499,9 @@ static int cifs_reopen_file(struct file *file, bool can_flush)
 		return -EBADF;
 
 	xid = GetXid();
-	down(&pCifsFile->fh_sem);
+	mutex_unlock(&pCifsFile->fh_mutex);
 	if (!pCifsFile->invalidHandle) {
-		up(&pCifsFile->fh_sem);
+		mutex_lock(&pCifsFile->fh_mutex);
 		FreeXid(xid);
 		return 0;
 	}
@@ -533,7 +532,7 @@ static int cifs_reopen_file(struct file *file, bool can_flush)
 	if (full_path == NULL) {
 		rc = -ENOMEM;
 reopen_error_exit:
-		up(&pCifsFile->fh_sem);
+		mutex_lock(&pCifsFile->fh_mutex);
 		FreeXid(xid);
 		return rc;
 	}
@@ -575,14 +574,14 @@ reopen_error_exit:
 			 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);
+		mutex_lock(&pCifsFile->fh_mutex);
 		pCifsInode = CIFS_I(inode);
 		if (pCifsInode) {
 			if (can_flush) {
Signature · 1 line
Jeff Layton <[email protected]>
#3 lore raw
Quoted 210 lines · Jeff Layton

On Wed, 1 Apr 2009 14:33:23 -0500
Shirish Pargaonkar <[email protected]> wrote:

Sorry I'm late. I know Steve has already committed this, but here are some comments. Maybe we can fix them in a later patch...

Quoted 32 lines
fs/cifs/cifsglob.h +0 -0
index 9fbf4df..9a8368f 100644
@@ -350,7 +350,7 @@ struct cifsFileInfo {

      bool invalidHandle:1;   /* file closed via session abend */ bool messageMode:1;     /* for pipes: message vs byte mode */ atomic_t wrtPending;   /* handle in use - defer close */
-     struct semaphore fh_sem; /* prevents reopen race after dead ses*/
+     struct mutex fh_mutex; /* prevents reopen race after dead ses*/ struct cifs_search_info srch_inf;
 };

fs/cifs/dir.c +11 -0
index f9b6f68..f473c73 100644
@@ -129,12 +129,64 @@ cifs_bp_rename_retry:
       return full_path;
  }
 
+static void
+cifs_fill_fileinfo(struct inode *newinode, __u16 fileHandle,
+                     struct cifsTconInfo *tcon, bool write_only)
+{
+     int oplock = 0;
+     struct cifsFileInfo *pCifsFile;
+     struct cifsInodeInfo *pCifsInode;
+
+     pCifsFile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
+
+     if (pCifsFile == NULL)

fs/cifs/dir.c cifs_bp_rename_retry: · line 152

Quoted 10 lines
fs/cifs/dir.c +10 -0
+             return;
+
+     if (oplockEnabled)
+             oplock = REQ_OPLOCK;
+
+     pCifsFile->netfid = fileHandle;
+     pCifsFile->pid = current->tgid;
+     pCifsFile->pInode = newinode;
+     pCifsFile->invalidHandle = false;
+     pCifsFile->closePend     = false;

                       ^^^^^^^
None of the other '=' are lined up. Why here?

Quoted 9 lines
fs/cifs/dir.c +9 -0
+     mutex_init(&pCifsFile->fh_mutex);
+     mutex_init(&pCifsFile->lock_mutex);
+     INIT_LIST_HEAD(&pCifsFile->llist);
+     atomic_set(&pCifsFile->wrtPending, 0);
+
+     /* set the following in open now
+                     pCifsFile->pfile = file; */
+     write_lock(&GlobalSMBSeslock);
+     list_add(&pCifsFile->tlist, &tcon->openFileList);

fs/cifs/dir.c cifs_bp_rename_retry: · line 171

Quoted 10 lines
fs/cifs/dir.c +10 -0
+     pCifsInode = CIFS_I(newinode);
+     if (pCifsInode) {
+             /* if readable file instance put first in list*/
+             if (write_only) {
+                     list_add_tail(&pCifsFile->flist,
+                                   &pCifsInode->openFileList);
+             } else {
+                     list_add(&pCifsFile->flist,
+                              &pCifsInode->openFileList);
+             }

