Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 1 | /* -*- mode: c; c-basic-offset: 8; -*- |
| 2 | * vim: noexpandtab sw=8 ts=8 sts=0: |
| 3 | * |
| 4 | * dlmfs.c |
| 5 | * |
| 6 | * Code which implements the kernel side of a minimal userspace |
| 7 | * interface to our DLM. This file handles the virtual file system |
| 8 | * used for communication with userspace. Credit should go to ramfs, |
| 9 | * which was a template for the fs side of this module. |
| 10 | * |
| 11 | * Copyright (C) 2003, 2004 Oracle. All rights reserved. |
| 12 | * |
| 13 | * This program is free software; you can redistribute it and/or |
| 14 | * modify it under the terms of the GNU General Public |
| 15 | * License as published by the Free Software Foundation; either |
| 16 | * version 2 of the License, or (at your option) any later version. |
| 17 | * |
| 18 | * This program is distributed in the hope that it will be useful, |
| 19 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 20 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 21 | * General Public License for more details. |
| 22 | * |
| 23 | * You should have received a copy of the GNU General Public |
| 24 | * License along with this program; if not, write to the |
| 25 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, |
| 26 | * Boston, MA 021110-1307, USA. |
| 27 | */ |
| 28 | |
| 29 | /* Simple VFS hooks based on: */ |
| 30 | /* |
| 31 | * Resizable simple ram filesystem for Linux. |
| 32 | * |
| 33 | * Copyright (C) 2000 Linus Torvalds. |
| 34 | * 2000 Transmeta Corp. |
| 35 | */ |
| 36 | |
| 37 | #include <linux/module.h> |
| 38 | #include <linux/fs.h> |
| 39 | #include <linux/pagemap.h> |
| 40 | #include <linux/types.h> |
| 41 | #include <linux/slab.h> |
| 42 | #include <linux/highmem.h> |
| 43 | #include <linux/init.h> |
| 44 | #include <linux/string.h> |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 45 | #include <linux/backing-dev.h> |
Joel Becker | 65b6f34 | 2010-01-26 22:14:20 -0800 | [diff] [blame] | 46 | #include <linux/poll.h> |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 47 | |
| 48 | #include <asm/uaccess.h> |
| 49 | |
Joel Becker | 0016eed | 2010-01-30 04:33:50 -0800 | [diff] [blame] | 50 | #include "stackglue.h" |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 51 | #include "userdlm.h" |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 52 | #include "dlmfsver.h" |
| 53 | |
| 54 | #define MLOG_MASK_PREFIX ML_DLMFS |
| 55 | #include "cluster/masklog.h" |
| 56 | |
Joel Becker | d24fbcd | 2008-01-25 17:02:21 -0800 | [diff] [blame] | 57 | |
Josef 'Jeff' Sipek | ee9b6d6 | 2007-02-12 00:55:41 -0800 | [diff] [blame] | 58 | static const struct super_operations dlmfs_ops; |
Arjan van de Ven | 00977a5 | 2007-02-12 00:55:34 -0800 | [diff] [blame] | 59 | static const struct file_operations dlmfs_file_operations; |
Arjan van de Ven | 92e1d5b | 2007-02-12 00:55:39 -0800 | [diff] [blame] | 60 | static const struct inode_operations dlmfs_dir_inode_operations; |
| 61 | static const struct inode_operations dlmfs_root_inode_operations; |
| 62 | static const struct inode_operations dlmfs_file_inode_operations; |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 63 | static struct kmem_cache *dlmfs_inode_cache; |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 64 | |
| 65 | struct workqueue_struct *user_dlm_worker; |
| 66 | |
Joel Becker | d24fbcd | 2008-01-25 17:02:21 -0800 | [diff] [blame] | 67 | |
Joel Becker | 14a437c | 2010-02-04 15:37:27 -0800 | [diff] [blame] | 68 | |
| 69 | /* |
| 70 | * These are the ABI capabilities of dlmfs. |
| 71 | * |
| 72 | * Over time, dlmfs has added some features that were not part of the |
| 73 | * initial ABI. Unfortunately, some of these features are not detectable |
| 74 | * via standard usage. For example, Linux's default poll always returns |
| 75 | * POLLIN, so there is no way for a caller of poll(2) to know when dlmfs |
| 76 | * added poll support. Instead, we provide this list of new capabilities. |
| 77 | * |
