134a9dd7eSJoel Becker /* -*- mode: c; c-basic-offset: 8; -*- 234a9dd7eSJoel Becker * vim: noexpandtab sw=8 ts=8 sts=0: 334a9dd7eSJoel Becker * 434a9dd7eSJoel Becker * dlmfs.c 534a9dd7eSJoel Becker * 634a9dd7eSJoel Becker * Code which implements the kernel side of a minimal userspace 734a9dd7eSJoel Becker * interface to our DLM. This file handles the virtual file system 834a9dd7eSJoel Becker * used for communication with userspace. Credit should go to ramfs, 934a9dd7eSJoel Becker * which was a template for the fs side of this module. 1034a9dd7eSJoel Becker * 1134a9dd7eSJoel Becker * Copyright (C) 2003, 2004 Oracle. All rights reserved. 1234a9dd7eSJoel Becker * 1334a9dd7eSJoel Becker * This program is free software; you can redistribute it and/or 1434a9dd7eSJoel Becker * modify it under the terms of the GNU General Public 1534a9dd7eSJoel Becker * License as published by the Free Software Foundation; either 1634a9dd7eSJoel Becker * version 2 of the License, or (at your option) any later version. 1734a9dd7eSJoel Becker * 1834a9dd7eSJoel Becker * This program is distributed in the hope that it will be useful, 1934a9dd7eSJoel Becker * but WITHOUT ANY WARRANTY; without even the implied warranty of 2034a9dd7eSJoel Becker * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 2134a9dd7eSJoel Becker * General Public License for more details. 2234a9dd7eSJoel Becker * 2334a9dd7eSJoel Becker * You should have received a copy of the GNU General Public 2434a9dd7eSJoel Becker * License along with this program; if not, write to the 2534a9dd7eSJoel Becker * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 2634a9dd7eSJoel Becker * Boston, MA 021110-1307, USA. 2734a9dd7eSJoel Becker */ 2834a9dd7eSJoel Becker 2934a9dd7eSJoel Becker /* Simple VFS hooks based on: */ 3034a9dd7eSJoel Becker /* 3134a9dd7eSJoel Becker * Resizable simple ram filesystem for Linux. 3234a9dd7eSJoel Becker * 3334a9dd7eSJoel Becker * Copyright (C) 2000 Linus Torvalds. 3434a9dd7eSJoel Becker * 2000 Transmeta Corp. 3534a9dd7eSJoel Becker */ 3634a9dd7eSJoel Becker 3734a9dd7eSJoel Becker #include <linux/module.h> 3834a9dd7eSJoel Becker #include <linux/fs.h> 3934a9dd7eSJoel Becker #include <linux/pagemap.h> 4034a9dd7eSJoel Becker #include <linux/types.h> 4134a9dd7eSJoel Becker #include <linux/slab.h> 4234a9dd7eSJoel Becker #include <linux/highmem.h> 4334a9dd7eSJoel Becker #include <linux/init.h> 4434a9dd7eSJoel Becker #include <linux/string.h> 4534a9dd7eSJoel Becker #include <linux/backing-dev.h> 4634a9dd7eSJoel Becker #include <linux/poll.h> 4734a9dd7eSJoel Becker 4834a9dd7eSJoel Becker #include <asm/uaccess.h> 4934a9dd7eSJoel Becker 500016eedcSJoel Becker #include "stackglue.h" 5134a9dd7eSJoel Becker #include "userdlm.h" 5234a9dd7eSJoel Becker #include "dlmfsver.h" 5334a9dd7eSJoel Becker 5434a9dd7eSJoel Becker #define MLOG_MASK_PREFIX ML_DLMFS 5534a9dd7eSJoel Becker #include "cluster/masklog.h" 5634a9dd7eSJoel Becker 5734a9dd7eSJoel Becker 5834a9dd7eSJoel Becker static const struct super_operations dlmfs_ops; 5934a9dd7eSJoel Becker static const struct file_operations dlmfs_file_operations; 6034a9dd7eSJoel Becker static const struct inode_operations dlmfs_dir_inode_operations; 6134a9dd7eSJoel Becker static const struct inode_operations dlmfs_root_inode_operations; 6234a9dd7eSJoel Becker static const struct inode_operations dlmfs_file_inode_operations; 6334a9dd7eSJoel Becker static struct kmem_cache *dlmfs_inode_cache; 6434a9dd7eSJoel Becker 6534a9dd7eSJoel Becker struct workqueue_struct *user_dlm_worker; 6634a9dd7eSJoel Becker 6734a9dd7eSJoel Becker 6834a9dd7eSJoel Becker 6934a9dd7eSJoel Becker /* 7034a9dd7eSJoel Becker * These are the ABI capabilities of dlmfs. 7134a9dd7eSJoel Becker * 7234a9dd7eSJoel Becker * Over time, dlmfs has added some features that were not part of the 7334a9dd7eSJoel Becker * initial ABI. Unfortunately, some of these features are not detectable 7434a9dd7eSJoel Becker * via standard usage. For example, Linux's default poll always returns 7534a9dd7eSJoel Becker * POLLIN, so there is no way for a caller of poll(2) to know when dlmfs 7634a9dd7eSJoel Becker * added poll support. Instead, we provide this list of new capabilities. 