               ^^^^^^^^
Are these brackets needed?

fs/cifs/dir.c cifs_bp_rename_retry: · line 179

Quoted 8 lines
fs/cifs/dir.c +8 -0
+             if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
+                     pCifsInode->clientCanCacheAll = true;
+                     pCifsInode->clientCanCacheRead = true;
+                     cFYI(1, ("Exclusive Oplock inode %p",
+                             newinode));
+             } else if ((oplock & 0xF) == OPLOCK_READ)
+                     pCifsInode->clientCanCacheRead = true;
+     }

Ditto           ^^^^^^^^^

Quoted 127 lines
fs/cifs/dir.c +8 -0
+     write_unlock(&GlobalSMBSeslock);
+}
+
  int cifs_posix_open(char *full_path, struct inode **pinode,
                   struct super_block *sb, int mode, int oflags,
                   int *poplock, __u16 *pnetfid, int xid)
  {
       int rc;
       __u32 oplock;
+     bool write_only = false;
       FILE_UNIX_BASIC_INFO *presp_data;
       __u32 posix_flags = 0;
       struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
@@ -172,6 +224,8 @@ int cifs_posix_open(char *full_path, struct inode **pinode,
       if (oflags & O_DIRECT)
               posix_flags |= SMB_O_DIRECT;
 
+     if (!(oflags & FMODE_READ))
+             write_only = true;
 
       rc = CIFSPOSIXCreate(xid, cifs_sb->tcon, posix_flags, mode,
                       pnetfid, presp_data, &oplock, full_path,
@@ -198,6 +252,8 @@ int cifs_posix_open(char *full_path, struct inode **pinode,
 
       posix_fill_in_inode(*pinode, presp_data, 1);
 
+     cifs_fill_fileinfo(*pinode, *pnetfid, cifs_sb->tcon, write_only);
+

 posix_open_ret:
      kfree(presp_data);
      return rc;

fs/cifs/dir.c +2 -39
@@ -239,7 +295,6 @@ cifs_create(struct inode *inode, struct dentry *direntry, int mode,
       char *full_path = NULL;
       FILE_ALL_INFO *buf = NULL;
       struct inode *newinode = NULL;
-     struct cifsInodeInfo *pCifsInode;
       int disposition = FILE_OVERWRITE_IF;
       bool write_only = false;
 
@@ -410,44 +465,8 @@ cifs_create_set_dentry:
               /* mknod case - do not leave file open */
               CIFSSMBClose(xid, tcon, fileHandle);
       } else if (newinode) {
-             struct cifsFileInfo *pCifsFile =
-                     kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
-
-             if (pCifsFile == NULL)
-                     goto cifs_create_out;
-             pCifsFile->netfid = fileHandle;
-             pCifsFile->pid = current->tgid;
-             pCifsFile->pInode = newinode;
-             pCifsFile->invalidHandle = false;
-             pCifsFile->closePend     = false;
-             init_MUTEX(&pCifsFile->fh_sem);
-             mutex_init(&pCifsFile->lock_mutex);
-             INIT_LIST_HEAD(&pCifsFile->llist);
-             atomic_set(&pCifsFile->wrtPending, 0);
-
-             /* set the following in open now
-                             pCifsFile->pfile = file; */
-             write_lock(&GlobalSMBSeslock);
-             list_add(&pCifsFile->tlist, &tcon->openFileList);
-             pCifsInode = CIFS_I(newinode);
-             if (pCifsInode) {
-                     /* if readable file instance put first in list*/
-                     if (write_only) {
-                             list_add_tail(&pCifsFile->flist,
-                                           &pCifsInode->openFileList);
-                     } else {
-                             list_add(&pCifsFile->flist,
-                                      &pCifsInode->openFileList);
-                     }
-                     if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
-                             pCifsInode->clientCanCacheAll = true;
-                             pCifsInode->clientCanCacheRead = true;
-                             cFYI(1, ("Exclusive Oplock inode %p",
-                                     newinode));
-                     } else if ((oplock & 0xF) == OPLOCK_READ)
-                             pCifsInode->clientCanCacheRead = true;
-             }
-             write_unlock(&GlobalSMBSeslock);
+                     cifs_fill_fileinfo(newinode, fileHandle,
+                                     cifs_sb->tcon, write_only);
       }

 cifs_create_out:
      kfree(buf);

fs/cifs/dir.c +0 -1
@@ -580,17 +599,21 @@ int cifs_mknod(struct inode *inode, struct dentry *direntry, int mode,
       return rc;
  }
 