| 78 | * Capabilities is a read-only attribute. We do it as a module parameter |
| 79 | * so we can discover it whether dlmfs is built in, loaded, or even not |
| 80 | * loaded. |
| 81 | * |
| 82 | * The ABI features are local to this machine's dlmfs mount. This is |
| 83 | * distinct from the locking protocol, which is concerned with inter-node |
| 84 | * interaction. |
Joel Becker | 65b6f34 | 2010-01-26 22:14:20 -0800 | [diff] [blame] | 85 | * |
| 86 | * Capabilities: |
| 87 | * - bast : POLLIN against the file descriptor of a held lock |
| 88 | * signifies a bast fired on the lock. |
Joel Becker | 14a437c | 2010-02-04 15:37:27 -0800 | [diff] [blame] | 89 | */ |
Joel Becker | cbe0e33 | 2010-01-30 06:02:10 -0800 | [diff] [blame] | 90 | #define DLMFS_CAPABILITIES "bast stackglue" |
Joel Becker | 14a437c | 2010-02-04 15:37:27 -0800 | [diff] [blame] | 91 | extern int param_set_dlmfs_capabilities(const char *val, |
| 92 | struct kernel_param *kp) |
| 93 | { |
| 94 | printk(KERN_ERR "%s: readonly parameter\n", kp->name); |
| 95 | return -EINVAL; |
| 96 | } |
| 97 | static int param_get_dlmfs_capabilities(char *buffer, |
| 98 | struct kernel_param *kp) |
| 99 | { |
| 100 | return strlcpy(buffer, DLMFS_CAPABILITIES, |
| 101 | strlen(DLMFS_CAPABILITIES) + 1); |
| 102 | } |
| 103 | module_param_call(capabilities, param_set_dlmfs_capabilities, |
| 104 | param_get_dlmfs_capabilities, NULL, 0444); |
| 105 | MODULE_PARM_DESC(capabilities, DLMFS_CAPABILITIES); |
| 106 | |
| 107 | |
Joel Becker | d24fbcd | 2008-01-25 17:02:21 -0800 | [diff] [blame] | 108 | /* |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 109 | * decodes a set of open flags into a valid lock level and a set of flags. |
| 110 | * returns < 0 if we have invalid flags |
| 111 | * flags which mean something to us: |
| 112 | * O_RDONLY -> PRMODE level |
| 113 | * O_WRONLY -> EXMODE level |
| 114 | * |
Tao Ma | efd647f | 2010-03-30 13:21:31 +0800 | [diff] [blame] | 115 | * O_NONBLOCK -> NOQUEUE |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 116 | */ |
| 117 | static int dlmfs_decode_open_flags(int open_flags, |
| 118 | int *level, |
| 119 | int *flags) |
| 120 | { |
| 121 | if (open_flags & (O_WRONLY|O_RDWR)) |
Tao Ma | efd647f | 2010-03-30 13:21:31 +0800 | [diff] [blame] | 122 | *level = DLM_LOCK_EX; |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 123 | else |
Tao Ma | efd647f | 2010-03-30 13:21:31 +0800 | [diff] [blame] | 124 | *level = DLM_LOCK_PR; |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 125 | |
| 126 | *flags = 0; |
| 127 | if (open_flags & O_NONBLOCK) |
Tao Ma | efd647f | 2010-03-30 13:21:31 +0800 | [diff] [blame] | 128 | *flags |= DLM_LKF_NOQUEUE; |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 129 | |
| 130 | return 0; |
| 131 | } |
| 132 | |
| 133 | static int dlmfs_file_open(struct inode *inode, |
| 134 | struct file *file) |
| 135 | { |
| 136 | int status, level, flags; |
| 137 | struct dlmfs_filp_private *fp = NULL; |
| 138 | struct dlmfs_inode_private *ip; |
| 139 | |
| 140 | if (S_ISDIR(inode->i_mode)) |
| 141 | BUG(); |
| 142 | |
| 143 | mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino, |
| 144 | file->f_flags); |
| 145 | |
| 146 | status = dlmfs_decode_open_flags(file->f_flags, &level, &flags); |
| 147 | if (status < 0) |
| 148 | goto bail; |
| 149 | |
| 150 | /* We don't want to honor O_APPEND at read/write time as it |
| 151 | * doesn't make sense for LVB writes. */ |
| 152 | file->f_flags &= ~O_APPEND; |
| 153 | |
Kurt Hackel | ad8100e | 2006-05-01 14:25:21 -0700 | [diff] [blame] | 154 | fp = kmalloc(sizeof(*fp), GFP_NOFS); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 155 | if (!fp) { |
| 156 | status = -ENOMEM; |
| 157 | goto bail; |
| 158 | } |
| 159 | fp->fp_lock_level = level; |
| 160 | |
| 161 | ip = DLMFS_I(inode); |
| 162 | |
| 163 | status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags); |
| 164 | if (status < 0) { |
| 165 | /* this is a strange error to return here but I want |
| 166 | * to be able userspace to be able to distinguish a |
