1328970deSThomas Gleixner // SPDX-License-Identifier: GPL-2.0-or-later 2fa60ce2cSMasahiro Yamada /* 334a9dd7eSJoel Becker * dlmfs.c 434a9dd7eSJoel Becker * 534a9dd7eSJoel Becker * Code which implements the kernel side of a minimal userspace 634a9dd7eSJoel Becker * interface to our DLM. This file handles the virtual file system 734a9dd7eSJoel Becker * used for communication with userspace. Credit should go to ramfs, 834a9dd7eSJoel Becker * which was a template for the fs side of this module. 934a9dd7eSJoel Becker * 1034a9dd7eSJoel Becker * Copyright (C) 2003, 2004 Oracle. All rights reserved. 1134a9dd7eSJoel Becker */ 1234a9dd7eSJoel Becker 1334a9dd7eSJoel Becker /* Simple VFS hooks based on: */ 1434a9dd7eSJoel Becker /* 1534a9dd7eSJoel Becker * Resizable simple ram filesystem for Linux. 1634a9dd7eSJoel Becker * 1734a9dd7eSJoel Becker * Copyright (C) 2000 Linus Torvalds. 1834a9dd7eSJoel Becker * 2000 Transmeta Corp. 1934a9dd7eSJoel Becker */ 2034a9dd7eSJoel Becker 2134a9dd7eSJoel Becker #include <linux/module.h> 2234a9dd7eSJoel Becker #include <linux/fs.h> 2334a9dd7eSJoel Becker #include <linux/pagemap.h> 2434a9dd7eSJoel Becker #include <linux/types.h> 2534a9dd7eSJoel Becker #include <linux/slab.h> 2634a9dd7eSJoel Becker #include <linux/highmem.h> 2734a9dd7eSJoel Becker #include <linux/init.h> 2834a9dd7eSJoel Becker #include <linux/string.h> 2934a9dd7eSJoel Becker #include <linux/backing-dev.h> 3034a9dd7eSJoel Becker #include <linux/poll.h> 3134a9dd7eSJoel Becker 327c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 3334a9dd7eSJoel Becker 34ca322fb6SMasahiro Yamada #include "../stackglue.h" 3534a9dd7eSJoel Becker #include "userdlm.h" 3634a9dd7eSJoel Becker 3734a9dd7eSJoel Becker #define MLOG_MASK_PREFIX ML_DLMFS 38ca322fb6SMasahiro Yamada #include "../cluster/masklog.h" 3934a9dd7eSJoel Becker 4034a9dd7eSJoel Becker 4134a9dd7eSJoel Becker static const struct super_operations dlmfs_ops; 4234a9dd7eSJoel Becker static const struct file_operations dlmfs_file_operations; 4334a9dd7eSJoel Becker static const struct inode_operations dlmfs_dir_inode_operations; 4434a9dd7eSJoel Becker static const struct inode_operations dlmfs_root_inode_operations; 4534a9dd7eSJoel Becker static const struct inode_operations dlmfs_file_inode_operations; 4634a9dd7eSJoel Becker static struct kmem_cache *dlmfs_inode_cache; 4734a9dd7eSJoel Becker 4834a9dd7eSJoel Becker struct workqueue_struct *user_dlm_worker; 4934a9dd7eSJoel Becker 5034a9dd7eSJoel Becker 5134a9dd7eSJoel Becker 5234a9dd7eSJoel Becker /* 5334a9dd7eSJoel Becker * These are the ABI capabilities of dlmfs. 5434a9dd7eSJoel Becker * 5534a9dd7eSJoel Becker * Over time, dlmfs has added some features that were not part of the 5634a9dd7eSJoel Becker * initial ABI. Unfortunately, some of these features are not detectable 5734a9dd7eSJoel Becker * via standard usage. For example, Linux's default poll always returns 58a9a08845SLinus Torvalds * EPOLLIN, so there is no way for a caller of poll(2) to know when dlmfs 5934a9dd7eSJoel Becker * added poll support. Instead, we provide this list of new capabilities. 6034a9dd7eSJoel Becker * 6134a9dd7eSJoel Becker * Capabilities is a read-only attribute. We do it as a module parameter 6234a9dd7eSJoel Becker * so we can discover it whether dlmfs is built in, loaded, or even not 6334a9dd7eSJoel Becker * loaded. 6434a9dd7eSJoel Becker * 6534a9dd7eSJoel Becker * The ABI features are local to this machine's dlmfs mount. This is 6634a9dd7eSJoel Becker * distinct from the locking protocol, which is concerned with inter-node 6734a9dd7eSJoel Becker * interaction. 