-

 struct dentry *
 cifs_lookup(struct inode *parent_dir_inode, struct dentry *direntry, struct nameidata *nd)
 {
      int xid;
      int rc = 0; /* to get around spurious gcc warning, set to zero here */
+     int oplock = 0;
+     int mode;
+     __u16 fileHandle = 0;
+     bool posix_open = false;
      struct cifs_sb_info *cifs_sb;
      struct cifsTconInfo *pTcon;
      struct inode *newInode = NULL;
      char *full_path = NULL;
+     struct file *filp;

      xid = GetXid();

fs/cifs/dir.c +10 -4
@@ -632,12 +655,27 @@ cifs_lookup(struct inode *parent_dir_inode, struct dentry *direntry,
       }
       cFYI(1, ("Full path: %s inode = 0x%p", full_path, direntry->d_inode));
 
-     if (pTcon->unix_ext)
-             rc = cifs_get_inode_info_unix(&newInode, full_path,
-                                           parent_dir_inode->i_sb, xid);
-     else
+     if (pTcon->unix_ext) {
+             if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY)) &&
+                             (nd->flags & LOOKUP_OPEN)) {
+                     if (!((nd->intent.open.flags & O_CREAT) &&
+                                     (nd->intent.open.flags & O_EXCL))) {
+                             mode = nd->intent.open.create_mode &
+                                             ~current->fs->umask;
+                             rc = cifs_posix_open(full_path, &newInode,
+                                     parent_dir_inode->i_sb, mode,
+                                     nd->intent.open.flags, &oplock,
Quoted 8 lines

fs/cifs/dir.c cifs_lookup(struct inode *parent_dir_inode, struct dentry *direntry, · line 669

Quoted 2 lines
fs/cifs/dir.c +2 -0
+                                     &fileHandle, xid);
+                             if ((rc != -EINVAL) && (rc != -EOPNOTSUPP))

                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ This looks wrong to me.  It's possible that you'd get a different return code here on an unsuccessful open call? Why would we want to instantiate the filp on any error from cifs_posix_open? For instance, that function can return -ENOMEM. I don't think we want to instantiate the filp in that case.

Jeff,

Thanks, will handle the identation/formatting errors in a subsequent patch for sure.

If there is any other error code returned besides these two error codes, it is handled as an error, so file pointer will not be instantiated (as posix_open
will be true).
These two errors are there to handle the bug in samba posix open and if either of these error codes are returned, cifs_lookup does what it had been doing (as posix_open will be false),.

Quoted 138 lines

Maybe that should just be: "if (!rc)" ?

Quoted 82 lines
fs/cifs/dir.c +8 -1
+                                     posix_open = true;
+                     }
+             }
+             if (!posix_open)
+                     rc = cifs_get_inode_info_unix(&newInode, full_path,
+                                             parent_dir_inode->i_sb, xid);
+     } else
               rc = cifs_get_inode_info(&newInode, full_path, NULL,
-                                      parent_dir_inode->i_sb, xid, NULL);
+                             parent_dir_inode->i_sb, xid, NULL);
 
       if ((rc == 0) && (newInode != NULL)) {
               if (pTcon->nocase)
@@ -645,7 +683,8 @@ cifs_lookup(struct inode *parent_dir_inode, struct dentry *direntry,

              else
                      direntry->d_op = &cifs_dentry_ops; d_add(direntry, newInode);
-
+             if (posix_open)
+                     filp = lookup_instantiate_filp(nd, direntry, NULL); /* since paths are not looked up by component - the parent directories are presumed to be good here */ renew_parental_timestamps(direntry);

fs/cifs/file.c +1 -1
index 81747ac..c3f51de 100644
@@ -46,7 +46,7 @@ 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 */

fs/cifs/file.c +15 -27
@@ -284,35 +284,34 @@ int cifs_open(struct inode *inode, struct file *file)
       cifs_sb = CIFS_SB(inode->i_sb);
       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;

fs/cifs/file.c int cifs_open(struct inode *inode, struct file *file) · line 311

Quoted 10 lines
fs/cifs/file.c +8 -0
               }
       }
+     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))

       ^^^^^ how about "else if" here and reduce the indentation below?