| 167 | * valid lock request from one that simply couldn't be |
| 168 | * granted. */ |
Tao Ma | efd647f | 2010-03-30 13:21:31 +0800 | [diff] [blame] | 169 | if (flags & DLM_LKF_NOQUEUE && status == -EAGAIN) |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 170 | status = -ETXTBSY; |
| 171 | kfree(fp); |
| 172 | goto bail; |
| 173 | } |
| 174 | |
| 175 | file->private_data = fp; |
| 176 | bail: |
| 177 | return status; |
| 178 | } |
| 179 | |
| 180 | static int dlmfs_file_release(struct inode *inode, |
| 181 | struct file *file) |
| 182 | { |
| 183 | int level, status; |
| 184 | struct dlmfs_inode_private *ip = DLMFS_I(inode); |
Joe Perches | 33fa1d9 | 2010-07-12 13:50:19 -0700 | [diff] [blame] | 185 | struct dlmfs_filp_private *fp = file->private_data; |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 186 | |
| 187 | if (S_ISDIR(inode->i_mode)) |
| 188 | BUG(); |
| 189 | |
| 190 | mlog(0, "close called on inode %lu\n", inode->i_ino); |
| 191 | |
| 192 | status = 0; |
| 193 | if (fp) { |
| 194 | level = fp->fp_lock_level; |
Tao Ma | efd647f | 2010-03-30 13:21:31 +0800 | [diff] [blame] | 195 | if (level != DLM_LOCK_IV) |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 196 | user_dlm_cluster_unlock(&ip->ip_lockres, level); |
| 197 | |
| 198 | kfree(fp); |
| 199 | file->private_data = NULL; |
| 200 | } |
| 201 | |
| 202 | return 0; |
| 203 | } |
| 204 | |
Joel Becker | e8fce48 | 2010-02-09 17:52:13 -0800 | [diff] [blame] | 205 | /* |
| 206 | * We do ->setattr() just to override size changes. Our size is the size |
| 207 | * of the LVB and nothing else. |
| 208 | */ |
| 209 | static int dlmfs_file_setattr(struct dentry *dentry, struct iattr *attr) |
| 210 | { |
| 211 | int error; |
| 212 | struct inode *inode = dentry->d_inode; |
| 213 | |
| 214 | attr->ia_valid &= ~ATTR_SIZE; |
| 215 | error = inode_change_ok(inode, attr); |
Christoph Hellwig | 1025774 | 2010-06-04 11:30:02 +0200 | [diff] [blame] | 216 | if (error) |
| 217 | return error; |
Joel Becker | e8fce48 | 2010-02-09 17:52:13 -0800 | [diff] [blame] | 218 | |
Christoph Hellwig | 1025774 | 2010-06-04 11:30:02 +0200 | [diff] [blame] | 219 | setattr_copy(inode, attr); |
| 220 | mark_inode_dirty(inode); |
| 221 | return 0; |
Joel Becker | e8fce48 | 2010-02-09 17:52:13 -0800 | [diff] [blame] | 222 | } |
| 223 | |
Joel Becker | 65b6f34 | 2010-01-26 22:14:20 -0800 | [diff] [blame] | 224 | static unsigned int dlmfs_file_poll(struct file *file, poll_table *wait) |
| 225 | { |
| 226 | int event = 0; |
| 227 | struct inode *inode = file->f_path.dentry->d_inode; |
| 228 | struct dlmfs_inode_private *ip = DLMFS_I(inode); |
| 229 | |
| 230 | poll_wait(file, &ip->ip_lockres.l_event, wait); |
| 231 | |
| 232 | spin_lock(&ip->ip_lockres.l_lock); |
| 233 | if (ip->ip_lockres.l_flags & USER_LOCK_BLOCKED) |
| 234 | event = POLLIN | POLLRDNORM; |
| 235 | spin_unlock(&ip->ip_lockres.l_lock); |
| 236 | |
| 237 | return event; |
| 238 | } |
| 239 | |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 240 | static ssize_t dlmfs_file_read(struct file *filp, |
| 241 | char __user *buf, |
| 242 | size_t count, |
| 243 | loff_t *ppos) |
| 244 | { |
| 245 | int bytes_left; |
Joel Becker | 0016eed | 2010-01-30 04:33:50 -0800 | [diff] [blame] | 246 | ssize_t readlen, got; |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 247 | char *lvb_buf; |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 248 | struct inode *inode = filp->f_path.dentry->d_inode; |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 249 | |
| 250 | mlog(0, "inode %lu, count = %zu, *ppos = %llu\n", |
| 251 | inode->i_ino, count, *ppos); |
| 252 | |
| 253 | if (*ppos >= i_size_read(inode)) |
| 254 | return 0; |
| 255 | |
| 256 | if (!count) |
| 257 | return 0; |
| 258 | |
| 259 | if (!access_ok(VERIFY_WRITE, buf, count)) |
| 260 | return -EFAULT; |
| 261 | |
| 262 | /* don't read past the lvb */ |
| 263 | if ((count + *ppos) > i_size_read(inode)) |
| 264 | readlen = i_size_read(inode) - *ppos; |
| 265 | else |
