Whamcloud - gitweb
- move the peter branch changes to the head
[fs/lustre-release.git] / lustre / llite / file.c
index 16d29e3..41ff0b8 100644 (file)
 
 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc);
 extern int ll_setattr(struct dentry *de, struct iattr *attr);
-extern inline struct obdo * ll_oa_from_inode(struct inode *inode,
-                                             unsigned long valid);
 
 static int ll_file_open(struct inode *inode, struct file *file)
 {
         int rc;
         struct ptlrpc_request *req = NULL;
         struct ll_file_data *fd;
-        struct obdo *oa;
+        struct obdo *oa = NULL;
+        struct lov_stripe_md *md = NULL; 
         struct ll_sb_info *sbi = ll_i2sbi(inode);
+        struct ll_inode_info *lli = ll_i2info(inode);
         ENTRY;
 
         if (file->private_data)
                 LBUG();
 
+        /*  delayed create of object (intent created inode) */
+        /*  XXX object needs to be cleaned up if mdc_open fails */
+        /*  XXX error handling appropriate here? */
+        if (lli->lli_smd == NULL || lli->lli_smd->lmd_object_id == 0) {
+                struct client_obd *mdc = sbi2mdc(ll_s2sbi(inode->i_sb));
+                struct inode * inode = file->f_dentry->d_inode;
+
+                lli->lli_smd = NULL; 
+                oa =  obdo_alloc();
+                if (!oa) { 
+                        RETURN(-ENOMEM);
+                }
+                oa->o_valid = OBD_MD_FLMODE;
+                oa->o_mode = S_IFREG | 0600;
+                oa->o_easize = mdc->cl_max_mdsize;
+                rc = obd_create(ll_i2obdconn(inode), oa, &lli->lli_smd);
+                if (rc)
+                        RETURN(rc);
+                md = lli->lli_smd;
+                lli->lli_flags &= ~OBD_FL_CREATEONOPEN;
+        }
+
         fd = kmem_cache_alloc(ll_file_data_slab, SLAB_KERNEL);
         if (!fd)
                 GOTO(out, rc = -ENOMEM);
         memset(fd, 0, sizeof(*fd));
 
-        rc = mdc_open(&sbi->ll_mds_client, sbi->ll_mds_conn, inode->i_ino,
-                      S_IFREG, file->f_flags, (__u64)(unsigned long)file, 
-                      &fd->fd_mdshandle, &req); 
+        rc = mdc_open(&sbi->ll_mdc_conn, inode->i_ino, S_IFREG | inode->i_mode,
+                      file->f_flags, md,
+                      (__u64)(unsigned long)file, &fd->fd_mdshandle, &req);
         fd->fd_req = req;
         ptlrpc_req_finished(req);
-        if (rc) {
-                if (rc > 0) 
-                        rc = -rc;
-                GOTO(out, rc);
+        if (rc)
+                GOTO(out_req, -abs(rc));
+        if (!fd->fd_mdshandle) {
+                CERROR("mdc_open didn't assign fd_mdshandle\n");
+                /* XXX handle this how, abort or is it non-fatal? */
         }
         if (!fd->fd_mdshandle)
                 CERROR("mdc_open didn't assign fd_mdshandle\n");
-        
 
-        oa = ll_oa_from_inode(inode, (OBD_MD_FLMODE | OBD_MD_FLID));
-        if (oa == NULL)
-                LBUG();
-        rc = obd_open(ll_i2obdconn(inode), oa);
-        obdo_free(oa);
-        if (rc) {
-                /* XXX: Need to do mdc_close here! */
-                GOTO(out, rc = abs(rc));
+        if (oa == NULL) {
+                oa =  obdo_alloc();
+                if (!oa)
+                        GOTO(out_mdc, rc = -EINVAL);
         }
+        oa->o_id = lli->lli_smd->lmd_object_id;
+        oa->o_mode = S_IFREG | inode->i_mode;
+        rc = obd_open(ll_i2obdconn(inode), oa, lli->lli_smd);
+        obdo_free(oa);
+        oa = NULL;
+
+        if (rc)
+                GOTO(out_mdc, rc = -abs(rc));
 
         file->private_data = fd;
 
         EXIT;
- out:
-        if (rc && fd) {
-                kmem_cache_free(ll_file_data_slab, fd);
-                file->private_data = NULL;
-        }
 