7734a9dd7eSJoel Becker * 7834a9dd7eSJoel Becker * Capabilities is a read-only attribute. We do it as a module parameter 7934a9dd7eSJoel Becker * so we can discover it whether dlmfs is built in, loaded, or even not 8034a9dd7eSJoel Becker * loaded. 8134a9dd7eSJoel Becker * 8234a9dd7eSJoel Becker * The ABI features are local to this machine's dlmfs mount. This is 8334a9dd7eSJoel Becker * distinct from the locking protocol, which is concerned with inter-node 8434a9dd7eSJoel Becker * interaction. 8534a9dd7eSJoel Becker * 8634a9dd7eSJoel Becker * Capabilities: 8734a9dd7eSJoel Becker * - bast : POLLIN against the file descriptor of a held lock 8834a9dd7eSJoel Becker * signifies a bast fired on the lock. 8934a9dd7eSJoel Becker */ 90cbe0e331SJoel Becker #define DLMFS_CAPABILITIES "bast stackglue" 9198ba073cSRobin Dong static int param_set_dlmfs_capabilities(const char *val, 9234a9dd7eSJoel Becker struct kernel_param *kp) 9334a9dd7eSJoel Becker { 9434a9dd7eSJoel Becker printk(KERN_ERR "%s: readonly parameter\n", kp->name); 9534a9dd7eSJoel Becker return -EINVAL; 9634a9dd7eSJoel Becker } 9734a9dd7eSJoel Becker static int param_get_dlmfs_capabilities(char *buffer, 9834a9dd7eSJoel Becker struct kernel_param *kp) 9934a9dd7eSJoel Becker { 10034a9dd7eSJoel Becker return strlcpy(buffer, DLMFS_CAPABILITIES, 10134a9dd7eSJoel Becker strlen(DLMFS_CAPABILITIES) + 1); 10234a9dd7eSJoel Becker } 10334a9dd7eSJoel Becker module_param_call(capabilities, param_set_dlmfs_capabilities, 10434a9dd7eSJoel Becker param_get_dlmfs_capabilities, NULL, 0444); 10534a9dd7eSJoel Becker MODULE_PARM_DESC(capabilities, DLMFS_CAPABILITIES); 10634a9dd7eSJoel Becker 10734a9dd7eSJoel Becker 10834a9dd7eSJoel Becker /* 10934a9dd7eSJoel Becker * decodes a set of open flags into a valid lock level and a set of flags. 11034a9dd7eSJoel Becker * returns < 0 if we have invalid flags 11134a9dd7eSJoel Becker * flags which mean something to us: 11234a9dd7eSJoel Becker * O_RDONLY -> PRMODE level 11334a9dd7eSJoel Becker * O_WRONLY -> EXMODE level 11434a9dd7eSJoel Becker * 115efd647f7STao Ma * O_NONBLOCK -> NOQUEUE 11634a9dd7eSJoel Becker */ 11734a9dd7eSJoel Becker static int dlmfs_decode_open_flags(int open_flags, 11834a9dd7eSJoel Becker int *level, 11934a9dd7eSJoel Becker int *flags) 12034a9dd7eSJoel Becker { 12134a9dd7eSJoel Becker if (open_flags & (O_WRONLY|O_RDWR)) 122efd647f7STao Ma *level = DLM_LOCK_EX; 12334a9dd7eSJoel Becker else 124efd647f7STao Ma *level = DLM_LOCK_PR; 12534a9dd7eSJoel Becker 12634a9dd7eSJoel Becker *flags = 0; 12734a9dd7eSJoel Becker if (open_flags & O_NONBLOCK) 128efd647f7STao Ma *flags |= DLM_LKF_NOQUEUE; 12934a9dd7eSJoel Becker 13034a9dd7eSJoel Becker return 0; 13134a9dd7eSJoel Becker } 13234a9dd7eSJoel Becker 13334a9dd7eSJoel Becker static int dlmfs_file_open(struct inode *inode, 13434a9dd7eSJoel Becker struct file *file) 13534a9dd7eSJoel Becker { 13634a9dd7eSJoel Becker int status, level, flags; 13734a9dd7eSJoel Becker struct dlmfs_filp_private *fp = NULL; 13834a9dd7eSJoel Becker struct dlmfs_inode_private *ip; 13934a9dd7eSJoel Becker 14034a9dd7eSJoel Becker if (S_ISDIR(inode->i_mode)) 14134a9dd7eSJoel Becker BUG(); 14234a9dd7eSJoel Becker 14334a9dd7eSJoel Becker mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino, 14434a9dd7eSJoel Becker file->f_flags); 14534a9dd7eSJoel Becker 14634a9dd7eSJoel Becker status = dlmfs_decode_open_flags(file->f_flags, &level, &flags); 14734a9dd7eSJoel Becker if (status < 0) 14834a9dd7eSJoel Becker goto bail; 14934a9dd7eSJoel Becker 15034a9dd7eSJoel Becker /* We don't want to honor O_APPEND at read/write time as it 15134a9dd7eSJoel Becker * doesn't make sense for LVB writes. */ 15234a9dd7eSJoel Becker file->f_flags &= ~O_APPEND; 15334a9dd7eSJoel Becker 15434a9dd7eSJoel Becker fp = kmalloc(sizeof(*fp), GFP_NOFS); 15534a9dd7eSJoel Becker if (!fp) { 15634a9dd7eSJoel Becker status = -ENOMEM; 15734a9dd7eSJoel Becker goto bail; 15834a9dd7eSJoel Becker } 15934a9dd7eSJoel Becker fp->fp_lock_level = level; 16034a9dd7eSJoel Becker 16134a9dd7eSJoel Becker ip = DLMFS_I(inode); 16234a9dd7eSJoel Becker 16334a9dd7eSJoel Becker status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags); 16434a9dd7eSJoel Becker if (status < 0) { 16534a9dd7eSJoel Becker /* this is a strange error to return here but I want 