Joel Becker | a36d515 | 2010-04-23 15:24:59 -0700 | [diff] [blame] | 266 | readlen = count; |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 267 | |
Kurt Hackel | ad8100e | 2006-05-01 14:25:21 -0700 | [diff] [blame] | 268 | lvb_buf = kmalloc(readlen, GFP_NOFS); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 269 | if (!lvb_buf) |
| 270 | return -ENOMEM; |
| 271 | |
Joel Becker | 0016eed | 2010-01-30 04:33:50 -0800 | [diff] [blame] | 272 | got = user_dlm_read_lvb(inode, lvb_buf, readlen); |
| 273 | if (got) { |
| 274 | BUG_ON(got != readlen); |
| 275 | bytes_left = __copy_to_user(buf, lvb_buf, readlen); |
| 276 | readlen -= bytes_left; |
| 277 | } else |
| 278 | readlen = 0; |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 279 | |
| 280 | kfree(lvb_buf); |
| 281 | |
| 282 | *ppos = *ppos + readlen; |
| 283 | |
| 284 | mlog(0, "read %zd bytes\n", readlen); |
| 285 | return readlen; |
| 286 | } |
| 287 | |
| 288 | static ssize_t dlmfs_file_write(struct file *filp, |
| 289 | const char __user *buf, |
| 290 | size_t count, |
| 291 | loff_t *ppos) |
| 292 | { |
| 293 | int bytes_left; |
| 294 | ssize_t writelen; |
| 295 | char *lvb_buf; |
Josef Sipek | d28c917 | 2006-12-08 02:37:25 -0800 | [diff] [blame] | 296 | struct inode *inode = filp->f_path.dentry->d_inode; |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 297 | |
| 298 | mlog(0, "inode %lu, count = %zu, *ppos = %llu\n", |
| 299 | inode->i_ino, count, *ppos); |
| 300 | |
| 301 | if (*ppos >= i_size_read(inode)) |
| 302 | return -ENOSPC; |
| 303 | |
| 304 | if (!count) |
| 305 | return 0; |
| 306 | |
| 307 | if (!access_ok(VERIFY_READ, buf, count)) |
| 308 | return -EFAULT; |
| 309 | |
| 310 | /* don't write past the lvb */ |
| 311 | if ((count + *ppos) > i_size_read(inode)) |
| 312 | writelen = i_size_read(inode) - *ppos; |
| 313 | else |
| 314 | writelen = count - *ppos; |
| 315 | |
Kurt Hackel | ad8100e | 2006-05-01 14:25:21 -0700 | [diff] [blame] | 316 | lvb_buf = kmalloc(writelen, GFP_NOFS); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 317 | if (!lvb_buf) |
| 318 | return -ENOMEM; |
| 319 | |
| 320 | bytes_left = copy_from_user(lvb_buf, buf, writelen); |
| 321 | writelen -= bytes_left; |
| 322 | if (writelen) |
| 323 | user_dlm_write_lvb(inode, lvb_buf, writelen); |
| 324 | |
| 325 | kfree(lvb_buf); |
| 326 | |
| 327 | *ppos = *ppos + writelen; |
| 328 | mlog(0, "wrote %zd bytes\n", writelen); |
| 329 | return writelen; |
| 330 | } |
| 331 | |
Alexey Dobriyan | 51cc506 | 2008-07-25 19:45:34 -0700 | [diff] [blame] | 332 | static void dlmfs_init_once(void *foo) |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 333 | { |
| 334 | struct dlmfs_inode_private *ip = |
| 335 | (struct dlmfs_inode_private *) foo; |
| 336 | |
Joel Becker | 0016eed | 2010-01-30 04:33:50 -0800 | [diff] [blame] | 337 | ip->ip_conn = NULL; |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 338 | ip->ip_parent = NULL; |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 339 | |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 340 | inode_init_once(&ip->ip_vfs_inode); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 341 | } |
| 342 | |
| 343 | static struct inode *dlmfs_alloc_inode(struct super_block *sb) |
| 344 | { |
| 345 | struct dlmfs_inode_private *ip; |
| 346 | |
Christoph Lameter | e6b4f8d | 2006-12-06 20:33:14 -0800 | [diff] [blame] | 347 | ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 348 | if (!ip) |
| 349 | return NULL; |
| 350 | |
| 351 | return &ip->ip_vfs_inode; |
| 352 | } |
| 353 | |
Nick Piggin | fa0d7e3 | 2011-01-07 17:49:49 +1100 | [diff] [blame] | 354 | static void dlmfs_i_callback(struct rcu_head *head) |
| 355 | { |
| 356 | struct inode *inode = container_of(head, struct inode, i_rcu); |
| 357 | INIT_LIST_HEAD(&inode->i_dentry); |
| 358 | kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode)); |
| 359 | } |
| 360 | |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 361 | static void dlmfs_destroy_inode(struct inode *inode) |
| 362 | { |
Nick Piggin | fa0d7e3 | 2011-01-07 17:49:49 +1100 | [diff] [blame] | 363 | call_rcu(&inode->i_rcu, dlmfs_i_callback); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 364 | } |
