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> 45 #include <linux/backing-dev.h> 46 #include <linux/poll.h> 47 48 #include <linux/uaccess.h> 49 50 #include "stackglue.h" 51 #include "userdlm.h" 52 53 #define MLOG_MASK_PREFIX ML_DLMFS 54 #include "cluster/masklog.h" 55 56 57 static const struct super_operations dlmfs_ops; 58 static const struct file_operations dlmfs_file_operations; 59 static const struct inode_operations dlmfs_dir_inode_operations; 60 static const struct inode_operations dlmfs_root_inode_operations; 61 static const struct inode_operations dlmfs_file_inode_operations; 62 static struct kmem_cache *dlmfs_inode_cache; 63 64 struct workqueue_struct *user_dlm_worker; 65 66 67 68 /* 69 * These are the ABI capabilities of dlmfs. 70 * 71 * Over time, dlmfs has added some features that were not part of the 72 * initial ABI. Unfortunately, some of these features are not detectable 73 * via standard usage. For example, Linux's default poll always returns 74 * EPOLLIN, so there is no way for a caller of poll(2) to know when dlmfs 75 * added poll support. Instead, we provide this list of new capabilities. 76 * 77 * Capabilities is a read-only attribute. We do it as a module parameter 78 * so we can discover it whether dlmfs is built in, loaded, or even not 79 * loaded. 80 * 81 * The ABI features are local to this machine's dlmfs mount. This is 82 * distinct from the locking protocol, which is concerned with inter-node 83 * interaction. 84 * 85 * Capabilities: 86 * - bast : EPOLLIN against the file descriptor of a held lock 87 * signifies a bast fired on the lock. 88 */ 89 #define DLMFS_CAPABILITIES "bast stackglue" 90 static int param_set_dlmfs_capabilities(const char *val, 91 const struct kernel_param *kp) 92 { 93 printk(KERN_ERR "%s: readonly parameter\n", kp->name); 94 return -EINVAL; 95 } 96 static int param_get_dlmfs_capabilities(char *buffer, 97 const struct kernel_param *kp) 98 { 99 return strlcpy(buffer, DLMFS_CAPABILITIES, 100 strlen(DLMFS_CAPABILITIES) + 1); 101 } 102 module_param_call(capabilities, param_set_dlmfs_capabilities, 103 param_get_dlmfs_capabilities, NULL, 0444); 104 MODULE_PARM_DESC(capabilities, DLMFS_CAPABILITIES); 105 106 107 /* 108 * decodes a set of open flags into a valid lock level and a set of flags. 109 * returns < 0 if we have invalid flags 110 * flags which mean something to us: 111 * O_RDONLY -> PRMODE level 112 * O_WRONLY -> EXMODE level 113 * 114 * O_NONBLOCK -> NOQUEUE 115 */ 116 static int dlmfs_decode_open_flags(int open_flags, 117 int *level, 118 int *flags) 119 { 120 if (open_flags & (O_WRONLY|O_RDWR)) 121 *level = DLM_LOCK_EX; 122 else 123 *level = DLM_LOCK_PR; 124 125 *flags = 0; 126 if (open_flags & O_NONBLOCK) 127 *flags |= DLM_LKF_NOQUEUE; 128 129 return 0; 130 } 131 132 static int dlmfs_file_open(struct inode *inode, 133 struct file *file) 134 { 135 int status, level, flags; 136 struct dlmfs_filp_private *fp = NULL; 137 struct dlmfs_inode_private *ip; 138 139 if (S_ISDIR(inode->i_mode)) 140 BUG(); 141 142 mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino, 143 file->f_flags); 144 145 status = dlmfs_decode_open_flags(file->f_flags, &level, &flags); 146 if (status < 0) 147 goto bail; 148 149 /* We don't want to honor O_APPEND at read/write time as it 150 * doesn't make sense for LVB writes. */ 151 file->f_flags &= ~O_APPEND; 152 153 fp = kmalloc(sizeof(*fp), GFP_NOFS); 154 if (!fp) { 155 status = -ENOMEM; 156 goto bail; 157 } 158 fp->fp_lock_level = level; 159 160 ip = DLMFS_I(inode); 161 162 status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags); 163 if (status < 0) { 164 /* this is a strange error to return here but I want 165 * to be able userspace to be able to distinguish a 166 * valid lock request from one that simply couldn't be 167 * granted. */ 168 if (flags & DLM_LKF_NOQUEUE && status == -EAGAIN) 169 status = -ETXTBSY; 170 kfree(fp); 171 goto bail; 172 } 173 174 file->private_data = fp; 175 bail: 176 return status; 177 } 178 179 static int dlmfs_file_release(struct inode *inode, 180 struct file *file) 181 { 182 int level; 183 struct dlmfs_inode_private *ip = DLMFS_I(inode); 184 struct dlmfs_filp_private *fp = file->private_data; 185 186 if (S_ISDIR(inode->i_mode)) 187 BUG(); 188 189 mlog(0, "close called on inode %lu\n", inode->i_ino); 190 191 if (fp) { 192 level = fp->fp_lock_level; 193 if (level != DLM_LOCK_IV) 194 user_dlm_cluster_unlock(&ip->ip_lockres, level); 195 196 kfree(fp); 197 file->private_data = NULL; 198 } 199 200 return 0; 201 } 202 203 /* 204 * We do ->setattr() just to override size changes. Our size is the size 205 * of the LVB and nothing else. 206 */ 207 static int dlmfs_file_setattr(struct dentry *dentry, struct iattr *attr) 208 { 209 int error; 210 struct inode *inode = d_inode(dentry); 211 212 attr->ia_valid &= ~ATTR_SIZE; 213 error = setattr_prepare(dentry, attr); 214 if (error) 215 return error; 216 217 setattr_copy(inode, attr); 218 mark_inode_dirty(inode); 219 return 0; 220 } 221 222 static __poll_t dlmfs_file_poll(struct file *file, poll_table *wait) 223 { 224 __poll_t event = 0; 225 struct inode *inode = file_inode(file); 226 struct dlmfs_inode_private *ip = DLMFS_I(inode); 227 228 poll_wait(file, &ip->ip_lockres.l_event, wait); 229 230 spin_lock(&ip->ip_lockres.l_lock); 231 if (ip->ip_lockres.l_flags & USER_LOCK_BLOCKED) 232 event = EPOLLIN | EPOLLRDNORM; 233 spin_unlock(&ip->ip_lockres.l_lock); 234 235 return event; 236 } 237 238 static ssize_t dlmfs_file_read(struct file *filp, 239 char __user *buf, 240 size_t count, 241 loff_t *ppos) 242 { 243 int bytes_left; 244 ssize_t readlen, got; 245 char *lvb_buf; 246 struct inode *inode = file_inode(filp); 247 248 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n", 249 inode->i_ino, count, *ppos); 250 251 if (*ppos >= i_size_read(inode)) 252 return 0; 253 254 if (!count) 255 return 0; 256 257 if (!access_ok(buf, count)) 258 return -EFAULT; 259 260 /* don't read past the lvb */ 261 if ((count + *ppos) > i_size_read(inode)) 262 readlen = i_size_read(inode) - *ppos; 263 else 264 readlen = count; 265 266 lvb_buf = kmalloc(readlen, GFP_NOFS); 267 if (!lvb_buf) 268 return -ENOMEM; 269 270 got = user_dlm_read_lvb(inode, lvb_buf, readlen); 271 if (got) { 272 BUG_ON(got != readlen); 273 bytes_left = __copy_to_user(buf, lvb_buf, readlen); 274 readlen -= bytes_left; 275 } else 276 readlen = 0; 277 278 kfree(lvb_buf); 279 280 *ppos = *ppos + readlen; 281 282 mlog(0, "read %zd bytes\n", readlen); 283 return readlen; 284 } 285 286 static ssize_t dlmfs_file_write(struct file *filp, 287 const char __user *buf, 288 size_t count, 289 loff_t *ppos) 290 { 291 int bytes_left; 292 ssize_t writelen; 293 