fs/ocfs2/dlm/dlmfs.c

   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/smp_lock.h>
  46 #include <linux/backing-dev.h>
  47
  48 #include <asm/uaccess.h>
  49
  50
  51 #include "cluster/nodemanager.h"
  52 #include "cluster/heartbeat.h"
  53 #include "cluster/tcp.h"
  54
  55 #include "dlmapi.h"
  56
  57 #include "userdlm.h"
  58
  59 #include "dlmfsver.h"
  60
  61 #define MLOG_MASK_PREFIX ML_DLMFS
  62 #include "cluster/masklog.h"
  63
  64 static const struct super_operations dlmfs_ops;
  65 static const struct file_operations dlmfs_file_operations;
  66 static const struct inode_operations dlmfs_dir_inode_operations;
  67 static const struct inode_operations dlmfs_root_inode_operations;
  68 static const struct inode_operations dlmfs_file_inode_operations;
  69 static struct kmem_cache *dlmfs_inode_cache;
  70
  71 struct workqueue_struct *user_dlm_worker;
  72
  73 /*
  74  * decodes a set of open flags into a valid lock level and a set of flags.
  75  * returns < 0 if we have invalid flags
  76  * flags which mean something to us:
  77  * O_RDONLY -> PRMODE level
  78  * O_WRONLY -> EXMODE level
  79  *
  80  * O_NONBLOCK -> LKM_NOQUEUE
  81  */
  82 static int dlmfs_decode_open_flags(int open_flags,
  83                                    int *level,
  84                                    int *flags)
  85 {
  86         if (open_flags & (O_WRONLY|O_RDWR))
  87                 *level = LKM_EXMODE;
  88         else
  89                 *level = LKM_PRMODE;
  90
  91         *flags = 0;
  92         if (open_flags & O_NONBLOCK)
  93                 *flags |= LKM_NOQUEUE;
  94
  95         return 0;
  96 }
  97
  98 static int dlmfs_file_open(struct inode *inode,
  99                            struct file *file)
 100 {
 101         int status, level, flags;
 102         struct dlmfs_filp_private *fp = NULL;
 103         struct dlmfs_inode_private *ip;
 104
 105         if (S_ISDIR(inode->i_mode))
 106                 BUG();
 107
 108         mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino,
 109                 file->f_flags);
 110
 111         status = dlmfs_decode_open_flags(file->f_flags, &level, &flags);
 112         if (status < 0)
 113                 goto bail;
 114
 115         /* We don't want to honor O_APPEND at read/write time as it
 116          * doesn't make sense for LVB writes. */
 117         file->f_flags &= ~O_APPEND;
 118
 119         fp = kmalloc(sizeof(*fp), GFP_NOFS);
 120         if (!fp) {
 121                 status = -ENOMEM;
 122                 goto bail;
 123         }
 124         fp->fp_lock_level = level;
 125
 126         ip = DLMFS_I(inode);
 127
 128         status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags);
 129         if (status < 0) {
 130                 /* this is a strange error to return here but I want
 131                  * to be able userspace to be able to distinguish a
 132                  * valid lock request from one that simply couldn't be
 133                  * granted. */
 134                 if (flags & LKM_NOQUEUE && status == -EAGAIN)
 135                         status = -ETXTBSY;
 136                 kfree(fp);
 137                 goto bail;
 138         }
 139
 140         file->private_data = fp;
 141 bail:
 142         return status;
 143 }
 144
 145 static int dlmfs_file_release(struct inode *inode,
 146                               struct file *file)
 147 {
 148         int level, status;
 149         struct dlmfs_inode_private *ip = DLMFS_I(inode);
 150         struct dlmfs_filp_private *fp =
 151                 (struct dlmfs_filp_private *) file->private_data;
 152
 153         if (S_ISDIR(inode->i_mode))
 154                 BUG();
 155
 156         mlog(0, "close called on inode %lu\n", inode->i_ino);
 157
 158         status = 0;
 159         if (fp) {
 160                 level = fp->fp_lock_level;
 161                 if (level != LKM_IVMODE)
 162                         user_dlm_cluster_unlock(&ip->ip_lockres, level);
 163
 164                 kfree(fp);
 165                 file->private_data = NULL;
 166         }
 167
 168         return 0;
 169 }
 170
 171 static ssize_t dlmfs_file_read(struct file *filp,
 172                                char __user *buf,
 173                                size_t count,
 174                                loff_t *ppos)
 175 {
 176         int bytes_left;
 177         ssize_t readlen;
 178         char *lvb_buf;
 179         struct inode *inode = filp->f_path.dentry->d_inode;
 180
 181         mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
 182                 inode->i_ino, count, *ppos);
 183
 184         if (*ppos >= i_size_read(inode))
 185                 return 0;
 186
 187         if (!count)