| 365 | |
Al Viro | b57922d | 2010-06-07 14:34:48 -0400 | [diff] [blame] | 366 | static void dlmfs_evict_inode(struct inode *inode) |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 367 | { |
| 368 | int status; |
| 369 | struct dlmfs_inode_private *ip; |
| 370 | |
Al Viro | b57922d | 2010-06-07 14:34:48 -0400 | [diff] [blame] | 371 | end_writeback(inode); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 372 | |
| 373 | mlog(0, "inode %lu\n", inode->i_ino); |
| 374 | |
| 375 | ip = DLMFS_I(inode); |
| 376 | |
| 377 | if (S_ISREG(inode->i_mode)) { |
| 378 | status = user_dlm_destroy_lock(&ip->ip_lockres); |
| 379 | if (status < 0) |
| 380 | mlog_errno(status); |
| 381 | iput(ip->ip_parent); |
| 382 | goto clear_fields; |
| 383 | } |
| 384 | |
Joel Becker | 0016eed | 2010-01-30 04:33:50 -0800 | [diff] [blame] | 385 | mlog(0, "we're a directory, ip->ip_conn = 0x%p\n", ip->ip_conn); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 386 | /* we must be a directory. If required, lets unregister the |
| 387 | * dlm context now. */ |
Joel Becker | 0016eed | 2010-01-30 04:33:50 -0800 | [diff] [blame] | 388 | if (ip->ip_conn) |
| 389 | user_dlm_unregister(ip->ip_conn); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 390 | clear_fields: |
| 391 | ip->ip_parent = NULL; |
Joel Becker | 0016eed | 2010-01-30 04:33:50 -0800 | [diff] [blame] | 392 | ip->ip_conn = NULL; |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 393 | } |
| 394 | |
| 395 | static struct backing_dev_info dlmfs_backing_dev_info = { |
Jens Axboe | d993831 | 2009-06-12 14:45:52 +0200 | [diff] [blame] | 396 | .name = "ocfs2-dlmfs", |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 397 | .ra_pages = 0, /* No readahead */ |
Miklos Szeredi | e4ad08f | 2008-04-30 00:54:37 -0700 | [diff] [blame] | 398 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 399 | }; |
| 400 | |
| 401 | static struct inode *dlmfs_get_root_inode(struct super_block *sb) |
| 402 | { |
| 403 | struct inode *inode = new_inode(sb); |
| 404 | int mode = S_IFDIR | 0755; |
| 405 | struct dlmfs_inode_private *ip; |
| 406 | |
| 407 | if (inode) { |
| 408 | ip = DLMFS_I(inode); |
| 409 | |
Christoph Hellwig | 85fe402 | 2010-10-23 11:19:54 -0400 | [diff] [blame] | 410 | inode->i_ino = get_next_ino(); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 411 | inode->i_mode = mode; |
David Howells | b19c2a3 | 2008-11-14 10:38:59 +1100 | [diff] [blame] | 412 | inode->i_uid = current_fsuid(); |
| 413 | inode->i_gid = current_fsgid(); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 414 | inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info; |
| 415 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
Dave Hansen | d8c76e6 | 2006-09-30 23:29:04 -0700 | [diff] [blame] | 416 | inc_nlink(inode); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 417 | |
| 418 | inode->i_fop = &simple_dir_operations; |
| 419 | inode->i_op = &dlmfs_root_inode_operations; |
| 420 | } |
| 421 | |
| 422 | return inode; |
| 423 | } |
| 424 | |
| 425 | static struct inode *dlmfs_get_inode(struct inode *parent, |
| 426 | struct dentry *dentry, |
| 427 | int mode) |
| 428 | { |
| 429 | struct super_block *sb = parent->i_sb; |
| 430 | struct inode * inode = new_inode(sb); |
| 431 | struct dlmfs_inode_private *ip; |
| 432 | |
| 433 | if (!inode) |
| 434 | return NULL; |
| 435 | |
Christoph Hellwig | 85fe402 | 2010-10-23 11:19:54 -0400 | [diff] [blame] | 436 | inode->i_ino = get_next_ino(); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 437 | inode->i_mode = mode; |
David Howells | b19c2a3 | 2008-11-14 10:38:59 +1100 | [diff] [blame] | 438 | inode->i_uid = current_fsuid(); |
| 439 | inode->i_gid = current_fsgid(); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 440 | inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info; |
| 441 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
| 442 | |
| 443 | ip = DLMFS_I(inode); |