char *lvb_buf; 294 struct inode *inode = file_inode(filp); 295 296 mlog(0, "inode %lu, count = %zu, *ppos = %llu\n", 297 inode->i_ino, count, *ppos); 298 299 if (*ppos >= i_size_read(inode)) 300 return -ENOSPC; 301 302 if (!count) 303 return 0; 304 305 if (!access_ok(buf, count)) 306 return -EFAULT; 307 308 /* don't write past the lvb */ 309 if ((count + *ppos) > i_size_read(inode)) 310 writelen = i_size_read(inode) - *ppos; 311 else 312 writelen = count - *ppos; 313 314 lvb_buf = kmalloc(writelen, GFP_NOFS); 315 if (!lvb_buf) 316 return -ENOMEM; 317 318 bytes_left = copy_from_user(lvb_buf, buf, writelen); 319 writelen -= bytes_left; 320 if (writelen) 321 user_dlm_write_lvb(inode, lvb_buf, writelen); 322 323 kfree(lvb_buf); 324 325 *ppos = *ppos + writelen; 326 mlog(0, "wrote %zd bytes\n", writelen); 327 return writelen; 328 } 329 330 static void dlmfs_init_once(void *foo) 331 { 332 struct dlmfs_inode_private *ip = 333 (struct dlmfs_inode_private *) foo; 334 335 ip->ip_conn = NULL; 336 ip->ip_parent = NULL; 337 338 inode_init_once(&ip->ip_vfs_inode); 339 } 340 341 static struct inode *dlmfs_alloc_inode(struct super_block *sb) 342 { 343 struct dlmfs_inode_private *ip; 344 345 ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS); 346 if (!ip) 347 return NULL; 348 349 return &ip->ip_vfs_inode; 350 } 351 352 static void dlmfs_free_inode(struct inode *inode) 353 { 354 kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode)); 355 } 356 357 static void dlmfs_evict_inode(struct inode *inode) 358 { 359 int status; 360 struct dlmfs_inode_private *ip; 361 362 clear_inode(inode); 363 364 mlog(0, "inode %lu\n", inode->i_ino); 365 366 ip = DLMFS_I(inode); 367 368 if (S_ISREG(inode->i_mode)) { 369 status = user_dlm_destroy_lock(&ip->ip_lockres); 370 if (status < 0) 371 mlog_errno(status); 372 iput(ip->ip_parent); 373 goto clear_fields; 374 } 375 376 mlog(0, "we're a directory, ip->ip_conn = 0x%p\n", ip->ip_conn); 377 /* we must be a directory. If required, lets unregister the 378 * dlm context now. */ 379 if (ip->ip_conn) 380 user_dlm_unregister(ip->ip_conn); 381 clear_fields: 382 ip->ip_parent = NULL; 383 ip->ip_conn = NULL; 384 } 385 386 static struct inode *dlmfs_get_root_inode(struct super_block *sb) 387 { 388 struct inode *inode = new_inode(sb); 389 umode_t mode = S_IFDIR | 0755; 390 391 if (inode) { 392 inode->i_ino = get_next_ino(); 393 inode_init_owner(inode, NULL, mode); 394 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 395 inc_nlink(inode); 396 397 inode->i_fop = &simple_dir_operations; 398 inode->i_op = &dlmfs_root_inode_operations; 399 } 400 401 return inode; 402 } 403 404 static struct inode *dlmfs_get_inode(struct inode *parent, 405 struct dentry *dentry, 406 umode_t mode) 407 { 408 struct super_block *sb = parent->i_sb; 409 struct inode * inode = new_inode(sb); 410 struct dlmfs_inode_private *ip; 411 412 if (!inode) 413 return NULL; 414 415 inode->i_ino = get_next_ino(); 416 inode_init_owner(inode, parent, mode); 417 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 418 419 ip = DLMFS_I(inode); 420 ip->ip_conn = DLMFS_I(parent)->ip_conn; 421 422 switch (mode & S_IFMT) { 423 default: 424 /* for now we don't support anything other than 425 * directories and regular files. */ 426 BUG(); 427 break; 428 case S_IFREG: 429 inode->i_op = &dlmfs_file_inode_operations; 