16634a9dd7eSJoel Becker * to be able userspace to be able to distinguish a 16734a9dd7eSJoel Becker * valid lock request from one that simply couldn't be 16834a9dd7eSJoel Becker * granted. */ 169efd647f7STao Ma if (flags & DLM_LKF_NOQUEUE && status == -EAGAIN) 17034a9dd7eSJoel Becker status = -ETXTBSY; 17134a9dd7eSJoel Becker kfree(fp); 17234a9dd7eSJoel Becker goto bail; 17334a9dd7eSJoel Becker } 17434a9dd7eSJoel Becker 17534a9dd7eSJoel Becker file->private_data = fp; 17634a9dd7eSJoel Becker bail: 17734a9dd7eSJoel Becker return status; 17834a9dd7eSJoel Becker } 17934a9dd7eSJoel Becker 18034a9dd7eSJoel Becker static int dlmfs_file_release(struct inode *inode, 18134a9dd7eSJoel Becker struct file *file) 18234a9dd7eSJoel Becker { 18334a9dd7eSJoel Becker int level, status; 18434a9dd7eSJoel Becker struct dlmfs_inode_private *ip = DLMFS_I(inode); 18533fa1d90SJoe Perches struct dlmfs_filp_private *fp = file->private_data; 18634a9dd7eSJoel Becker 18734a9dd7eSJoel Becker if (S_ISDIR(inode->i_mode)) 18834a9dd7eSJoel Becker BUG(); 18934a9dd7eSJoel Becker 19034a9dd7eSJoel Becker mlog(0, "close called on inode %lu\n", inode->i_ino); 19134a9dd7eSJoel Becker 19234a9dd7eSJoel Becker status = 0; 19334a9dd7eSJoel Becker if (fp) { 19434a9dd7eSJoel Becker level = fp->fp_lock_level; 195efd647f7STao Ma if (level != DLM_LOCK_IV) 19634a9dd7eSJoel Becker user_dlm_cluster_unlock(&ip->ip_lockres, level); 19734a9dd7eSJoel Becker 19834a9dd7eSJoel Becker kfree(fp); 19934a9dd7eSJoel Becker file->private_data = NULL; 20034a9dd7eSJoel Becker } 20134a9dd7eSJoel Becker 20234a9dd7eSJoel Becker return 0; 20334a9dd7eSJoel Becker } 20434a9dd7eSJoel Becker 205e8fce482SJoel Becker /* 206e8fce482SJoel Becker * We do ->setattr() just to override size changes. Our size is the size 207e8fce482SJoel Becker * of the LVB and nothing else. 208e8fce482SJoel Becker */ 209e8fce482SJoel Becker static int dlmfs_file_setattr(struct dentry *dentry, struct iattr *attr) 210e8fce482SJoel Becker { 211e8fce482SJoel Becker int error; 212e8fce482SJoel Becker struct inode *inode = dentry->d_inode; 213e8fce482SJoel Becker 214e8fce482SJoel Becker attr->ia_valid &= ~ATTR_SIZE; 215e8fce482SJoel Becker error = inode_change_ok(inode, attr); 2161025774cSChristoph Hellwig if (error) 217e8fce482SJoel Becker return error; 2181025774cSChristoph Hellwig 2191025774cSChristoph Hellwig setattr_copy(inode, attr); 2201025774cSChristoph Hellwig mark_inode_dirty(inode); 2211025774cSChristoph Hellwig return 0; 222e8fce482SJoel Becker } 223e8fce482SJoel Becker 22434a9dd7eSJoel Becker static unsigned int dlmfs_file_poll(struct file *file, poll_table *wait) 22534a9dd7eSJoel Becker { 22634a9dd7eSJoel Becker int event = 0; 22734a9dd7eSJoel Becker struct inode *inode = file->f_path.dentry->d_inode; 22834a9dd7eSJoel Becker struct dlmfs_inode_private *ip = DLMFS_I(inode); 22934a9dd7eSJoel Becker 23034a9dd7eSJoel Becker poll_wait(file, &ip->ip_lockres.l_event, wait); 23134a9dd7eSJoel Becker 23234a9dd7eSJoel Becker spin_lock(&ip->ip_lockres.l_lock); 23334a9dd7eSJoel Becker if (ip->ip_lockres.l_flags & USER_LOCK_BLOCKED) 23434a9dd7eSJoel Becker event = POLLIN | POLLRDNORM; 23534a9dd7eSJoel Becker spin_unlock(&ip->ip_lockres.l_lock); 23634a9dd7eSJoel Becker 23734a9dd7eSJoel Becker return event; 23834a9dd7eSJoel Becker } 23934a9dd7eSJoel Becker 24034a9dd7eSJoel Becker static ssize_t dlmfs_file_read(struct file *filp, 24134a9dd7eSJoel Becker char __user *buf, 24234a9dd7eSJoel Becker size_t count, 24334a9dd7eSJoel Becker loff_t *ppos) 24434a9dd7eSJoel Becker { 24534a9dd7eSJoel Becker int bytes_left; 2460016eedcSJoel Becker ssize_t readlen, got; 24734a9dd7eSJoel Becker char *lvb_buf; 24834a9dd7eSJoel Becker struct inode *inode = filp->f_path.dentry->d_inode; 24934a9dd7eSJoel Becker 25034a9dd7eSJoel Becker mlog(0, "inode %lu, count = %zu, *ppos = %llu\n", 25134a9dd7eSJoel Becker inode->i_ino, count, *ppos); 25234a9dd7eSJoel Becker 25334a9dd7eSJoel Becker if (*ppos >= i_size_read(inode)) 25434a9dd7eSJoel Becker return 0; 25534a9dd7eSJoel Becker 25634a9dd7eSJoel Becker if (!count) 25734a9dd7eSJoel Becker return 0; 25834a9dd7eSJoel Becker 25934a9dd7eSJoel Becker if (!access_ok(VERIFY_WRITE, buf, count)) 26034a9dd7eSJoel Becker return -EFAULT; 26134a9dd7eSJoel