Joel Becker | 0016eed | 2010-01-30 04:33:50 -0800 | [diff] [blame] | 444 | ip->ip_conn = DLMFS_I(parent)->ip_conn; |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 445 | |
| 446 | switch (mode & S_IFMT) { |
| 447 | default: |
| 448 | /* for now we don't support anything other than |
| 449 | * directories and regular files. */ |
| 450 | BUG(); |
| 451 | break; |
| 452 | case S_IFREG: |
| 453 | inode->i_op = &dlmfs_file_inode_operations; |
| 454 | inode->i_fop = &dlmfs_file_operations; |
| 455 | |
| 456 | i_size_write(inode, DLM_LVB_LEN); |
| 457 | |
| 458 | user_dlm_lock_res_init(&ip->ip_lockres, dentry); |
| 459 | |
| 460 | /* released at clear_inode time, this insures that we |
| 461 | * get to drop the dlm reference on each lock *before* |
| 462 | * we call the unregister code for releasing parent |
| 463 | * directories. */ |
| 464 | ip->ip_parent = igrab(parent); |
| 465 | BUG_ON(!ip->ip_parent); |
| 466 | break; |
| 467 | case S_IFDIR: |
| 468 | inode->i_op = &dlmfs_dir_inode_operations; |
| 469 | inode->i_fop = &simple_dir_operations; |
| 470 | |
| 471 | /* directory inodes start off with i_nlink == |
| 472 | * 2 (for "." entry) */ |
Dave Hansen | d8c76e6 | 2006-09-30 23:29:04 -0700 | [diff] [blame] | 473 | inc_nlink(inode); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 474 | break; |
| 475 | } |
| 476 | |
| 477 | if (parent->i_mode & S_ISGID) { |
| 478 | inode->i_gid = parent->i_gid; |
| 479 | if (S_ISDIR(mode)) |
| 480 | inode->i_mode |= S_ISGID; |
| 481 | } |
| 482 | |
| 483 | return inode; |
| 484 | } |
| 485 | |
| 486 | /* |
| 487 | * File creation. Allocate an inode, and we're done.. |
| 488 | */ |
| 489 | /* SMP-safe */ |
| 490 | static int dlmfs_mkdir(struct inode * dir, |
| 491 | struct dentry * dentry, |
| 492 | int mode) |
| 493 | { |
| 494 | int status; |
| 495 | struct inode *inode = NULL; |
| 496 | struct qstr *domain = &dentry->d_name; |
| 497 | struct dlmfs_inode_private *ip; |
Joel Becker | 0016eed | 2010-01-30 04:33:50 -0800 | [diff] [blame] | 498 | struct ocfs2_cluster_connection *conn; |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 499 | |
| 500 | mlog(0, "mkdir %.*s\n", domain->len, domain->name); |
| 501 | |
| 502 | /* verify that we have a proper domain */ |
Joel Becker | 0016eed | 2010-01-30 04:33:50 -0800 | [diff] [blame] | 503 | if (domain->len >= GROUP_NAME_MAX) { |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 504 | status = -EINVAL; |
| 505 | mlog(ML_ERROR, "invalid domain name for directory.\n"); |
| 506 | goto bail; |
| 507 | } |
| 508 | |
| 509 | inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR); |
| 510 | if (!inode) { |
| 511 | status = -ENOMEM; |
| 512 | mlog_errno(status); |
| 513 | goto bail; |
| 514 | } |
| 515 | |
| 516 | ip = DLMFS_I(inode); |
| 517 | |
Joel Becker | 0016eed | 2010-01-30 04:33:50 -0800 | [diff] [blame] | 518 | conn = user_dlm_register(domain); |
| 519 | if (IS_ERR(conn)) { |
| 520 | status = PTR_ERR(conn); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 521 | mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n", |
| 522 | status, domain->len, domain->name); |
| 523 | goto bail; |
| 524 | } |
Joel Becker | 0016eed | 2010-01-30 04:33:50 -0800 | [diff] [blame] | 525 | ip->ip_conn = conn; |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 526 | |
Dave Hansen | d8c76e6 | 2006-09-30 23:29:04 -0700 | [diff] [blame] | 527 | inc_nlink(dir); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 528 | d_instantiate(dentry, inode); |
| 529 | dget(dentry); /* Extra count - pin the dentry in core */ |
| 530 | |
| 531 | status = 0; |
| 532 | bail: |
| 533 | if (status < 0) |
| 534 | iput(inode); |
| 535 | return status; |
| 536 | } |
| 537 | |
| 538 | static int dlmfs_create(struct inode *dir, |
| 539 | struct dentry *dentry, |
| 540 | int mode, |
| 541 | struct nameidata *nd) |
| 542 | { |
| 543 | int status = 0; |
| 544 | struct inode *inode; |
| 545 | struct qstr *name = &dentry->d_name; |
| 546 | |
| 547 | mlog(0, "create %.*s\n", name->len, name->name); |
| 548 | |
| 549 | /* verify name is valid and doesn't contain any dlm reserved |