6834a9dd7eSJoel Becker * 6934a9dd7eSJoel Becker * Capabilities: 70a9a08845SLinus Torvalds * - bast : EPOLLIN against the file descriptor of a held lock 7134a9dd7eSJoel Becker * signifies a bast fired on the lock. 7234a9dd7eSJoel Becker */ 73cbe0e331SJoel Becker #define DLMFS_CAPABILITIES "bast stackglue" 7498ba073cSRobin Dong static int param_set_dlmfs_capabilities(const char *val, 75e4dca7b7SKees Cook const struct kernel_param *kp) 7634a9dd7eSJoel Becker { 7734a9dd7eSJoel Becker printk(KERN_ERR "%s: readonly parameter\n", kp->name); 7834a9dd7eSJoel Becker return -EINVAL; 7934a9dd7eSJoel Becker } 8034a9dd7eSJoel Becker static int param_get_dlmfs_capabilities(char *buffer, 81e4dca7b7SKees Cook const struct kernel_param *kp) 8234a9dd7eSJoel Becker { 8334a9dd7eSJoel Becker return strlcpy(buffer, DLMFS_CAPABILITIES, 8434a9dd7eSJoel Becker strlen(DLMFS_CAPABILITIES) + 1); 8534a9dd7eSJoel Becker } 8634a9dd7eSJoel Becker module_param_call(capabilities, param_set_dlmfs_capabilities, 8734a9dd7eSJoel Becker param_get_dlmfs_capabilities, NULL, 0444); 8834a9dd7eSJoel Becker MODULE_PARM_DESC(capabilities, DLMFS_CAPABILITIES); 8934a9dd7eSJoel Becker 9034a9dd7eSJoel Becker 9134a9dd7eSJoel Becker /* 9234a9dd7eSJoel Becker * decodes a set of open flags into a valid lock level and a set of flags. 9334a9dd7eSJoel Becker * returns < 0 if we have invalid flags 9434a9dd7eSJoel Becker * flags which mean something to us: 9534a9dd7eSJoel Becker * O_RDONLY -> PRMODE level 9634a9dd7eSJoel Becker * O_WRONLY -> EXMODE level 9734a9dd7eSJoel Becker * 98efd647f7STao Ma * O_NONBLOCK -> NOQUEUE 9934a9dd7eSJoel Becker */ 10034a9dd7eSJoel Becker static int dlmfs_decode_open_flags(int open_flags, 10134a9dd7eSJoel Becker int *level, 10234a9dd7eSJoel Becker int *flags) 10334a9dd7eSJoel Becker { 10434a9dd7eSJoel Becker if (open_flags & (O_WRONLY|O_RDWR)) 105efd647f7STao Ma *level = DLM_LOCK_EX; 10634a9dd7eSJoel Becker else 107efd647f7STao Ma *level = DLM_LOCK_PR; 10834a9dd7eSJoel Becker 10934a9dd7eSJoel Becker *flags = 0; 11034a9dd7eSJoel Becker if (open_flags & O_NONBLOCK) 111efd647f7STao Ma *flags |= DLM_LKF_NOQUEUE; 11234a9dd7eSJoel Becker 11334a9dd7eSJoel Becker return 0; 11434a9dd7eSJoel Becker } 11534a9dd7eSJoel Becker 11634a9dd7eSJoel Becker static int dlmfs_file_open(struct inode *inode, 11734a9dd7eSJoel Becker struct file *file) 11834a9dd7eSJoel Becker { 11934a9dd7eSJoel Becker int status, level, flags; 12034a9dd7eSJoel Becker struct dlmfs_filp_private *fp = NULL; 12134a9dd7eSJoel Becker struct dlmfs_inode_private *ip; 12234a9dd7eSJoel Becker 12334a9dd7eSJoel Becker if (S_ISDIR(inode->i_mode)) 12434a9dd7eSJoel Becker BUG(); 12534a9dd7eSJoel Becker 12634a9dd7eSJoel Becker mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino, 12734a9dd7eSJoel Becker file->f_flags); 12834a9dd7eSJoel Becker 12934a9dd7eSJoel Becker status = dlmfs_decode_open_flags(file->f_flags, &level, &flags); 13034a9dd7eSJoel Becker if (status < 0) 13134a9dd7eSJoel Becker goto bail; 13234a9dd7eSJoel Becker 13334a9dd7eSJoel Becker /* We don't want to honor O_APPEND at read/write time as it 13434a9dd7eSJoel Becker * doesn't make sense for LVB writes. */ 13534a9dd7eSJoel Becker file->f_flags &= ~O_APPEND; 13634a9dd7eSJoel Becker 13734a9dd7eSJoel Becker fp = kmalloc(sizeof(*fp), GFP_NOFS); 13834a9dd7eSJoel Becker if (!fp) { 13934a9dd7eSJoel Becker status = -ENOMEM; 14034a9dd7eSJoel Becker goto bail; 14134a9dd7eSJoel Becker } 14234a9dd7eSJoel Becker fp->fp_lock_level = level; 14334a9dd7eSJoel Becker 14434a9dd7eSJoel Becker ip = DLMFS_I(inode); 14534a9dd7eSJoel Becker 14634a9dd7eSJoel Becker status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags); 14734a9dd7eSJoel Becker if (status < 0) { 14834a9dd7eSJoel Becker /* this is a strange error to