 188                 return 0;
 189
 190         if (!access_ok(VERIFY_WRITE, buf, count))
 191                 return -EFAULT;
 192
 193         /* don't read past the lvb */
 194         if ((count + *ppos) > i_size_read(inode))
 195                 readlen = i_size_read(inode) - *ppos;
 196         else
 197                 readlen = count - *ppos;
 198
 199         lvb_buf = kmalloc(readlen, GFP_NOFS);
 200         if (!lvb_buf)
 201                 return -ENOMEM;
 202
 203         user_dlm_read_lvb(inode, lvb_buf, readlen);
 204         bytes_left = __copy_to_user(buf, lvb_buf, readlen);
 205         readlen -= bytes_left;
 206
 207         kfree(lvb_buf);
 208
 209         *ppos = *ppos + readlen;
 210
 211         mlog(0, "read %zd bytes\n", readlen);
 212         return readlen;
 213 }
 214
 215 static ssize_t dlmfs_file_write(struct file *filp,
 216                                 const char __user *buf,
 217                                 size_t count,
 218                                 loff_t *ppos)
 219 {
 220         int bytes_left;
 221         ssize_t writelen;
 222         char *lvb_buf;
 223         struct inode *inode = filp->f_path.dentry->d_inode;
 224
 225         mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
 226                 inode->i_ino, count, *ppos);
 227
 228         if (*ppos >= i_size_read(inode))
 229                 return -ENOSPC;
 230
 231         if (!count)
 232                 return 0;
 233
 234         if (!access_ok(VERIFY_READ, buf, count))
 235                 return -EFAULT;
 236
 237         /* don't write past the lvb */
 238         if ((count + *ppos) > i_size_read(inode))
 239                 writelen = i_size_read(inode) - *ppos;
 240         else
 241                 writelen = count - *ppos;
 242
 243         lvb_buf = kmalloc(writelen, GFP_NOFS);
 244         if (!lvb_buf)
 245                 return -ENOMEM;
 246
 247         bytes_left = copy_from_user(lvb_buf, buf, writelen);
 248         writelen -= bytes_left;
 249         if (writelen)
 250                 user_dlm_write_lvb(inode, lvb_buf, writelen);
 251
 252         kfree(lvb_buf);
 253
 254         *ppos = *ppos + writelen;
 255         mlog(0, "wrote %zd bytes\n", writelen);
 256         return writelen;
 257 }
 258
 259 static void dlmfs_init_once(void *foo,
 260                             struct kmem_cache *cachep,
 261                             unsigned long flags)
 262 {
 263         struct dlmfs_inode_private *ip =
 264                 (struct dlmfs_inode_private *) foo;
 265
 266         if (flags & SLAB_CTOR_CONSTRUCTOR) {
 267                 ip->ip_dlm = NULL;
 268                 ip->ip_parent = NULL;
 269
 270                 inode_init_once(&ip->ip_vfs_inode);
 271         }
 272 }
 273
 274 static struct inode *dlmfs_alloc_inode(struct super_block *sb)
 275 {
 276         struct dlmfs_inode_private *ip;
 277
 278         ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS);
 279         if (!ip)
 280                 return NULL;
 281
 282         return &ip->ip_vfs_inode;
 283 }
 284
 285 static void dlmfs_destroy_inode(struct inode *inode)
 286 {
 287         kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode));
 288 }
 289
 290 static void dlmfs_clear_inode(struct inode *inode)
 291 {
 292         int status;
 293         struct dlmfs_inode_private *ip;
 294
 295         if (!inode)
 296                 return;
 297
 298         mlog(0, "inode %lu\n", inode->i_ino);
 299
 300         ip = DLMFS_I(inode);
 301
 302         if (S_ISREG(inode->i_mode)) {
 303                 status = user_dlm_destroy_lock(&ip->ip_lockres);
 304                 if (status < 0)
 305                         mlog_errno(status);
 306                 iput(ip->ip_parent);
 307                 goto clear_fields;
 308         }
 309
 310         mlog(0, "we're a directory, ip->ip_dlm = 0x%p\n", ip->ip_dlm);
 311         /* we must be a directory. If required, lets unregister the
 312          * dlm context now. */
 313         if (ip->ip_dlm)
 314                 user_dlm_unregister_context(ip->ip_dlm);
 315 clear_fields:
 316         ip->ip_parent = NULL;
 317         ip->ip_dlm = NULL;
 318 }
 319
 320 static struct backing_dev_info dlmfs_backing_dev_info = {
 321         .ra_pages       = 0,    /* No readahead */
 322         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
 323 };
 324
 325 static struct inode *dlmfs_get_root_inode(struct super_block *sb)
 326 {
 327         struct inode *inode = new_inode(sb);
 328         int mode = S_IFDIR | 0755;
 329         struct dlmfs_inode_private *ip;
 330
 331         if (inode) {
 332                 ip = DLMFS_I(inode);
 333
 334                 inode->i_mode = mode;