Becker 26234a9dd7eSJoel Becker /* don't read past the lvb */ 26334a9dd7eSJoel Becker if ((count + *ppos) > i_size_read(inode)) 26434a9dd7eSJoel Becker readlen = i_size_read(inode) - *ppos; 26534a9dd7eSJoel Becker else 266a36d515cSJoel Becker readlen = count; 26734a9dd7eSJoel Becker 26834a9dd7eSJoel Becker lvb_buf = kmalloc(readlen, GFP_NOFS); 26934a9dd7eSJoel Becker if (!lvb_buf) 27034a9dd7eSJoel Becker return -ENOMEM; 27134a9dd7eSJoel Becker 2720016eedcSJoel Becker got = user_dlm_read_lvb(inode, lvb_buf, readlen); 2730016eedcSJoel Becker if (got) { 2740016eedcSJoel Becker BUG_ON(got != readlen); 27534a9dd7eSJoel Becker bytes_left = __copy_to_user(buf, lvb_buf, readlen); 27634a9dd7eSJoel Becker readlen -= bytes_left; 2770016eedcSJoel Becker } else 2780016eedcSJoel Becker readlen = 0; 27934a9dd7eSJoel Becker 28034a9dd7eSJoel Becker kfree(lvb_buf); 28134a9dd7eSJoel Becker 28234a9dd7eSJoel Becker *ppos = *ppos + readlen; 28334a9dd7eSJoel Becker 28434a9dd7eSJoel Becker mlog(0, "read %zd bytes\n", readlen); 28534a9dd7eSJoel Becker return readlen; 28634a9dd7eSJoel Becker } 28734a9dd7eSJoel Becker 28834a9dd7eSJoel Becker static ssize_t dlmfs_file_write(struct file *filp, 28934a9dd7eSJoel Becker const char __user *buf, 29034a9dd7eSJoel Becker size_t count, 29134a9dd7eSJoel Becker loff_t *ppos) 29234a9dd7eSJoel Becker { 29334a9dd7eSJoel Becker int bytes_left; 29434a9dd7eSJoel Becker ssize_t writelen; 29534a9dd7eSJoel Becker char *lvb_buf; 29634a9dd7eSJoel Becker struct inode *inode = filp->f_path.dentry->d_inode; 29734a9dd7eSJoel Becker 29834a9dd7eSJoel Becker mlog(0, "inode %lu, count = %zu, *ppos = %llu\n", 29934a9dd7eSJoel Becker inode->i_ino, count, *ppos); 30034a9dd7eSJoel Becker 30134a9dd7eSJoel Becker if (*ppos >= i_size_read(inode)) 30234a9dd7eSJoel Becker return -ENOSPC; 30334a9dd7eSJoel Becker 30434a9dd7eSJoel Becker if (!count) 30534a9dd7eSJoel Becker return 0; 30634a9dd7eSJoel Becker 30734a9dd7eSJoel Becker if (!access_ok(VERIFY_READ, buf, count)) 30834a9dd7eSJoel Becker return -EFAULT; 30934a9dd7eSJoel Becker 31034a9dd7eSJoel Becker /* don't write past the lvb */ 31134a9dd7eSJoel Becker if ((count + *ppos) > i_size_read(inode)) 31234a9dd7eSJoel Becker writelen = i_size_read(inode) - *ppos; 31334a9dd7eSJoel Becker else 31434a9dd7eSJoel Becker writelen = count - *ppos; 31534a9dd7eSJoel Becker 31634a9dd7eSJoel Becker lvb_buf = kmalloc(writelen, GFP_NOFS); 31734a9dd7eSJoel Becker if (!lvb_buf) 31834a9dd7eSJoel Becker return -ENOMEM; 31934a9dd7eSJoel Becker 32034a9dd7eSJoel Becker bytes_left = copy_from_user(lvb_buf, buf, writelen); 32134a9dd7eSJoel Becker writelen -= bytes_left; 32234a9dd7eSJoel Becker if (writelen) 32334a9dd7eSJoel Becker user_dlm_write_lvb(inode, lvb_buf, writelen); 32434a9dd7eSJoel Becker 32534a9dd7eSJoel Becker kfree(lvb_buf); 32634a9dd7eSJoel Becker 32734a9dd7eSJoel Becker *ppos = *ppos + writelen; 32834a9dd7eSJoel Becker mlog(0, "wrote %zd bytes\n", writelen); 32934a9dd7eSJoel Becker return writelen; 33034a9dd7eSJoel Becker } 33134a9dd7eSJoel Becker 33234a9dd7eSJoel Becker static void dlmfs_init_once(void *foo) 33334a9dd7eSJoel Becker { 33434a9dd7eSJoel Becker struct dlmfs_inode_private *ip = 33534a9dd7eSJoel Becker (struct dlmfs_inode_private *) foo; 33634a9dd7eSJoel Becker 3370016eedcSJoel Becker ip->ip_conn = NULL; 33834a9dd7eSJoel Becker ip->ip_parent = NULL; 33934a9dd7eSJoel Becker 34034a9dd7eSJoel Becker inode_init_once(&ip->ip_vfs_inode); 34134a9dd7eSJoel Becker } 34234a9dd7eSJoel Becker 34334a9dd7eSJoel Becker static struct inode *dlmfs_alloc_inode(struct super_block *sb) 34434a9dd7eSJoel Becker { 34534a9dd7eSJoel Becker struct dlmfs_inode_private *ip; 34634a9dd7eSJoel Becker 34734a9dd7eSJoel Becker ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS); 34834a9dd7eSJoel Becker if (!ip) 34934a9dd7eSJoel Becker return NULL; 35034a9dd7eSJoel Becker 35134a9dd7eSJoel Becker return &ip->ip_vfs_inode; 35234a9dd7eSJoel Becker } 35334a9dd7eSJoel Becker 354fa0d7e3dSNick Piggin static void dlmfs_i_callback(struct rcu_head *head) 355fa0d7e3dSNick Piggin { 356fa0d7e3dSNick Piggin struct inode *inode = container_of(head, struct inode, i_rcu); 357fa0d7e3dSNick Piggin INIT_LIST_HEAD(&inode->i_dentry); 358fa0d7e3dSNick Piggin kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode)); 359fa0d7e3dSNick Piggin } 360fa0d7e3dSNick Piggin 36134a9dd7eSJoel Becker static void dlmfs_destroy_inode(struct inode *inode) 36234a9dd7eSJoel Becker { 363fa0d7e3dSNick Piggin call_rcu(&inode->i_rcu, dlmfs_i_callback); 36434a9dd7eSJoel Becker } 36534a9dd7eSJoel Becker 366b57922d9SAl Viro static void dlmfs_evict_inode(struct inode *inode) 36734a9dd7eSJoel Becker { 36834a9dd7eSJoel Becker int status; 36934a9dd7eSJoel Becker struct dlmfs_inode_private *ip; 37034a9dd7eSJoel Becker 371b57922d9SAl Viro end_writeback(inode); 37234a9dd7eSJoel Becker 37334a9dd7eSJoel Becker mlog(0, "inode %lu\n", inode->i_ino); 37434a9dd7eSJoel Becker 37534a9dd7eSJoel Becker ip = DLMFS_I(inode); 37634a9dd7eSJoel Becker 37734a9dd7eSJoel Becker if (S_ISREG(inode->i_mode)) { 37834a9dd7eSJoel Becker status = user_dlm_destroy_lock(&ip->ip_lockres); 37934a9dd7eSJoel Becker if (status < 0) 38034a9dd7eSJoel Becker mlog_errno(status); 38134a9dd7eSJoel Becker iput(ip->ip_parent); 38234a9dd7eSJoel Becker goto clear_fields; 38334a9dd7eSJoel Becker } 38434a9dd7eSJoel Becker 3850016eedcSJoel Becker mlog(0, "we're a directory, ip->ip_conn = 0x%p\n", ip->ip_conn); 38634a9dd7eSJoel Becker /* we must be a directory. If required, lets unregister the 38734a9dd7eSJoel Becker * dlm context now. */ 3880016eedcSJoel Becker if (ip->ip_conn) 3890016eedcSJoel Becker user_dlm_unregister(ip->ip_conn); 39034a9dd7eSJoel Becker clear_fields: 39134a9dd7eSJoel Becker ip->ip_parent = NULL; 3920016eedcSJoel Becker ip->ip_conn = NULL; 39334a9dd7eSJoel Becker } 39434a9dd7eSJoel Becker 39534a9dd7eSJoel Becker static struct backing_dev_info dlmfs_backing_dev_info = { 39634a9dd7eSJoel Becker .name = "ocfs2-dlmfs", 39734a9dd7eSJoel Becker .ra_pages = 0, /* No readahead */ 39834a9dd7eSJoel Becker .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, 39934a9dd7eSJoel Becker }; 40034a9dd7eSJoel Becker 40134a9dd7eSJoel Becker static struct inode *dlmfs_get_root_inode(struct super_block *sb) 40234a9dd7eSJoel Becker { 40334a9dd7eSJoel Becker struct inode *inode = new_inode(sb); 40434a9dd7eSJoel Becker int mode = S_IFDIR | 0755; 40534a9dd7eSJoel Becker struct dlmfs_inode_private *ip; 40634a9dd7eSJoel Becker 40734a9dd7eSJoel Becker if (inode) { 40834a9dd7eSJoel Becker ip = DLMFS_I(inode); 40934a9dd7eSJoel Becker 41085fe4025SChristoph Hellwig inode->i_ino = get_next_ino(); 41134a9dd7eSJoel Becker inode->i_mode = mode; 41234a9dd7eSJoel Becker inode->i_uid = current_fsuid(); 41334a9dd7eSJoel Becker inode->i_gid = current_fsgid(); 41434a9dd7eSJoel Becker inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info; 41534a9dd7eSJoel Becker inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 41634a9dd7eSJoel Becker inc_nlink(inode); 41734a9dd7eSJoel Becker 41834a9dd7eSJoel Becker inode->i_fop = &simple_dir_operations; 41934a9dd7eSJoel Becker inode->i_op = &dlmfs_root_inode_operations; 42034a9dd7eSJoel Becker } 42134a9dd7eSJoel Becker 42234a9dd7eSJoel Becker return inode; 42334a9dd7eSJoel Becker } 42434a9dd7eSJoel Becker 42534a9dd7eSJoel Becker static struct inode *dlmfs_get_inode(struct inode *parent, 42634a9dd7eSJoel Becker struct dentry *dentry, 42734a9dd7eSJoel Becker int mode) 42834a9dd7eSJoel Becker { 42934a9dd7eSJoel Becker struct super_block *sb = parent->i_sb; 43034a9dd7eSJoel Becker struct inode * inode = new_inode(sb); 43134a9dd7eSJoel Becker struct dlmfs_inode_private *ip; 43234a9dd7eSJoel Becker 43334a9dd7eSJoel Becker if (!inode) 43434a9dd7eSJoel Becker return NULL; 43534a9dd7eSJoel Becker 43685fe4025SChristoph Hellwig inode->i_ino = get_next_ino(); 43734a9dd7eSJoel Becker inode->i_mode = mode; 43834a9dd7eSJoel Becker inode->i_uid = current_fsuid(); 43934a9dd7eSJoel Becker inode->i_gid = current_fsgid(); 44034a9dd7eSJoel Becker inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info; 44134a9dd7eSJoel Becker inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 44234a9dd7eSJoel Becker 44334a9dd7eSJoel Becker ip = DLMFS_I(inode); 4440016eedcSJoel Becker ip->ip_conn = DLMFS_I(parent)->ip_conn; 44534a9dd7eSJoel Becker 44634a9dd7eSJoel Becker switch (mode & S_IFMT) { 44734a9dd7eSJoel Becker default: 44834a9dd7eSJoel Becker /* for now we don't support