[CIFS] Fix rsize calculation so that large readx flag is checked.
authorSteve French <sfrench@us.ibm.com>
Mon, 10 Oct 2005 17:57:19 +0000 (10:57 -0700)
committerSteve French <sfrench@us.ibm.com>
Mon, 10 Oct 2005 17:57:19 +0000 (10:57 -0700)
Signed-off-by: Steve French (sfrench@us.ibm.com)
fs/cifs/cifs_debug.c
fs/cifs/connect.c
fs/cifs/file.c

index b7fb064..6f78109 100644 (file)
@@ -267,7 +267,7 @@ cifs_stats_read(char *buf, char **beginBuffer, off_t offset,
                        atomic_read(&tcon->num_oplock_brks));
                buf += item_length;
                length += item_length;
-               item_length = sprintf(buf, "\nReads:  %d Bytes %lld",
+               item_length = sprintf(buf, "\nReads:  %d Bytes: %lld",
                        atomic_read(&tcon->num_reads),
                        (long long)(tcon->bytes_read));
                buf += item_length;
index 31eb9a3..d1c6ace 100644 (file)
@@ -30,6 +30,7 @@
 #include <linux/mempool.h>
 #include <linux/delay.h>
 #include <linux/completion.h>
+#include <linux/pagevec.h>
 #include <asm/uaccess.h>
 #include <asm/processor.h>
 #include "cifspdu.h"
@@ -188,6 +189,7 @@ cifs_reconnect(struct TCP_Server_Info *server)
                                        server->server_RFC1001_name);
                }
                if(rc) {
+                       cERROR(1,("reconnect error %d",rc));
                        msleep(3000);
                } else {
                        atomic_inc(&tcpSesReconnectCount);
@@ -469,6 +471,7 @@ cifs_demultiplex_thread(struct TCP_Server_Info *server)
                        } else {
                                /* give server a second to
                                clean up before reconnect attempt */
+                               cERROR(1,("sleep before reconnect"));
                                msleep(1000);
                                /* always try 445 first on reconnect
                                since we get NACK on some if we ever
@@ -556,6 +559,7 @@ cifs_demultiplex_thread(struct TCP_Server_Info *server)
                dump_smb(smb_buffer, length);
                if (checkSMB (smb_buffer, smb_buffer->Mid, total_read+4)) {
                        cERROR(1, ("Bad SMB Received "));
+                       cifs_dump_mem("smb: ", smb_buffer, 48);
                        continue;
                }
 
@@ -1383,7 +1387,9 @@ ipv4_connect(struct sockaddr_in *psin_server, struct socket **csocket,
                the default. sock_setsockopt not used because it expects 
                user space buffer */
        (*csocket)->sk->sk_rcvtimeo = 7 * HZ;
-
+        cERROR(1,("sndbuf %d rcvbuf %d reset to 200K each",(*csocket)->sk->sk_sndbuf, (*csocket)->sk->sk_rcvbuf));
+       (*csocket)->sk->sk_sndbuf = 300 * 1024;
+       (*csocket)->sk->sk_rcvbuf = 200 * 1024;
        /* send RFC1001 sessinit */
 
        if(psin_server->sin_port == htons(RFC1001_PORT)) {
@@ -1736,11 +1742,20 @@ cifs_mount(struct super_block *sb, struct cifs_sb_info *cifs_sb,
     
        /* search for existing tcon to this server share */
        if (!rc) {
-               if((volume_info.rsize) && (volume_info.rsize <= CIFSMaxBufSize))
+               if(volume_info.rsize > CIFSMaxBufSize) {
+                       cERROR(1,("rsize %d too large, using MaxBufSize",
+                               volume_info.rsize));
+                       cifs_sb->rsize = CIFSMaxBufSize;
+               } else if((volume_info.rsize) && (volume_info.rsize <= CIFSMaxBufSize))
                        cifs_sb->rsize = volume_info.rsize;
-               else
-                       cifs_sb->rsize = srvTcp->maxBuf - MAX_CIFS_HDR_SIZE; /* default */
-               if(volume_info.wsize)
+               else /* default */
+                       cifs_sb->rsize = CIFSMaxBufSize;
+
+               if(volume_info.wsize > PAGEVEC_SIZE * PAGE_CACHE_SIZE) {
+                       cERROR(1,("wsize %d too large using 4096 instead",
+                                 volume_info.wsize));
+                       cifs_sb->wsize = 4096;
+               } else if(volume_info.wsize)
                        cifs_sb->wsize = volume_info.wsize;
                else
                        cifs_sb->wsize = CIFSMaxBufSize; /* default */
@@ -1895,6 +1910,10 @@ cifs_mount(struct super_block *sb, struct cifs_sb_info *cifs_sb,
                        cifs_sb->wsize = min(cifs_sb->wsize,
                                             (tcon->ses->server->maxBuf -
                                              MAX_CIFS_HDR_SIZE));
+               if (!(tcon->ses->capabilities & CAP_LARGE_READ_X))
+                        cifs_sb->rsize = min(cifs_sb->rsize,
+                                             (tcon->ses->server->maxBuf -
+                                              MAX_CIFS_HDR_SIZE));
        }
 
        /* volume_info.password is freed above when existing session found
index 39b23f4..11806c8 100644 (file)
@@ -855,7 +855,7 @@ static ssize_t cifs_write(struct file *file, const char *write_data,
                                struct kvec iov[2];
                                unsigned int len;
 
-                               len = min(cifs_sb->wsize,
+                               len = min((size_t)cifs_sb->wsize,
                                          write_size - total_written);
                                /* iov[0] is reserved for smb header */
                                iov[1].iov_base = (char *)write_data +
@@ -920,7 +920,8 @@ struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode)
                    ((open_file->pfile->f_flags & O_RDWR) ||
                     (open_file->pfile->f_flags & O_WRONLY))) {
                        read_unlock(&GlobalSMBSeslock);
-                       if(open_file->invalidHandle) {
+                       if((open_file->invalidHandle) && 
+                          (!open_file->closePend)) {
                                rc = cifs_reopen_file(&cifs_inode->vfs_inode, 
                                                      open_file->pfile, FALSE);
                                /* if it fails, try another handle - might be */