Quoted 40 lines
fs/cifs/file.c +5 -2
+                     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) {
@@ -500,9 +499,9 @@ static int cifs_reopen_file(struct file *file, bool can_flush)
               return -EBADF;
 
       xid = GetXid();
-     down(&pCifsFile->fh_sem);
+     mutex_unlock(&pCifsFile->fh_mutex);
       if (!pCifsFile->invalidHandle) {
-             up(&pCifsFile->fh_sem);
+             mutex_lock(&pCifsFile->fh_mutex);
               FreeXid(xid);
               return 0;
       }
@@ -533,7 +532,7 @@ static int cifs_reopen_file(struct file *file, bool can_flush)
       if (full_path == NULL) {
               rc = -ENOMEM;

 reopen_error_exit:
-             up(&pCifsFile->fh_sem);
+             mutex_lock(&pCifsFile->fh_mutex);
              FreeXid(xid);
              return rc;
      }

fs/cifs/file.c +0 -0
@@ -575,14 +574,14 @@ reopen_error_exit:

                       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);

Quoted 9 lines
Quoted 4 lines

+             mutex_lock(&pCifsFile->fh_mutex);
              pCifsInode = CIFS_I(inode);
              if (pCifsInode) {
                      if (can_flush) {

--
Jeff Layton <[email protected]>

--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#4 lore raw
Quoted 32 lines · On Thu, 9 Apr 2009 08:46:07 -0500 Shirish Pargaonkar…
Quoted 18 lines
Quoted 18 lines
fs/cifs/dir.c +10 -4
@@ -632,12 +655,27 @@ cifs_lookup(struct inode *parent_dir_inode, struct dentry *direntry,
       }
       cFYI(1, ("Full path: %s inode = 0x%p", full_path, direntry->d_inode));
 
-     if (pTcon->unix_ext)
-             rc = cifs_get_inode_info_unix(&newInode, full_path,
-                                           parent_dir_inode->i_sb, xid);
-     else
+     if (pTcon->unix_ext) {
+             if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY)) &&
+                             (nd->flags & LOOKUP_OPEN)) {
+                     if (!((nd->intent.open.flags & O_CREAT) &&
+                                     (nd->intent.open.flags & O_EXCL))) {
+                             mode = nd->intent.open.create_mode &
+                                             ~current->fs->umask;
+                             rc = cifs_posix_open(full_path, &newInode,
+                                     parent_dir_inode->i_sb, mode,
+                                     nd->intent.open.flags, &oplock,
Quoted 8 lines

fs/cifs/dir.c cifs_lookup(struct inode *parent_dir_inode, struct dentry *direntry, · line 669

Quoted 2 lines
fs/cifs/dir.c +2 -0
+                                     &fileHandle, xid);
+                             if ((rc != -EINVAL) && (rc != -EOPNOTSUPP))

                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ This looks wrong to me.  It's possible that you'd get a different return code here on an unsuccessful open call? Why would we want to instantiate the filp on any error from cifs_posix_open? For instance, that function can return -ENOMEM. I don't think we want to instantiate the filp in that case.

Jeff,

Thanks, will handle the identation/formatting errors in a subsequent patch for sure.

Quoted 8 lines

If there is any other error code returned besides these two error codes, it is handled as an error, so file pointer will not be instantiated (as posix_open
will be true).
These two errors are there to handle the bug in samba posix open and if either of these error codes are returned, cifs_lookup does what it had been doing (as posix_open will be false),.

Ok, thanks for clarifying that. Maybe a comment in there to explain that would be helpful too?

Signature · 1 line
Jeff Layton <[email protected]>