anything other than 44934a9dd7eSJoel Becker * directories and regular files. */ 45034a9dd7eSJoel Becker BUG(); 45134a9dd7eSJoel Becker break; 45234a9dd7eSJoel Becker case S_IFREG: 45334a9dd7eSJoel Becker inode->i_op = &dlmfs_file_inode_operations; 45434a9dd7eSJoel Becker inode->i_fop = &dlmfs_file_operations; 45534a9dd7eSJoel Becker 45634a9dd7eSJoel Becker i_size_write(inode, DLM_LVB_LEN); 45734a9dd7eSJoel Becker 45834a9dd7eSJoel Becker user_dlm_lock_res_init(&ip->ip_lockres, dentry); 45934a9dd7eSJoel Becker 46034a9dd7eSJoel Becker /* released at clear_inode time, this insures that we 46134a9dd7eSJoel Becker * get to drop the dlm reference on each lock *before* 46234a9dd7eSJoel Becker * we call the unregister code for releasing parent 46334a9dd7eSJoel Becker * directories. */ 46434a9dd7eSJoel Becker ip->ip_parent = igrab(parent); 46534a9dd7eSJoel Becker BUG_ON(!ip->ip_parent); 46634a9dd7eSJoel Becker break; 46734a9dd7eSJoel Becker case S_IFDIR: 46834a9dd7eSJoel Becker inode->i_op = &dlmfs_dir_inode_operations; 46934a9dd7eSJoel Becker inode->i_fop = &simple_dir_operations; 47034a9dd7eSJoel Becker 47134a9dd7eSJoel Becker /* directory inodes start off with i_nlink == 47234a9dd7eSJoel Becker * 2 (for "." entry) */ 47334a9dd7eSJoel Becker inc_nlink(inode); 47434a9dd7eSJoel Becker break; 47534a9dd7eSJoel Becker } 47634a9dd7eSJoel Becker 47734a9dd7eSJoel Becker if (parent->i_mode & S_ISGID) { 47834a9dd7eSJoel Becker inode->i_gid = parent->i_gid; 47934a9dd7eSJoel Becker if (S_ISDIR(mode)) 48034a9dd7eSJoel Becker inode->i_mode |= S_ISGID; 48134a9dd7eSJoel Becker } 48234a9dd7eSJoel Becker 48334a9dd7eSJoel Becker return inode; 48434a9dd7eSJoel Becker } 48534a9dd7eSJoel Becker 48634a9dd7eSJoel Becker /* 48734a9dd7eSJoel Becker * File creation. Allocate an inode, and we're done.. 48834a9dd7eSJoel Becker */ 48934a9dd7eSJoel Becker /* SMP-safe */ 49034a9dd7eSJoel Becker static int dlmfs_mkdir(struct inode * dir, 49134a9dd7eSJoel Becker struct dentry * dentry, 49234a9dd7eSJoel Becker int mode) 49334a9dd7eSJoel Becker { 49434a9dd7eSJoel Becker int status; 49534a9dd7eSJoel Becker struct inode *inode = NULL; 49634a9dd7eSJoel Becker struct qstr *domain = &dentry->d_name; 49734a9dd7eSJoel Becker struct dlmfs_inode_private *ip; 4980016eedcSJoel Becker struct ocfs2_cluster_connection *conn; 49934a9dd7eSJoel Becker 50034a9dd7eSJoel Becker mlog(0, "mkdir %.*s\n", domain->len, domain->name); 50134a9dd7eSJoel Becker 50234a9dd7eSJoel Becker /* verify that we have a proper domain */ 5030016eedcSJoel Becker if (domain->len >= GROUP_NAME_MAX) { 50434a9dd7eSJoel Becker status = -EINVAL; 50534a9dd7eSJoel Becker mlog(ML_ERROR, "invalid domain name for directory.\n"); 50634a9dd7eSJoel Becker goto bail; 50734a9dd7eSJoel Becker } 50834a9dd7eSJoel Becker 50934a9dd7eSJoel Becker inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR); 51034a9dd7eSJoel Becker if (!inode) { 51134a9dd7eSJoel Becker status = -ENOMEM; 51234a9dd7eSJoel Becker mlog_errno(status); 51334a9dd7eSJoel Becker goto bail; 51434a9dd7eSJoel Becker } 51534a9dd7eSJoel Becker 51634a9dd7eSJoel Becker ip = DLMFS_I(inode); 51734a9dd7eSJoel Becker 5180016eedcSJoel Becker conn = user_dlm_register(domain); 5190016eedcSJoel Becker if (IS_ERR(conn)) { 5200016eedcSJoel Becker status = PTR_ERR(conn); 52134a9dd7eSJoel Becker mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n", 52234a9dd7eSJoel Becker status, domain->len, domain->name); 52334a9dd7eSJoel Becker goto bail; 52434a9dd7eSJoel Becker } 5250016eedcSJoel Becker ip->ip_conn = conn; 52634a9dd7eSJoel Becker 52734a9dd7eSJoel Becker inc_nlink(dir); 52834a9dd7eSJoel Becker d_instantiate(dentry, inode); 52934a9dd7eSJoel Becker dget(dentry); /* Extra count - pin the dentry in core */ 53034a9dd7eSJoel Becker 53134a9dd7eSJoel Becker status = 0; 53234a9dd7eSJoel Becker bail: 53334a9dd7eSJoel Becker if (status < 0) 53434a9dd7eSJoel Becker iput(inode); 53534a9dd7eSJoel Becker return status; 53634a9dd7eSJoel Becker } 53734a9dd7eSJoel Becker 53834a9dd7eSJoel Becker static int dlmfs_create(struct inode *dir, 53934a9dd7eSJoel Becker struct dentry *dentry, 54034a9dd7eSJoel Becker int mode, 54134a9dd7eSJoel Becker struct nameidata *nd) 54234a9dd7eSJoel Becker { 54334a9dd7eSJoel Becker int status = 0; 54434a9dd7eSJoel Becker struct inode *inode; 54534a9dd7eSJoel Becker struct qstr *name = &dentry->d_name; 54634a9dd7eSJoel Becker 54734a9dd7eSJoel Becker mlog(0, "create %.