+        return 0;
+out_mdc:
+        mdc_close(&sbi->ll_mdc_conn, inode->i_ino,
+                  S_IFREG, fd->fd_mdshandle, &req);
+out_req:
+        ptlrpc_free_req(req);
+//out_fd:
+        if (oa)
+                obdo_free(oa);
+        kmem_cache_free(ll_file_data_slab, fd);
+        file->private_data = NULL;
+out:
         return rc;
 }
 
@@ -93,7 +126,7 @@ static int ll_file_release(struct inode *inode, struct file *file)
         struct ll_file_data *fd;
         struct obdo *oa;
         struct ll_sb_info *sbi = ll_i2sbi(inode);
-        struct iattr iattr;
+        struct ll_inode_info *lli = ll_i2info(inode);
 
         ENTRY;
 
@@ -103,46 +136,48 @@ static int ll_file_release(struct inode *inode, struct file *file)
                 GOTO(out, rc = -EINVAL);
         }
 
-        oa = ll_oa_from_inode(inode, (OBD_MD_FLMODE | OBD_MD_FLID));
-        if (oa == NULL)
+        oa = obdo_alloc();
+        if (oa == NULL) {
                 LBUG();
-        rc = obd_close(ll_i2obdconn(inode), oa);
-        obdo_free(oa);
+                GOTO(out_fd, rc = -ENOENT);
+        }
+        rc = obd_close(ll_i2obdconn(inode), oa, lli->lli_smd);
+        obdo_free(oa); 
         if (rc)
-                GOTO(out, abs(rc));
+                GOTO(out_fd, abs(rc));
 
-        if (file->f_flags & O_WRONLY) {
+        if (file->f_mode & FMODE_WRITE) {
                 struct iattr attr;
                 attr.ia_valid = ATTR_MTIME | ATTR_CTIME | ATTR_ATIME | ATTR_SIZE;
                 attr.ia_mtime = inode->i_mtime;
                 attr.ia_ctime = inode->i_ctime;
                 attr.ia_atime = inode->i_atime;
-                iattr.ia_size = inode->i_size;
+                attr.ia_size = inode->i_size;
 
-                rc = ll_inode_setattr(inode, &iattr, 0);
+                /* XXX: this introduces a small race that we should evaluate */
+                rc = ll_inode_setattr(inode, &attr, 0);
                 if (rc) {
                         CERROR("failed - %d.\n", rc);
                         rc = -EIO; /* XXX - GOTO(out)? -phil */
                 }
         }
 
-        rc = mdc_close(&sbi->ll_mds_client, sbi->ll_mds_conn, inode->i_ino,
+        rc = mdc_close(&sbi->ll_mdc_conn, inode->i_ino,
                        S_IFREG, fd->fd_mdshandle, &req);
         ptlrpc_req_finished(req);
-        if (rc) { 
-                if (rc > 0) 
+        if (rc) {
+                if (rc > 0)
                         rc = -rc;
                 GOTO(out, rc);
         }
         ptlrpc_free_req(fd->fd_req);
 
-        EXIT; 
+        EXIT;
 
- out:
-        if (!rc && fd) {
-                kmem_cache_free(ll_file_data_slab, fd);
-                file->private_data = NULL;
-        }
+out_fd:
+        kmem_cache_free(ll_file_data_slab, fd);
+        file->private_data = NULL;
+out:
         return rc;
 }
 
@@ -162,44 +197,95 @@ static inline void ll_remove_suid(struct inode *inode)
         }
 }
 
+static void ll_update_atime(struct inode *inode)
+{
+        struct iattr attr;
+
+        attr.ia_atime = CURRENT_TIME;
+        attr.ia_valid = ATTR_ATIME;
+
+        if (inode->i_atime == attr.ia_atime) return;
+        if (IS_RDONLY(inode)) return;
+        if (IS_NOATIME(inode)) return;
+
+        /* ll_inode_setattr() sets inode->i_atime from attr.ia_atime */
+        ll_inode_setattr(inode, &attr, 0);
+}
+
+static int ll_lock_callback(struct lustre_handle *lockh,
+                            struct ldlm_lock_desc *new,
+                            void *data, __u32 data_len,
+                            struct ptlrpc_request **reqp)
+{
+        struct inode *inode = data;
+        ENTRY;
+
+        if (new == NULL) {
+                /* Completion AST.  Do nothing. */
+                RETURN(0);
+        }
+
+        if (data_len != sizeof(struct inode))
+                LBUG();
+
+        /* FIXME: do something better than throwing away everything */
+        if (inode == NULL)
+                LBUG();
+        down(&inode->i_sem);
+        CDEBUG(D_INODE, "invalidating obdo/inode %ld\n", inode->i_ino);
+        invalidate_inode_pages(inode);
+        up(&inode->i_sem);
+
+        if (ldlm_cli_cancel(lockh) < 0)
+                LBUG();
+        RETURN(0);
+}
+
 static ssize_t ll_file_read(struct file *filp, char *buf, size_t count,
                             loff_t *ppos)
 {
+        struct ll_file_data *fd = (struct ll_file_data *)filp->private_data;
         struct inode *inode = filp->f_dentry->d_inode;
         struct ll_sb_info *sbi = ll_i2sbi(inode);
         struct ldlm_extent extent;
-        struct ldlm_handle lockh;
+        struct lustre_handle lockh;
         __u64 res_id[RES_NAME_SIZE] = {inode->i_ino};
         int flags = 0;
         ldlm_error_t err;
         ssize_t retval;
         ENTRY;
 