return here but I want 14934a9dd7eSJoel Becker * to be able userspace to be able to distinguish a 15034a9dd7eSJoel Becker * valid lock request from one that simply couldn't be 15134a9dd7eSJoel Becker * granted. */ 152efd647f7STao Ma if (flags & DLM_LKF_NOQUEUE && status == -EAGAIN) 15334a9dd7eSJoel Becker status = -ETXTBSY; 15434a9dd7eSJoel Becker kfree(fp); 15534a9dd7eSJoel Becker goto bail; 15634a9dd7eSJoel Becker } 15734a9dd7eSJoel Becker 15834a9dd7eSJoel Becker file->private_data = fp; 15934a9dd7eSJoel Becker bail: 16034a9dd7eSJoel Becker return status; 16134a9dd7eSJoel Becker } 16234a9dd7eSJoel Becker 16334a9dd7eSJoel Becker static int dlmfs_file_release(struct inode *inode, 16434a9dd7eSJoel Becker struct file *file) 16534a9dd7eSJoel Becker { 166cb6a8fd7Szhong jiang int level; 16734a9dd7eSJoel Becker struct dlmfs_inode_private *ip = DLMFS_I(inode); 16833fa1d90SJoe Perches struct dlmfs_filp_private *fp = file->private_data; 16934a9dd7eSJoel Becker 17034a9dd7eSJoel Becker if (S_ISDIR(inode->i_mode)) 17134a9dd7eSJoel Becker BUG(); 17234a9dd7eSJoel Becker 17334a9dd7eSJoel Becker mlog(0, "close called on inode %lu\n", inode->i_ino); 17434a9dd7eSJoel Becker 17534a9dd7eSJoel Becker if (fp) { 17634a9dd7eSJoel Becker level = fp->fp_lock_level; 177efd647f7STao Ma if (level != DLM_LOCK_IV) 17834a9dd7eSJoel Becker user_dlm_cluster_unlock(&ip->ip_lockres, level); 17934a9dd7eSJoel Becker 18034a9dd7eSJoel Becker kfree(fp); 18134a9dd7eSJoel Becker file->private_data = NULL; 18234a9dd7eSJoel Becker } 18334a9dd7eSJoel Becker 18434a9dd7eSJoel Becker return 0; 18534a9dd7eSJoel Becker } 18634a9dd7eSJoel Becker 187e8fce482SJoel Becker /* 188e8fce482SJoel Becker * We do ->setattr() just to override size changes. Our size is the size 189e8fce482SJoel Becker * of the LVB and nothing else. 190e8fce482SJoel Becker */ 191*c1632a0fSChristian Brauner static int dlmfs_file_setattr(struct mnt_idmap *idmap, 192549c7297SChristian Brauner struct dentry *dentry, struct iattr *attr) 193e8fce482SJoel Becker { 194e8fce482SJoel Becker int error; 1952b0143b5SDavid Howells struct inode *inode = d_inode(dentry); 196e8fce482SJoel Becker 197e8fce482SJoel Becker attr->ia_valid &= ~ATTR_SIZE; 198*c1632a0fSChristian Brauner error = setattr_prepare(&nop_mnt_idmap, dentry, attr); 1991025774cSChristoph Hellwig if (error) 200e8fce482SJoel Becker return error; 2011025774cSChristoph Hellwig 202*c1632a0fSChristian Brauner setattr_copy(&nop_mnt_idmap, inode, attr); 2031025774cSChristoph Hellwig mark_inode_dirty(inode); 2041025774cSChristoph Hellwig return 0; 205e8fce482SJoel Becker } 206e8fce482SJoel Becker 207076ccb76SAl Viro static __poll_t dlmfs_file_poll(struct file *file, poll_table *wait) 20834a9dd7eSJoel Becker { 209076ccb76SAl Viro __poll_t event = 0; 210496ad9aaSAl Viro struct inode *inode = file_inode(file); 21134a9dd7eSJoel Becker struct dlmfs_inode_private *ip = DLMFS_I(inode); 21234a9dd7eSJoel Becker 21334a9dd7eSJoel Becker poll_wait(file, &ip->ip_lockres.l_event, wait); 21434a9dd7eSJoel Becker 21534a9dd7eSJoel Becker spin_lock(&ip->ip_lockres.l_lock); 21634a9dd7eSJoel Becker if (ip->ip_lockres.l_flags & USER_LOCK_BLOCKED) 217a9a08845SLinus Torvalds event = EPOLLIN | EPOLLRDNORM; 21834a9dd7eSJoel Becker spin_unlock(&ip->ip_lockres.l_lock); 21934a9dd7eSJoel Becker 22034a9dd7eSJoel Becker return event; 22134a9dd7eSJoel Becker } 22234a9dd7eSJoel Becker 223067c054fSAl Viro static ssize_t dlmfs_file_read(struct file *file, 22434a9dd7eSJoel Becker char __user *buf, 22534a9dd7eSJoel Becker size_t count, 22634a9dd7eSJoel Becker loff_t *ppos) 22734a9dd7eSJoel Becker { 228067c054fSAl Viro char lvb[DLM_LVB_LEN]; 22934a9dd7eSJoel Becker 230067c054fSAl