*s\n", name->len, name->name); 54834a9dd7eSJoel Becker 54934a9dd7eSJoel Becker /* verify name is valid and doesn't contain any dlm reserved 55034a9dd7eSJoel Becker * characters */ 55134a9dd7eSJoel Becker if (name->len >= USER_DLM_LOCK_ID_MAX_LEN || 55234a9dd7eSJoel Becker name->name[0] == '$') { 55334a9dd7eSJoel Becker status = -EINVAL; 55434a9dd7eSJoel Becker mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len, 55534a9dd7eSJoel Becker name->name); 55634a9dd7eSJoel Becker goto bail; 55734a9dd7eSJoel Becker } 55834a9dd7eSJoel Becker 55934a9dd7eSJoel Becker inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG); 56034a9dd7eSJoel Becker if (!inode) { 56134a9dd7eSJoel Becker status = -ENOMEM; 56234a9dd7eSJoel Becker mlog_errno(status); 56334a9dd7eSJoel Becker goto bail; 56434a9dd7eSJoel Becker } 56534a9dd7eSJoel Becker 56634a9dd7eSJoel Becker d_instantiate(dentry, inode); 56734a9dd7eSJoel Becker dget(dentry); /* Extra count - pin the dentry in core */ 56834a9dd7eSJoel Becker bail: 56934a9dd7eSJoel Becker return status; 57034a9dd7eSJoel Becker } 57134a9dd7eSJoel Becker 57234a9dd7eSJoel Becker static int dlmfs_unlink(struct inode *dir, 57334a9dd7eSJoel Becker struct dentry *dentry) 57434a9dd7eSJoel Becker { 57534a9dd7eSJoel Becker int status; 57634a9dd7eSJoel Becker struct inode *inode = dentry->d_inode; 57734a9dd7eSJoel Becker 57834a9dd7eSJoel Becker mlog(0, "unlink inode %lu\n", inode->i_ino); 57934a9dd7eSJoel Becker 58034a9dd7eSJoel Becker /* if there are no current holders, or none that are waiting 58134a9dd7eSJoel Becker * to acquire a lock, this basically destroys our lockres. */ 58234a9dd7eSJoel Becker status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres); 58334a9dd7eSJoel Becker if (status < 0) { 58434a9dd7eSJoel Becker mlog(ML_ERROR, "unlink %.*s, error %d from destroy\n", 58534a9dd7eSJoel Becker dentry->d_name.len, dentry->d_name.name, status); 58634a9dd7eSJoel Becker goto bail; 58734a9dd7eSJoel Becker } 58834a9dd7eSJoel Becker status = simple_unlink(dir, dentry); 58934a9dd7eSJoel Becker bail: 59034a9dd7eSJoel Becker return status; 59134a9dd7eSJoel Becker } 59234a9dd7eSJoel Becker 59334a9dd7eSJoel Becker static int dlmfs_fill_super(struct super_block * sb, 59434a9dd7eSJoel Becker void * data, 59534a9dd7eSJoel Becker int silent) 59634a9dd7eSJoel Becker { 59734a9dd7eSJoel Becker struct inode * inode; 59834a9dd7eSJoel Becker struct dentry * root; 59934a9dd7eSJoel Becker 60034a9dd7eSJoel Becker sb->s_maxbytes = MAX_LFS_FILESIZE; 60134a9dd7eSJoel Becker sb->s_blocksize = PAGE_CACHE_SIZE; 60234a9dd7eSJoel Becker sb->s_blocksize_bits = PAGE_CACHE_SHIFT; 60334a9dd7eSJoel Becker sb->s_magic = DLMFS_MAGIC; 60434a9dd7eSJoel Becker sb->s_op = &dlmfs_ops; 60534a9dd7eSJoel Becker inode = dlmfs_get_root_inode(sb); 60634a9dd7eSJoel Becker if (!inode) 60734a9dd7eSJoel Becker return -ENOMEM; 60834a9dd7eSJoel Becker 60934a9dd7eSJoel Becker root = d_alloc_root(inode); 61034a9dd7eSJoel Becker if (!root) { 61134a9dd7eSJoel Becker iput(inode); 61234a9dd7eSJoel Becker return -ENOMEM; 61334a9dd7eSJoel Becker } 61434a9dd7eSJoel Becker sb->s_root = root; 61534a9dd7eSJoel Becker return 0; 61634a9dd7eSJoel Becker } 61734a9dd7eSJoel Becker 61834a9dd7eSJoel Becker static const struct file_operations dlmfs_file_operations = { 61934a9dd7eSJoel Becker .open = dlmfs_file_open, 62034a9dd7eSJoel Becker .release = dlmfs_file_release, 62134a9dd7eSJoel Becker .poll = dlmfs_file_poll, 62234a9dd7eSJoel Becker .read = dlmfs_file_read, 62334a9dd7eSJoel Becker .write = dlmfs_file_write, 6246038f373SArnd Bergmann .llseek = default_llseek, 62534a9dd7eSJoel Becker }; 62634a9dd7eSJoel Becker 62734a9dd7eSJoel Becker static const struct inode_operations dlmfs_dir_inode_operations = { 62834a9dd7eSJoel Becker .create = dlmfs_create, 62934a9dd7eSJoel Becker .lookup = simple_lookup, 63034a9dd7eSJoel Becker .unlink = dlmfs_unlink, 63134a9dd7eSJoel Becker }; 63234a9dd7eSJoel Becker 63334a9dd7eSJoel Becker /* this way we can restrict mkdir to only the toplevel of the fs. */ 63434a9dd7eSJoel Becker static