| 550 | * characters */ |
| 551 | if (name->len >= USER_DLM_LOCK_ID_MAX_LEN || |
| 552 | name->name[0] == '$') { |
| 553 | status = -EINVAL; |
| 554 | mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len, |
| 555 | name->name); |
| 556 | goto bail; |
| 557 | } |
| 558 | |
| 559 | inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG); |
| 560 | if (!inode) { |
| 561 | status = -ENOMEM; |
| 562 | mlog_errno(status); |
| 563 | goto bail; |
| 564 | } |
| 565 | |
| 566 | d_instantiate(dentry, inode); |
| 567 | dget(dentry); /* Extra count - pin the dentry in core */ |
| 568 | bail: |
| 569 | return status; |
| 570 | } |
| 571 | |
| 572 | static int dlmfs_unlink(struct inode *dir, |
| 573 | struct dentry *dentry) |
| 574 | { |
| 575 | int status; |
| 576 | struct inode *inode = dentry->d_inode; |
| 577 | |
| 578 | mlog(0, "unlink inode %lu\n", inode->i_ino); |
| 579 | |
| 580 | /* if there are no current holders, or none that are waiting |
| 581 | * to acquire a lock, this basically destroys our lockres. */ |
| 582 | status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres); |
| 583 | if (status < 0) { |
| 584 | mlog(ML_ERROR, "unlink %.*s, error %d from destroy\n", |
| 585 | dentry->d_name.len, dentry->d_name.name, status); |
| 586 | goto bail; |
| 587 | } |
| 588 | status = simple_unlink(dir, dentry); |
| 589 | bail: |
| 590 | return status; |
| 591 | } |
| 592 | |
| 593 | static int dlmfs_fill_super(struct super_block * sb, |
| 594 | void * data, |
| 595 | int silent) |
| 596 | { |
| 597 | struct inode * inode; |
| 598 | struct dentry * root; |
| 599 | |
| 600 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
| 601 | sb->s_blocksize = PAGE_CACHE_SIZE; |
| 602 | sb->s_blocksize_bits = PAGE_CACHE_SHIFT; |
| 603 | sb->s_magic = DLMFS_MAGIC; |
| 604 | sb->s_op = &dlmfs_ops; |
| 605 | inode = dlmfs_get_root_inode(sb); |
| 606 | if (!inode) |
| 607 | return -ENOMEM; |
| 608 | |
| 609 | root = d_alloc_root(inode); |
| 610 | if (!root) { |
| 611 | iput(inode); |
| 612 | return -ENOMEM; |
| 613 | } |
| 614 | sb->s_root = root; |
| 615 | return 0; |
| 616 | } |
| 617 | |
Arjan van de Ven | 00977a5 | 2007-02-12 00:55:34 -0800 | [diff] [blame] | 618 | static const struct file_operations dlmfs_file_operations = { |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 619 | .open = dlmfs_file_open, |
| 620 | .release = dlmfs_file_release, |
Joel Becker | 65b6f34 | 2010-01-26 22:14:20 -0800 | [diff] [blame] | 621 | .poll = dlmfs_file_poll, |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 622 | .read = dlmfs_file_read, |
| 623 | .write = dlmfs_file_write, |
Arnd Bergmann | 6038f37 | 2010-08-15 18:52:59 +0200 | [diff] [blame] | 624 | .llseek = default_llseek, |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 625 | }; |
| 626 | |
Arjan van de Ven | 92e1d5b | 2007-02-12 00:55:39 -0800 | [diff] [blame] | 627 | static const struct inode_operations dlmfs_dir_inode_operations = { |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 628 | .create = dlmfs_create, |
| 629 | .lookup = simple_lookup, |
| 630 | .unlink = dlmfs_unlink, |
| 631 | }; |
| 632 | |
| 633 | /* this way we can restrict mkdir to only the toplevel of the fs. */ |
Arjan van de Ven | 92e1d5b | 2007-02-12 00:55:39 -0800 | [diff] [blame] | 634 | static const struct inode_operations dlmfs_root_inode_operations = { |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 635 | .lookup = simple_lookup, |
| 636 | .mkdir = dlmfs_mkdir, |
| 637 | .rmdir = simple_rmdir, |
| 638 | }; |
| 639 | |
Josef 'Jeff' Sipek | ee9b6d6 | 2007-02-12 00:55:41 -0800 | [diff] [blame] | 640 | static const struct super_operations dlmfs_ops = { |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 641 | .statfs = simple_statfs, |
| 642 | .alloc_inode = dlmfs_alloc_inode, |
| 643 | .destroy_inode = dlmfs_destroy_inode, |
Al Viro | b57922d | 2010-06-07 14:34:48 -0400 | [diff] [blame] | 644 | .evict_inode = dlmfs_evict_inode, |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 645 | .drop_inode = generic_delete_inode, |