Viro if (!user_dlm_read_lvb(file_inode(file), lvb)) 23134a9dd7eSJoel Becker return 0; 23234a9dd7eSJoel Becker 233067c054fSAl Viro return simple_read_from_buffer(buf, count, ppos, lvb, sizeof(lvb)); 23434a9dd7eSJoel Becker } 23534a9dd7eSJoel Becker 23634a9dd7eSJoel Becker static ssize_t dlmfs_file_write(struct file *filp, 23734a9dd7eSJoel Becker const char __user *buf, 23834a9dd7eSJoel Becker size_t count, 23934a9dd7eSJoel Becker loff_t *ppos) 24034a9dd7eSJoel Becker { 241067c054fSAl Viro char lvb_buf[DLM_LVB_LEN]; 24234a9dd7eSJoel Becker int bytes_left; 243496ad9aaSAl Viro struct inode *inode = file_inode(filp); 24434a9dd7eSJoel Becker 24534a9dd7eSJoel Becker mlog(0, "inode %lu, count = %zu, *ppos = %llu\n", 24634a9dd7eSJoel Becker inode->i_ino, count, *ppos); 24734a9dd7eSJoel Becker 248067c054fSAl Viro if (*ppos >= DLM_LVB_LEN) 24934a9dd7eSJoel Becker return -ENOSPC; 25034a9dd7eSJoel Becker 2513815f1beSAl Viro /* don't write past the lvb */ 252067c054fSAl Viro if (count > DLM_LVB_LEN - *ppos) 253067c054fSAl Viro count = DLM_LVB_LEN - *ppos; 2543815f1beSAl Viro 25534a9dd7eSJoel Becker if (!count) 25634a9dd7eSJoel Becker return 0; 25734a9dd7eSJoel Becker 2583815f1beSAl Viro bytes_left = copy_from_user(lvb_buf, buf, count); 2593815f1beSAl Viro count -= bytes_left; 2603815f1beSAl Viro if (count) 2613815f1beSAl Viro user_dlm_write_lvb(inode, lvb_buf, count); 26234a9dd7eSJoel Becker 2633815f1beSAl Viro *ppos = *ppos + count; 2643815f1beSAl Viro mlog(0, "wrote %zu bytes\n", count); 2653815f1beSAl Viro return count; 26634a9dd7eSJoel Becker } 26734a9dd7eSJoel Becker 26834a9dd7eSJoel Becker static void dlmfs_init_once(void *foo) 26934a9dd7eSJoel Becker { 27034a9dd7eSJoel Becker struct dlmfs_inode_private *ip = 27134a9dd7eSJoel Becker (struct dlmfs_inode_private *) foo; 27234a9dd7eSJoel Becker 2730016eedcSJoel Becker ip->ip_conn = NULL; 27434a9dd7eSJoel Becker ip->ip_parent = NULL; 27534a9dd7eSJoel Becker 27634a9dd7eSJoel Becker inode_init_once(&ip->ip_vfs_inode); 27734a9dd7eSJoel Becker } 27834a9dd7eSJoel Becker 27934a9dd7eSJoel Becker static struct inode *dlmfs_alloc_inode(struct super_block *sb) 28034a9dd7eSJoel Becker { 28134a9dd7eSJoel Becker struct dlmfs_inode_private *ip; 28234a9dd7eSJoel Becker 283fd60b288SMuchun Song ip = alloc_inode_sb(sb, dlmfs_inode_cache, GFP_NOFS); 28434a9dd7eSJoel Becker if (!ip) 28534a9dd7eSJoel Becker return NULL; 28634a9dd7eSJoel Becker 28734a9dd7eSJoel Becker return &ip->ip_vfs_inode; 28834a9dd7eSJoel Becker } 28934a9dd7eSJoel Becker 2909fbc0007SAl Viro static void dlmfs_free_inode(struct inode *inode) 291fa0d7e3dSNick Piggin { 292fa0d7e3dSNick Piggin kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode)); 293fa0d7e3dSNick Piggin } 294fa0d7e3dSNick Piggin 295b57922d9SAl Viro static void dlmfs_evict_inode(struct inode *inode) 29634a9dd7eSJoel Becker { 29734a9dd7eSJoel Becker int status; 29834a9dd7eSJoel Becker struct dlmfs_inode_private *ip; 29953fd5ffbSJunxiao Bi struct user_lock_res *lockres; 30053fd5ffbSJunxiao Bi int teardown; 30134a9dd7eSJoel Becker 302dbd5768fSJan Kara clear_inode(inode); 30334a9dd7eSJoel Becker 30434a9dd7eSJoel Becker mlog(0, "inode %lu\n", inode->i_ino); 30534a9dd7eSJoel Becker 30634a9dd7eSJoel Becker ip = DLMFS_I(inode); 30753fd5ffbSJunxiao Bi lockres = &ip->ip_lockres; 30834a9dd7eSJoel Becker 30934a9dd7eSJoel Becker if (S_ISREG(inode->i_mode)) { 31053fd5ffbSJunxiao Bi spin_lock(&lockres->l_lock); 31153fd5ffbSJunxiao Bi teardown = !!