const struct inode_operations dlmfs_root_inode_operations = { 63534a9dd7eSJoel Becker .lookup = simple_lookup, 63634a9dd7eSJoel Becker .mkdir = dlmfs_mkdir, 63734a9dd7eSJoel Becker .rmdir = simple_rmdir, 63834a9dd7eSJoel Becker }; 63934a9dd7eSJoel Becker 64034a9dd7eSJoel Becker static const struct super_operations dlmfs_ops = { 64134a9dd7eSJoel Becker .statfs = simple_statfs, 64234a9dd7eSJoel Becker .alloc_inode = dlmfs_alloc_inode, 64334a9dd7eSJoel Becker .destroy_inode = dlmfs_destroy_inode, 644b57922d9SAl Viro .evict_inode = dlmfs_evict_inode, 64534a9dd7eSJoel Becker .drop_inode = generic_delete_inode, 64634a9dd7eSJoel Becker }; 64734a9dd7eSJoel Becker 64834a9dd7eSJoel Becker static const struct inode_operations dlmfs_file_inode_operations = { 64934a9dd7eSJoel Becker .getattr = simple_getattr, 650e8fce482SJoel Becker .setattr = dlmfs_file_setattr, 65134a9dd7eSJoel Becker }; 65234a9dd7eSJoel Becker 6533c26ff6eSAl Viro static struct dentry *dlmfs_mount(struct file_system_type *fs_type, 6543c26ff6eSAl Viro int flags, const char *dev_name, void *data) 65534a9dd7eSJoel Becker { 6563c26ff6eSAl Viro return mount_nodev(fs_type, flags, data, dlmfs_fill_super); 65734a9dd7eSJoel Becker } 65834a9dd7eSJoel Becker 65934a9dd7eSJoel Becker static struct file_system_type dlmfs_fs_type = { 66034a9dd7eSJoel Becker .owner = THIS_MODULE, 66134a9dd7eSJoel Becker .name = "ocfs2_dlmfs", 6623c26ff6eSAl Viro .mount = dlmfs_mount, 66334a9dd7eSJoel Becker .kill_sb = kill_litter_super, 66434a9dd7eSJoel Becker }; 66534a9dd7eSJoel Becker 66634a9dd7eSJoel Becker static int __init init_dlmfs_fs(void) 66734a9dd7eSJoel Becker { 66834a9dd7eSJoel Becker int status; 66934a9dd7eSJoel Becker int cleanup_inode = 0, cleanup_worker = 0; 67034a9dd7eSJoel Becker 67134a9dd7eSJoel Becker dlmfs_print_version(); 67234a9dd7eSJoel Becker 67334a9dd7eSJoel Becker status = bdi_init(&dlmfs_backing_dev_info); 67434a9dd7eSJoel Becker if (status) 67534a9dd7eSJoel Becker return status; 67634a9dd7eSJoel Becker 67734a9dd7eSJoel Becker dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache", 67834a9dd7eSJoel Becker sizeof(struct dlmfs_inode_private), 67934a9dd7eSJoel Becker 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| 68034a9dd7eSJoel Becker SLAB_MEM_SPREAD), 68134a9dd7eSJoel Becker dlmfs_init_once); 68234a9dd7eSJoel Becker if (!dlmfs_inode_cache) { 68334a9dd7eSJoel Becker status = -ENOMEM; 68434a9dd7eSJoel Becker goto bail; 68534a9dd7eSJoel Becker } 68634a9dd7eSJoel Becker cleanup_inode = 1; 68734a9dd7eSJoel Becker 68834a9dd7eSJoel Becker user_dlm_worker = create_singlethread_workqueue("user_dlm"); 68934a9dd7eSJoel Becker if (!user_dlm_worker) { 69034a9dd7eSJoel Becker status = -ENOMEM; 69134a9dd7eSJoel Becker goto bail; 69234a9dd7eSJoel Becker } 69334a9dd7eSJoel Becker cleanup_worker = 1; 69434a9dd7eSJoel Becker 6950016eedcSJoel Becker user_dlm_set_locking_protocol(); 69634a9dd7eSJoel Becker status = register_filesystem(&dlmfs_fs_type); 69734a9dd7eSJoel Becker bail: 69834a9dd7eSJoel Becker if (status) { 69934a9dd7eSJoel Becker if (cleanup_inode) 70034a9dd7eSJoel Becker kmem_cache_destroy(dlmfs_inode_cache); 70134a9dd7eSJoel Becker if (cleanup_worker) 70234a9dd7eSJoel Becker destroy_workqueue(user_dlm_worker); 70334a9dd7eSJoel Becker bdi_destroy(&dlmfs_backing_dev_info); 70434a9dd7eSJoel Becker } else 70534a9dd7eSJoel Becker printk("OCFS2 User DLM kernel interface loaded\n"); 70634a9dd7eSJoel Becker return status; 70734a9dd7eSJoel Becker } 70834a9dd7eSJoel Becker 70934a9dd7eSJoel Becker static void __exit exit_dlmfs_fs(void) 71034a9dd7eSJoel Becker { 71134a9dd7eSJoel Becker unregister_filesystem(&dlmfs_fs_type); 71234a9dd7eSJoel Becker 71334a9dd7eSJoel Becker flush_workqueue(user_dlm_worker); 71434a9dd7eSJoel Becker destroy_workqueue(user_dlm_worker); 71534a9dd7eSJoel Becker 71634a9dd7eSJoel Becker kmem_cache_destroy(dlmfs_inode_cache); 71734a9dd7eSJoel Becker 71834a9dd7eSJoel Becker bdi_destroy(&dlmfs_backing_dev_info); 71934a9dd7eSJoel Becker } 72034a9dd7eSJoel Becker 72134a9dd7eSJoel Becker MODULE_AUTHOR("Oracle"); 72234a9dd7eSJoel Becker MODULE_LICENSE("GPL"); 72334a9dd7eSJoel Becker 72434a9dd7eSJoel Becker module_init(init_dlmfs_fs) 72534a9dd7eSJoel Becker module_exit(exit_dlmfs_fs) 726