| 646 | }; |
| 647 | |
Arjan van de Ven | 92e1d5b | 2007-02-12 00:55:39 -0800 | [diff] [blame] | 648 | static const struct inode_operations dlmfs_file_inode_operations = { |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 649 | .getattr = simple_getattr, |
Joel Becker | e8fce48 | 2010-02-09 17:52:13 -0800 | [diff] [blame] | 650 | .setattr = dlmfs_file_setattr, |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 651 | }; |
| 652 | |
Al Viro | 3c26ff6 | 2010-07-25 11:46:36 +0400 | [diff] [blame] | 653 | static struct dentry *dlmfs_mount(struct file_system_type *fs_type, |
| 654 | int flags, const char *dev_name, void *data) |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 655 | { |
Al Viro | 3c26ff6 | 2010-07-25 11:46:36 +0400 | [diff] [blame] | 656 | return mount_nodev(fs_type, flags, data, dlmfs_fill_super); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 657 | } |
| 658 | |
| 659 | static struct file_system_type dlmfs_fs_type = { |
| 660 | .owner = THIS_MODULE, |
| 661 | .name = "ocfs2_dlmfs", |
Al Viro | 3c26ff6 | 2010-07-25 11:46:36 +0400 | [diff] [blame] | 662 | .mount = dlmfs_mount, |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 663 | .kill_sb = kill_litter_super, |
| 664 | }; |
| 665 | |
| 666 | static int __init init_dlmfs_fs(void) |
| 667 | { |
| 668 | int status; |
| 669 | int cleanup_inode = 0, cleanup_worker = 0; |
| 670 | |
| 671 | dlmfs_print_version(); |
| 672 | |
Peter Zijlstra | e0bf68d | 2007-10-16 23:25:46 -0700 | [diff] [blame] | 673 | status = bdi_init(&dlmfs_backing_dev_info); |
| 674 | if (status) |
| 675 | return status; |
| 676 | |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 677 | dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache", |
| 678 | sizeof(struct dlmfs_inode_private), |
Paul Jackson | fffb60f | 2006-03-24 03:16:06 -0800 | [diff] [blame] | 679 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
| 680 | SLAB_MEM_SPREAD), |
Paul Mundt | 20c2df8 | 2007-07-20 10:11:58 +0900 | [diff] [blame] | 681 | dlmfs_init_once); |
Coly Li | 07d9a39 | 2008-11-17 12:38:22 +0800 | [diff] [blame] | 682 | if (!dlmfs_inode_cache) { |
| 683 | status = -ENOMEM; |
Peter Zijlstra | e0bf68d | 2007-10-16 23:25:46 -0700 | [diff] [blame] | 684 | goto bail; |
Coly Li | 07d9a39 | 2008-11-17 12:38:22 +0800 | [diff] [blame] | 685 | } |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 686 | cleanup_inode = 1; |
| 687 | |
| 688 | user_dlm_worker = create_singlethread_workqueue("user_dlm"); |
| 689 | if (!user_dlm_worker) { |
| 690 | status = -ENOMEM; |
| 691 | goto bail; |
| 692 | } |
| 693 | cleanup_worker = 1; |
| 694 | |
Joel Becker | 0016eed | 2010-01-30 04:33:50 -0800 | [diff] [blame] | 695 | user_dlm_set_locking_protocol(); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 696 | status = register_filesystem(&dlmfs_fs_type); |
| 697 | bail: |
| 698 | if (status) { |
| 699 | if (cleanup_inode) |
| 700 | kmem_cache_destroy(dlmfs_inode_cache); |
| 701 | if (cleanup_worker) |
| 702 | destroy_workqueue(user_dlm_worker); |
Peter Zijlstra | e0bf68d | 2007-10-16 23:25:46 -0700 | [diff] [blame] | 703 | bdi_destroy(&dlmfs_backing_dev_info); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 704 | } else |
| 705 | printk("OCFS2 User DLM kernel interface loaded\n"); |
| 706 | return status; |
| 707 | } |
| 708 | |
| 709 | static void __exit exit_dlmfs_fs(void) |
| 710 | { |
| 711 | unregister_filesystem(&dlmfs_fs_type); |
| 712 | |
| 713 | flush_workqueue(user_dlm_worker); |
| 714 | destroy_workqueue(user_dlm_worker); |
| 715 | |
Alexey Dobriyan | 1a1d92c | 2006-09-27 01:49:40 -0700 | [diff] [blame] | 716 | kmem_cache_destroy(dlmfs_inode_cache); |
Peter Zijlstra | e0bf68d | 2007-10-16 23:25:46 -0700 | [diff] [blame] | 717 | |
| 718 | bdi_destroy(&dlmfs_backing_dev_info); |
Mark Fasheh | 8df08c8 | 2005-12-15 14:31:23 -0800 | [diff] [blame] | 719 | } |
| 720 | |
| 721 | MODULE_AUTHOR("Oracle"); |
| 722 | MODULE_LICENSE("GPL"); |
| 723 | |
| 724 | module_init(init_dlmfs_fs) |
| 725 | module_exit(exit_dlmfs_fs) |