(lockres->l_flags & USER_LOCK_IN_TEARDOWN); 31253fd5ffbSJunxiao Bi spin_unlock(&lockres->l_lock); 31353fd5ffbSJunxiao Bi if (!teardown) { 31453fd5ffbSJunxiao Bi status = user_dlm_destroy_lock(lockres); 31534a9dd7eSJoel Becker if (status < 0) 31634a9dd7eSJoel Becker mlog_errno(status); 31753fd5ffbSJunxiao Bi } 31834a9dd7eSJoel Becker iput(ip->ip_parent); 31934a9dd7eSJoel Becker goto clear_fields; 32034a9dd7eSJoel Becker } 32134a9dd7eSJoel Becker 3220016eedcSJoel Becker mlog(0, "we're a directory, ip->ip_conn = 0x%p\n", ip->ip_conn); 32334a9dd7eSJoel Becker /* we must be a directory. If required, lets unregister the 32434a9dd7eSJoel Becker * dlm context now. */ 3250016eedcSJoel Becker if (ip->ip_conn) 3260016eedcSJoel Becker user_dlm_unregister(ip->ip_conn); 32734a9dd7eSJoel Becker clear_fields: 32834a9dd7eSJoel Becker ip->ip_parent = NULL; 3290016eedcSJoel Becker ip->ip_conn = NULL; 33034a9dd7eSJoel Becker } 33134a9dd7eSJoel Becker 33234a9dd7eSJoel Becker static struct inode *dlmfs_get_root_inode(struct super_block *sb) 33334a9dd7eSJoel Becker { 33434a9dd7eSJoel Becker struct inode *inode = new_inode(sb); 33567697cbdSAl Viro umode_t mode = S_IFDIR | 0755; 33634a9dd7eSJoel Becker 33734a9dd7eSJoel Becker if (inode) { 33885fe4025SChristoph Hellwig inode->i_ino = get_next_ino(); 33921cb47beSChristian Brauner inode_init_owner(&init_user_ns, inode, NULL, mode); 340078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 34134a9dd7eSJoel Becker inc_nlink(inode); 34234a9dd7eSJoel Becker 34334a9dd7eSJoel Becker inode->i_fop = &simple_dir_operations; 34434a9dd7eSJoel Becker inode->i_op = &dlmfs_root_inode_operations; 34534a9dd7eSJoel Becker } 34634a9dd7eSJoel Becker 34734a9dd7eSJoel Becker return inode; 34834a9dd7eSJoel Becker } 34934a9dd7eSJoel Becker 35034a9dd7eSJoel Becker static struct inode *dlmfs_get_inode(struct inode *parent, 35134a9dd7eSJoel Becker struct dentry *dentry, 35267697cbdSAl Viro umode_t mode) 35334a9dd7eSJoel Becker { 35434a9dd7eSJoel Becker struct super_block *sb = parent->i_sb; 35534a9dd7eSJoel Becker struct inode * inode = new_inode(sb); 35634a9dd7eSJoel Becker struct dlmfs_inode_private *ip; 35734a9dd7eSJoel Becker 35834a9dd7eSJoel Becker if (!inode) 35934a9dd7eSJoel Becker return NULL; 36034a9dd7eSJoel Becker 36185fe4025SChristoph Hellwig inode->i_ino = get_next_ino(); 36221cb47beSChristian Brauner inode_init_owner(&init_user_ns, inode, parent, mode); 363078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 36434a9dd7eSJoel Becker 36534a9dd7eSJoel Becker ip = DLMFS_I(inode); 3660016eedcSJoel Becker ip->ip_conn = DLMFS_I(parent)->ip_conn; 36734a9dd7eSJoel Becker 36834a9dd7eSJoel Becker switch (mode & S_IFMT) { 36934a9dd7eSJoel Becker default: 37034a9dd7eSJoel Becker /* for now we don't support anything other than 37134a9dd7eSJoel Becker * directories and regular files. */ 37234a9dd7eSJoel Becker BUG(); 37334a9dd7eSJoel Becker break; 37434a9dd7eSJoel Becker case S_IFREG: 37534a9dd7eSJoel Becker inode->i_op = &dlmfs_file_inode_operations; 37634a9dd7eSJoel Becker inode->i_fop = &dlmfs_file_operations; 37734a9dd7eSJoel Becker 37834a9dd7eSJoel Becker i_size_write(inode, DLM_LVB_LEN); 37934a9dd7eSJoel Becker 38034a9dd7eSJoel Becker user_dlm_lock_res_init(&ip->ip_lockres, dentry); 38134a9dd7eSJoel Becker 38234a9dd7eSJoel Becker /* released at clear_inode time, this insures that we 38334a9dd7eSJoel Becker * get to drop the dlm reference on each lock *before* 38434a9dd7eSJoel Becker * we call the unregister code for releasing parent 38534a9dd7eSJoel Becker * directories. */ 38634a9dd7eSJoel Becker ip->ip_parent = igrab(parent); 38734a9dd7eSJoel Becker BUG_ON(!ip->ip_parent); 38834a9dd7eSJoel Becker break; 38934a9dd7eSJoel Becker case S_IFDIR: 39034a9dd7eSJoel Becker inode->i_op = &dlmfs_dir_inode_operations; 39134a9dd7eSJoel Becker inode->i_fop = &simple_dir_operations; 39234a9dd7eSJoel Becker 39334a9dd7eSJoel