-        extent.start = *ppos;
-        extent.end = *ppos + count;
-        CDEBUG(D_INFO, "Locking inode %ld, start %Lu end %Lu\n",
-               inode->i_ino, extent.start, extent.end);
-
-        err = obd_enqueue(&sbi->ll_conn, sbi->ll_namespace, NULL, res_id,
-                          LDLM_EXTENT, &extent, LCK_PR, &flags, inode,
-                          sizeof(*inode), &lockh);
-        if (err != ELDLM_OK)
-                CERROR("lock enqueue: err: %d\n", err);
-        ldlm_lock_dump((void *)(unsigned long)lockh.addr);
+        if (!(fd->fd_flags & LL_FILE_IGNORE_LOCK)) {
+                extent.start = *ppos;
+                extent.end = *ppos + count;
+                CDEBUG(D_INFO, "Locking inode %ld, start %Lu end %Lu\n",
+                       inode->i_ino, extent.start, extent.end);
+
+                err = obd_enqueue(&sbi->ll_osc_conn, NULL, res_id, LDLM_EXTENT,
+                                  &extent, sizeof(extent), LCK_PR, &flags,
+                                  ll_lock_callback, inode, sizeof(*inode),
+                                  &lockh);
+                if (err != ELDLM_OK) {
+                        CERROR("lock enqueue: err: %d\n", err);
+                        RETURN(err);
+                }
+                ldlm_lock_dump((void *)(unsigned long)lockh.addr);
+        }
 
         CDEBUG(D_INFO, "Reading inode %ld, %d bytes, offset %Ld\n",
                inode->i_ino, count, *ppos);
         retval = generic_file_read(filp, buf, count, ppos);
-        if (retval > 0) {
-                struct iattr attr;
-                attr.ia_valid = ATTR_ATIME;
-                attr.ia_atime = CURRENT_TIME;
-                ll_setattr(filp->f_dentry, &attr);
-        }
 
-        err = obd_cancel(&sbi->ll_conn, LCK_PR, &lockh);
-        if (err != ELDLM_OK)
-                CERROR("lock cancel: err: %d\n", err);
+        if (retval > 0)
+                ll_update_atime(inode);
+
+        if (!(fd->fd_flags & LL_FILE_IGNORE_LOCK)) {
+                err = obd_cancel(&sbi->ll_osc_conn, LCK_PR, &lockh);
+                if (err != ELDLM_OK) {
+                        CERROR("lock cancel: err: %d\n", err);
+                        RETURN(err);
+                }
+        }
 
         RETURN(retval);
 }
@@ -210,40 +296,92 @@ static ssize_t ll_file_read(struct file *filp, char *buf, size_t count,
 static ssize_t
 ll_file_write(struct file *file, const char *buf, size_t count, loff_t *ppos)
 {
+        struct ll_file_data *fd = (struct ll_file_data *)file->private_data;
         struct inode *inode = file->f_dentry->d_inode;
         struct ll_sb_info *sbi = ll_i2sbi(inode);
         struct ldlm_extent extent;
-        struct ldlm_handle lockh;
+        struct lustre_handle lockh;
         __u64 res_id[RES_NAME_SIZE] = {inode->i_ino};
         int flags = 0;
         ldlm_error_t err;
         ssize_t retval;
         ENTRY;
 