430 inode->i_fop = &dlmfs_file_operations; 431 432 i_size_write(inode, DLM_LVB_LEN); 433 434 user_dlm_lock_res_init(&ip->ip_lockres, dentry); 435 436 /* released at clear_inode time, this insures that we 437 * get to drop the dlm reference on each lock *before* 438 * we call the unregister code for releasing parent 439 * directories. */ 440 ip->ip_parent = igrab(parent); 441 BUG_ON(!ip->ip_parent); 442 break; 443 case S_IFDIR: 444 inode->i_op = &dlmfs_dir_inode_operations; 445 inode->i_fop = &simple_dir_operations; 446 447 /* directory inodes start off with i_nlink == 448 * 2 (for "." entry) */ 449 inc_nlink(inode); 450 break; 451 } 452 return inode; 453 } 454 455 /* 456 * File creation. Allocate an inode, and we're done.. 457 */ 458 /* SMP-safe */ 459 static int dlmfs_mkdir(struct inode * dir, 460 struct dentry * dentry, 461 umode_t mode) 462 { 463 int status; 464 struct inode *inode = NULL; 465 const struct qstr *domain = &dentry->d_name; 466 struct dlmfs_inode_private *ip; 467 struct ocfs2_cluster_connection *conn; 468 469 mlog(0, "mkdir %.*s\n", domain->len, domain->name); 470 471 /* verify that we have a proper domain */ 472 if (domain->len >= GROUP_NAME_MAX) { 473 status = -EINVAL; 474 mlog(ML_ERROR, "invalid domain name for directory.\n"); 475 goto bail; 476 } 477 478 inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR); 479 if (!inode) { 480 status = -ENOMEM; 481 mlog_errno(status); 482 goto bail; 483 } 484 485 ip = DLMFS_I(inode); 486 487 conn = user_dlm_register(domain); 488 if (IS_ERR(conn)) { 489 status = PTR_ERR(conn); 490 mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n", 491 status, domain->len, domain->name); 492 goto bail; 493 } 494 ip->ip_conn = conn; 495 496 inc_nlink(dir); 497 d_instantiate(dentry, inode); 498 dget(dentry); /* Extra count - pin the dentry in core */ 499 500 status = 0; 501 bail: 502 if (status < 0) 503 iput(inode); 504 return status; 505 } 506 507 static int dlmfs_create(struct inode *dir, 508 struct dentry *dentry, 509 umode_t mode, 510 bool excl) 511 { 512 int status = 0; 513 struct inode *inode; 514 const struct qstr *name = &dentry->d_name; 515 516 mlog(0, "create %.*s\n", name->len, name->name); 517 518 /* verify name is valid and doesn't contain any dlm reserved 519 * characters */ 520 if (name->len >= USER_DLM_LOCK_ID_MAX_LEN || 521 name->name[0] == '$') { 522 status = -EINVAL; 523 mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len, 524 name->name); 525 goto bail; 526 } 527 528 inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG); 529 if (!inode) { 530 status = -ENOMEM; 531 mlog_errno(status); 532 goto bail; 533 } 534 535 d_instantiate(dentry, inode); 536 dget(dentry); /* Extra count - pin the dentry in core */ 537 bail: 538 return status; 539 } 540 541 static int dlmfs_unlink(struct inode *dir, 542 struct dentry *dentry) 543 { 544 int status; 545 struct inode *inode = d_inode(dentry); 546 547 mlog(0, "unlink inode %lu\n", inode->i_ino); 548 549 /* if there are no current holders, or none that are waiting 550 * to acquire a lock, this basically destroys our lockres. */ 551 status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres); 552 if (status < 0) { 553 mlog(ML_ERROR, "unlink %pd, error %d from destroy\n", 554 dentry, status); 555 goto bail; 556 } 557 status = simple_unlink(dir, dentry); 558 