Becker /* directory inodes start off with i_nlink == 39434a9dd7eSJoel Becker * 2 (for "." entry) */ 39534a9dd7eSJoel Becker inc_nlink(inode); 39634a9dd7eSJoel Becker break; 39734a9dd7eSJoel Becker } 39834a9dd7eSJoel Becker return inode; 39934a9dd7eSJoel Becker } 40034a9dd7eSJoel Becker 40134a9dd7eSJoel Becker /* 40234a9dd7eSJoel Becker * File creation. Allocate an inode, and we're done.. 40334a9dd7eSJoel Becker */ 40434a9dd7eSJoel Becker /* SMP-safe */ 405549c7297SChristian Brauner static int dlmfs_mkdir(struct user_namespace * mnt_userns, 406549c7297SChristian Brauner struct inode * dir, 40734a9dd7eSJoel Becker struct dentry * dentry, 40818bb1db3SAl Viro umode_t mode) 40934a9dd7eSJoel Becker { 41034a9dd7eSJoel Becker int status; 41134a9dd7eSJoel Becker struct inode *inode = NULL; 412612645f7SAl Viro const struct qstr *domain = &dentry->d_name; 41334a9dd7eSJoel Becker struct dlmfs_inode_private *ip; 4140016eedcSJoel Becker struct ocfs2_cluster_connection *conn; 41534a9dd7eSJoel Becker 41634a9dd7eSJoel Becker mlog(0, "mkdir %.*s\n", domain->len, domain->name); 41734a9dd7eSJoel Becker 41834a9dd7eSJoel Becker /* verify that we have a proper domain */ 4190016eedcSJoel Becker if (domain->len >= GROUP_NAME_MAX) { 42034a9dd7eSJoel Becker status = -EINVAL; 42134a9dd7eSJoel Becker mlog(ML_ERROR, "invalid domain name for directory.\n"); 42234a9dd7eSJoel Becker goto bail; 42334a9dd7eSJoel Becker } 42434a9dd7eSJoel Becker 42534a9dd7eSJoel Becker inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR); 42634a9dd7eSJoel Becker if (!inode) { 42734a9dd7eSJoel Becker status = -ENOMEM; 42834a9dd7eSJoel Becker mlog_errno(status); 42934a9dd7eSJoel Becker goto bail; 43034a9dd7eSJoel Becker } 43134a9dd7eSJoel Becker 43234a9dd7eSJoel Becker ip = DLMFS_I(inode); 43334a9dd7eSJoel Becker 4340016eedcSJoel Becker conn = user_dlm_register(domain); 4350016eedcSJoel Becker if (IS_ERR(conn)) { 4360016eedcSJoel Becker status = PTR_ERR(conn); 43734a9dd7eSJoel Becker mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n", 43834a9dd7eSJoel Becker status, domain->len, domain->name); 43934a9dd7eSJoel Becker goto bail; 44034a9dd7eSJoel Becker } 4410016eedcSJoel Becker ip->ip_conn = conn; 44234a9dd7eSJoel Becker 44334a9dd7eSJoel Becker inc_nlink(dir); 44434a9dd7eSJoel Becker d_instantiate(dentry, inode); 44534a9dd7eSJoel Becker dget(dentry); /* Extra count - pin the dentry in core */ 44634a9dd7eSJoel Becker 44734a9dd7eSJoel Becker status = 0; 44834a9dd7eSJoel Becker bail: 44934a9dd7eSJoel Becker if (status < 0) 45034a9dd7eSJoel Becker iput(inode); 45134a9dd7eSJoel Becker return status; 45234a9dd7eSJoel Becker } 45334a9dd7eSJoel Becker 454549c7297SChristian Brauner static int dlmfs_create(struct user_namespace *mnt_userns, 455549c7297SChristian Brauner struct inode *dir, 45634a9dd7eSJoel Becker struct dentry *dentry, 4574acdaf27SAl Viro umode_t mode, 458ebfc3b49SAl Viro bool excl) 45934a9dd7eSJoel Becker { 46034a9dd7eSJoel Becker int status = 0; 46134a9dd7eSJoel Becker struct inode *inode; 462612645f7SAl Viro const struct qstr *name = &dentry->d_name; 46334a9dd7eSJoel Becker 46434a9dd7eSJoel Becker mlog(0, "create %.