-        extent.start = *ppos;
-        extent.end = *ppos + count;
-        CDEBUG(D_INFO, "Locking inode %ld, start %Lu end %Lu\n",
-               inode->i_ino, extent.start, extent.end);
-
-        err = obd_enqueue(&sbi->ll_conn, sbi->ll_namespace, NULL, res_id,
-                          LDLM_EXTENT, &extent, LCK_PW, &flags, inode,
-                          sizeof(*inode), &lockh);
-        if (err != ELDLM_OK)
-                CERROR("lock enqueue: err: %d\n", err);
-        ldlm_lock_dump((void *)(unsigned long)lockh.addr);
+        if (!(fd->fd_flags & LL_FILE_IGNORE_LOCK)) {
+                /* FIXME: this should check whether O_APPEND is set and adjust
+                 * extent.start accordingly */
+                extent.start = *ppos;
+                extent.end = *ppos + count;
+                CDEBUG(D_INFO, "Locking inode %ld, start %Lu end %Lu\n",
+                       inode->i_ino, extent.start, extent.end);
+
+                err = obd_enqueue(&sbi->ll_osc_conn, NULL, res_id, LDLM_EXTENT,
+                                  &extent, sizeof(extent), LCK_PW, &flags,
+                                  ll_lock_callback, inode, sizeof(*inode),
+                                  &lockh);
+                if (err != ELDLM_OK) {
+                        CERROR("lock enqueue: err: %d\n", err);
+                        RETURN(err);
+                }
+                ldlm_lock_dump((void *)(unsigned long)lockh.addr);
+        }
 
         CDEBUG(D_INFO, "Writing inode %ld, %ld bytes, offset %Ld\n",
                inode->i_ino, (long)count, *ppos);
 
         retval = generic_file_write(file, buf, count, ppos);
 
-        err = obd_cancel(&sbi->ll_conn, LCK_PW, &lockh);
-        if (err != ELDLM_OK)
-                CERROR("lock cancel: err: %d\n", err);
+        if (!(fd->fd_flags & LL_FILE_IGNORE_LOCK)) {
+                err = obd_cancel(&sbi->ll_osc_conn, LCK_PW, &lockh);
+                if (err != ELDLM_OK) {
+                        CERROR("lock cancel: err: %d\n", err);
+                        RETURN(err);
+                }
+        }
 
         RETURN(retval);
 }
 
+int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
+                  unsigned long arg)
+{
+        struct ll_file_data *fd = (struct ll_file_data *)file->private_data;
+        int flags;
+
+        switch(cmd) {
+        case LL_IOC_GETFLAGS:
+                /* Get the current value of the file flags */
+                return put_user(fd->fd_flags, (int *)arg);
+        case LL_IOC_SETFLAGS:
+        case LL_IOC_CLRFLAGS:
+                /* Set or clear specific file flags */
+                /* XXX This probably needs checks to ensure the flags are
+                 *     not abused, and to handle any flag side effects.
+                 */
+                if (get_user(flags, (int *) arg))
+                        return -EFAULT;
+
+                if (cmd == LL_IOC_SETFLAGS)
+                        fd->fd_flags |= flags;
+                else
+                        fd->fd_flags &= ~flags;
+                return 0;
+
+        /* We need to special case any other ioctls we want to handle,
+         * to send them to the MDS/OST as appropriate and to properly
+         * network encode the arg field.
+        case EXT2_IOC_GETFLAGS:
+        case EXT2_IOC_SETFLAGS:
+        case EXT2_IOC_GETVERSION_OLD:
+        case EXT2_IOC_GETVERSION_NEW:
+        case EXT2_IOC_SETVERSION_OLD:
+        case EXT2_IOC_SETVERSION_NEW:
+        */
+        default:
+                return -ENOTTY;
+        }
+}
+
 /* XXX this does not need to do anything for data, it _does_ need to
    call setattr */
 int ll_fsync(struct file *file, struct dentry *dentry, int data)
@@ -252,12 +390,13 @@ int ll_fsync(struct file *file, struct dentry *dentry, int data)
 }
 
 struct file_operations ll_file_operations = {
-        read: ll_file_read,
-        write: ll_file_write,
-        open: ll_file_open,
-        release: ll_file_release,
-        mmap: generic_file_mmap,
-        fsync: NULL
+        read:           ll_file_read,
+        write:          ll_file_write,
+        ioctl:          ll_file_ioctl,
+        open:           ll_file_open,
+        release:        ll_file_release,
+        mmap:           generic_file_mmap,
+        fsync:          NULL
 };
 
 struct inode_operations ll_file_inode_operations = {