bail: 559 return status; 560 } 561 562 static int dlmfs_fill_super(struct super_block * sb, 563 void * data, 564 int silent) 565 { 566 sb->s_maxbytes = MAX_LFS_FILESIZE; 567 sb->s_blocksize = PAGE_SIZE; 568 sb->s_blocksize_bits = PAGE_SHIFT; 569 sb->s_magic = DLMFS_MAGIC; 570 sb->s_op = &dlmfs_ops; 571 sb->s_root = d_make_root(dlmfs_get_root_inode(sb)); 572 if (!sb->s_root) 573 return -ENOMEM; 574 return 0; 575 } 576 577 static const struct file_operations dlmfs_file_operations = { 578 .open = dlmfs_file_open, 579 .release = dlmfs_file_release, 580 .poll = dlmfs_file_poll, 581 .read = dlmfs_file_read, 582 .write = dlmfs_file_write, 583 .llseek = default_llseek, 584 }; 585 586 static const struct inode_operations dlmfs_dir_inode_operations = { 587 .create = dlmfs_create, 588 .lookup = simple_lookup, 589 .unlink = dlmfs_unlink, 590 }; 591 592 /* this way we can restrict mkdir to only the toplevel of the fs. */ 593 static const struct inode_operations dlmfs_root_inode_operations = { 594 .lookup = simple_lookup, 595 .mkdir = dlmfs_mkdir, 596 .rmdir = simple_rmdir, 597 }; 598 599 static const struct super_operations dlmfs_ops = { 600 .statfs = simple_statfs, 601 .alloc_inode = dlmfs_alloc_inode, 602 .free_inode = dlmfs_free_inode, 603 .evict_inode = dlmfs_evict_inode, 604 .drop_inode = generic_delete_inode, 605 }; 606 607 static const struct inode_operations dlmfs_file_inode_operations = { 608 .getattr = simple_getattr, 609 .setattr = dlmfs_file_setattr, 610 }; 611 612 static struct dentry *dlmfs_mount(struct file_system_type *fs_type, 613 int flags, const char *dev_name, void *data) 614 { 615 return mount_nodev(fs_type, flags, data, dlmfs_fill_super); 616 } 617 618 static struct file_system_type dlmfs_fs_type = { 619 .owner = THIS_MODULE, 620 .name = "ocfs2_dlmfs", 621 .mount = dlmfs_mount, 622 .kill_sb = kill_litter_super, 623 }; 624 MODULE_ALIAS_FS("ocfs2_dlmfs"); 625 626 static int __init init_dlmfs_fs(void) 627 { 628 int status; 629 int cleanup_inode = 0, cleanup_worker = 0; 630 631 dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache", 632 sizeof(struct dlmfs_inode_private), 633 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| 634 SLAB_MEM_SPREAD|SLAB_ACCOUNT), 635 dlmfs_init_once); 636 if (!dlmfs_inode_cache) { 637 status = -ENOMEM; 638 goto bail; 639 } 640 cleanup_inode = 1; 641 642 user_dlm_worker = alloc_workqueue("user_dlm", WQ_MEM_RECLAIM, 0); 643 if (!user_dlm_worker) { 644 status = -ENOMEM; 645 goto bail; 646 } 647 cleanup_worker = 1; 648 649 user_dlm_set_locking_protocol(); 650 status = register_filesystem(&dlmfs_fs_type); 651 bail: 652 if (status) { 653 if (cleanup_inode) 654 kmem_cache_destroy(dlmfs_inode_cache); 655 if (cleanup_worker) 656 destroy_workqueue(user_dlm_worker); 657 } else 658 printk("OCFS2 User DLM kernel interface loaded\n"); 659 return status; 660 } 661 662 static void __exit exit_dlmfs_fs(void) 663 { 664 unregister_filesystem(&dlmfs_fs_type); 665 666 destroy_workqueue(user_dlm_worker); 667 668 /* 669 * Make sure all delayed rcu free inodes are flushed before we 670 * destroy cache. 671 */ 672 rcu_barrier(); 673 kmem_cache_destroy(dlmfs_inode_cache); 674 675 } 676 677 MODULE_AUTHOR("Oracle"); 678 MODULE_LICENSE("GPL"); 679 MODULE_DESCRIPTION("OCFS2 DLM-Filesystem"); 680 681 module_init(init_dlmfs_fs) 682 module_exit(exit_dlmfs_fs) 683