*s\n", name->len, name->name); 46534a9dd7eSJoel Becker 46634a9dd7eSJoel Becker /* verify name is valid and doesn't contain any dlm reserved 46734a9dd7eSJoel Becker * characters */ 46834a9dd7eSJoel Becker if (name->len >= USER_DLM_LOCK_ID_MAX_LEN || 46934a9dd7eSJoel Becker name->name[0] == '$') { 47034a9dd7eSJoel Becker status = -EINVAL; 47134a9dd7eSJoel Becker mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len, 47234a9dd7eSJoel Becker name->name); 47334a9dd7eSJoel Becker goto bail; 47434a9dd7eSJoel Becker } 47534a9dd7eSJoel Becker 47634a9dd7eSJoel Becker inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG); 47734a9dd7eSJoel Becker if (!inode) { 47834a9dd7eSJoel Becker status = -ENOMEM; 47934a9dd7eSJoel Becker mlog_errno(status); 48034a9dd7eSJoel Becker goto bail; 48134a9dd7eSJoel Becker } 48234a9dd7eSJoel Becker 48334a9dd7eSJoel Becker d_instantiate(dentry, inode); 48434a9dd7eSJoel Becker dget(dentry); /* Extra count - pin the dentry in core */ 48534a9dd7eSJoel Becker bail: 48634a9dd7eSJoel Becker return status; 48734a9dd7eSJoel Becker } 48834a9dd7eSJoel Becker 48934a9dd7eSJoel Becker static int dlmfs_unlink(struct inode *dir, 49034a9dd7eSJoel Becker struct dentry *dentry) 49134a9dd7eSJoel Becker { 49234a9dd7eSJoel Becker int status; 4932b0143b5SDavid Howells struct inode *inode = d_inode(dentry); 49434a9dd7eSJoel Becker 49534a9dd7eSJoel Becker mlog(0, "unlink inode %lu\n", inode->i_ino); 49634a9dd7eSJoel Becker 49734a9dd7eSJoel Becker /* if there are no current holders, or none that are waiting 49834a9dd7eSJoel Becker * to acquire a lock, this basically destroys our lockres. */ 49934a9dd7eSJoel Becker status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres); 50034a9dd7eSJoel Becker if (status < 0) { 501a455589fSAl Viro mlog(ML_ERROR, "unlink %pd, error %d from destroy\n", 502a455589fSAl Viro dentry, status); 50334a9dd7eSJoel Becker goto bail; 50434a9dd7eSJoel Becker } 50534a9dd7eSJoel Becker status = simple_unlink(dir, dentry); 50634a9dd7eSJoel Becker bail: 50734a9dd7eSJoel Becker return status; 50834a9dd7eSJoel Becker } 50934a9dd7eSJoel Becker 51034a9dd7eSJoel Becker static int dlmfs_fill_super(struct super_block * sb, 51134a9dd7eSJoel Becker void * data, 51234a9dd7eSJoel Becker int silent) 51334a9dd7eSJoel Becker { 51434a9dd7eSJoel Becker sb->s_maxbytes = MAX_LFS_FILESIZE; 51509cbfeafSKirill A. Shutemov sb->s_blocksize = PAGE_SIZE; 51609cbfeafSKirill A. Shutemov sb->s_blocksize_bits = PAGE_SHIFT; 51734a9dd7eSJoel Becker sb->s_magic = DLMFS_MAGIC; 51834a9dd7eSJoel Becker sb->s_op = &dlmfs_ops; 51948fde701SAl Viro sb->s_root = d_make_root(dlmfs_get_root_inode(sb)); 52048fde701SAl Viro if (!sb->s_root) 52134a9dd7eSJoel Becker return -ENOMEM; 52234a9dd7eSJoel Becker return 0; 52334a9dd7eSJoel Becker } 52434a9dd7eSJoel Becker 52534a9dd7eSJoel Becker static const struct file_operations dlmfs_file_operations = { 52634a9dd7eSJoel Becker .open = dlmfs_file_open, 52734a9dd7eSJoel Becker .release = dlmfs_file_release, 52834a9dd7eSJoel Becker .poll = dlmfs_file_poll, 52934a9dd7eSJoel Becker .read = dlmfs_file_read, 53034a9dd7eSJoel Becker .write = dlmfs_file_write, 5316038f373SArnd Bergmann .llseek = default_llseek, 53234a9dd7eSJoel Becker }; 53334a9dd7eSJoel Becker 53434a9dd7eSJoel Becker static const struct inode_operations dlmfs_dir_inode_operations = { 53534a9dd7eSJoel Becker .create = dlmfs_create, 53634a9dd7eSJoel Becker .lookup = simple_lookup, 53734a9dd7eSJoel Becker .unlink = dlmfs_unlink, 53834a9dd7eSJoel Becker }; 53934a9dd7eSJoel Becker 54034a9dd7eSJoel Becker /* this way we can restrict mkdir to only the toplevel of the fs. */ 54134a9dd7eSJoel Becker static const struct inode_operations dlmfs_root_inode_operations = { 54234a9dd7eSJoel Becker .lookup = simple_lookup, 54334a9dd7eSJoel Becker .mkdir = dlmfs_mkdir, 54434a9dd7eSJoel Becker .rmdir = simple_rmdir, 54534a9dd7eSJoel Becker }; 54634a9dd7eSJoel Becker 54734a9dd7eSJoel Becker static const struct super_operations dlmfs_ops = { 54834a9dd7eSJoel Becker .statfs = simple_statfs, 54934a9dd7eSJoel Becker .alloc_inode = dlmfs_alloc_inode, 5509fbc0007SAl Viro .free_inode = dlmfs_free_inode, 551b57922d9SAl Viro .evict_inode = dlmfs_evict_inode, 55234a9dd7eSJoel Becker .drop_inode = generic_delete_inode, 55334a9dd7eSJoel Becker }; 55434a9dd7eSJoel Becker 55534a9dd7eSJoel Becker static const struct inode_operations dlmfs_file_inode_operations = { 55634a9dd7eSJoel Becker .getattr = simple_getattr, 557e8fce482SJoel Becker .setattr = dlmfs_file_setattr, 55834a9dd7eSJoel Becker }; 55934a9dd7eSJoel Becker 5603c26ff6eSAl Viro static struct dentry *dlmfs_mount(struct file_system_type *fs_type, 5613c26ff6eSAl Viro int flags, const char *dev_name, void *data) 56234a9dd7eSJoel Becker { 5633c26ff6eSAl Viro return mount_nodev(fs_type, flags, data, dlmfs_fill_super); 56434a9dd7eSJoel Becker } 56534a9dd7eSJoel Becker 56634a9dd7eSJoel Becker static struct file_system_type dlmfs_fs_type = { 56734a9dd7eSJoel Becker .owner = THIS_MODULE, 56834a9dd7eSJoel Becker .name = "ocfs2_dlmfs", 5693c26ff6eSAl Viro .mount = dlmfs_mount, 57034a9dd7eSJoel Becker .kill_sb = kill_litter_super, 57134a9dd7eSJoel Becker }; 5727f78e035SEric W. Biederman MODULE_ALIAS_FS("ocfs2_dlmfs"); 57334a9dd7eSJoel Becker 57434a9dd7eSJoel Becker static int __init init_dlmfs_fs(void) 57534a9dd7eSJoel Becker { 57634a9dd7eSJoel Becker int status; 57734a9dd7eSJoel Becker int cleanup_inode = 0, cleanup_worker = 0; 57834a9dd7eSJoel Becker 57934a9dd7eSJoel Becker dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache", 58034a9dd7eSJoel Becker sizeof(struct dlmfs_inode_private), 58134a9dd7eSJoel Becker 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| 5825d097056SVladimir Davydov SLAB_MEM_SPREAD|SLAB_ACCOUNT), 58334a9dd7eSJoel Becker dlmfs_init_once); 58434a9dd7eSJoel Becker if (!dlmfs_inode_cache) { 58534a9dd7eSJoel Becker status = -ENOMEM; 58634a9dd7eSJoel Becker goto bail; 58734a9dd7eSJoel Becker } 58834a9dd7eSJoel Becker cleanup_inode = 1; 58934a9dd7eSJoel Becker 5900b41be07SBhaktipriya Shridhar user_dlm_worker = alloc_workqueue("user_dlm", WQ_MEM_RECLAIM, 0); 59134a9dd7eSJoel Becker if (!user_dlm_worker) { 59234a9dd7eSJoel Becker status = -ENOMEM; 59334a9dd7eSJoel Becker goto bail; 59434a9dd7eSJoel Becker } 59534a9dd7eSJoel Becker cleanup_worker = 1; 59634a9dd7eSJoel Becker 5970016eedcSJoel Becker user_dlm_set_locking_protocol(); 59834a9dd7eSJoel Becker status = register_filesystem(&dlmfs_fs_type); 59934a9dd7eSJoel Becker bail: 60034a9dd7eSJoel Becker if (status) { 60134a9dd7eSJoel Becker if (cleanup_inode) 60234a9dd7eSJoel Becker kmem_cache_destroy(dlmfs_inode_cache); 60334a9dd7eSJoel Becker if (cleanup_worker) 60434a9dd7eSJoel Becker destroy_workqueue(user_dlm_worker); 60534a9dd7eSJoel Becker } else 60634a9dd7eSJoel Becker printk("OCFS2 User DLM kernel interface loaded\n"); 60734a9dd7eSJoel Becker return status; 60834a9dd7eSJoel Becker } 60934a9dd7eSJoel Becker 61034a9dd7eSJoel Becker static void __exit exit_dlmfs_fs(void) 61134a9dd7eSJoel Becker { 61234a9dd7eSJoel Becker unregister_filesystem(&dlmfs_fs_type); 61334a9dd7eSJoel Becker 61434a9dd7eSJoel Becker destroy_workqueue(user_dlm_worker); 61534a9dd7eSJoel Becker 6168c0a8537SKirill A. Shutemov /* 6178c0a8537SKirill A. Shutemov * Make sure all delayed rcu free inodes are flushed before we 6188c0a8537SKirill A. Shutemov * destroy cache. 6198c0a8537SKirill A. Shutemov */ 6208c0a8537SKirill A. Shutemov rcu_barrier(); 62134a9dd7eSJoel Becker kmem_cache_destroy(dlmfs_inode_cache); 62234a9dd7eSJoel Becker 62334a9dd7eSJoel Becker } 62434a9dd7eSJoel Becker 62534a9dd7eSJoel Becker MODULE_AUTHOR("Oracle"); 62634a9dd7eSJoel Becker MODULE_LICENSE("GPL"); 627ff8fb335SGoldwyn Rodrigues MODULE_DESCRIPTION("OCFS2 DLM-Filesystem"); 62834a9dd7eSJoel Becker 62934a9dd7eSJoel Becker module_init(init_dlmfs_fs) 63034a9dd7eSJoel Becker module_exit(exit_dlmfs_fs) 631