 335                 inode->i_uid = current->fsuid;
 336                 inode->i_gid = current->fsgid;
 337                 inode->i_blocks = 0;
 338                 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
 339                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 340                 inc_nlink(inode);
 341
 342                 inode->i_fop = &simple_dir_operations;
 343                 inode->i_op = &dlmfs_root_inode_operations;
 344         }
 345
 346         return inode;
 347 }
 348
 349 static struct inode *dlmfs_get_inode(struct inode *parent,
 350                                      struct dentry *dentry,
 351                                      int mode)
 352 {
 353         struct super_block *sb = parent->i_sb;
 354         struct inode * inode = new_inode(sb);
 355         struct dlmfs_inode_private *ip;
 356
 357         if (!inode)
 358                 return NULL;
 359
 360         inode->i_mode = mode;
 361         inode->i_uid = current->fsuid;
 362         inode->i_gid = current->fsgid;
 363         inode->i_blocks = 0;
 364         inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
 365         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 366
 367         ip = DLMFS_I(inode);
 368         ip->ip_dlm = DLMFS_I(parent)->ip_dlm;
 369
 370         switch (mode & S_IFMT) {
 371         default:
 372                 /* for now we don't support anything other than
 373                  * directories and regular files. */
 374                 BUG();
 375                 break;
 376         case S_IFREG:
 377                 inode->i_op = &dlmfs_file_inode_operations;
 378                 inode->i_fop = &dlmfs_file_operations;
 379
 380                 i_size_write(inode,  DLM_LVB_LEN);
 381
 382                 user_dlm_lock_res_init(&ip->ip_lockres, dentry);
 383
 384                 /* released at clear_inode time, this insures that we
 385                  * get to drop the dlm reference on each lock *before*
 386                  * we call the unregister code for releasing parent
 387                  * directories. */
 388                 ip->ip_parent = igrab(parent);
 389                 BUG_ON(!ip->ip_parent);
 390                 break;
 391         case S_IFDIR:
 392                 inode->i_op = &dlmfs_dir_inode_operations;
 393                 inode->i_fop = &simple_dir_operations;
 394
 395                 /* directory inodes start off with i_nlink ==
 396                  * 2 (for "." entry) */
 397                 inc_nlink(inode);
 398                 break;
 399         }
 400
 401         if (parent->i_mode & S_ISGID) {
 402                 inode->i_gid = parent->i_gid;
 403                 if (S_ISDIR(mode))
 404                         inode->i_mode |= S_ISGID;
 405         }
 406
 407         return inode;
 408 }
 409
 410 /*
 411  * File creation. Allocate an inode, and we're done..
 412  */
 413 /* SMP-safe */
 414 static int dlmfs_mkdir(struct inode * dir,
 415                        struct dentry * dentry,
 416                        int mode)
 417 {
 418         int status;
 419         struct inode *inode = NULL;
 420         struct qstr *domain = &dentry->d_name;
 421         struct dlmfs_inode_private *ip;
 422         struct dlm_ctxt *dlm;
 423
 424         mlog(0, "mkdir %.*s\n", domain->len, domain->name);
 425
 426         /* verify that we have a proper domain */
 427         if (domain->len >= O2NM_MAX_NAME_LEN) {
 428                 status = -EINVAL;
 429                 mlog(ML_ERROR, "invalid domain name for directory.\n");
 430                 goto bail;
 431         }
 432
 433         inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR);
 434         if (!inode) {
 435                 status = -ENOMEM;
 436                 mlog_errno(status);
 437                 goto bail;
 438         }
 439
 440         ip = DLMFS_I(inode);
 441
 442         dlm = user_dlm_register_context(domain);
 443         if (IS_ERR(dlm)) {
 444                 status = PTR_ERR(dlm);
 445                 mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n",
 446                      status, domain->len, domain->name);
 447                 goto bail;
 448         }
 449         ip->ip_dlm = dlm;
 450
 451         inc_nlink(dir);
 452         d_instantiate(dentry, inode);
 453         dget(dentry);   /* Extra count - pin the dentry in core */
 454
 455         status = 0;
 456 bail:
 457         if (status < 0)
 458                 iput(inode);
 459         return status;
 460 }
 461
 462 static int dlmfs_create(struct inode *dir,
 463                         struct dentry *dentry,
 464                         int mode,
 465                         struct nameidata *nd)
 466 {
 467         int status = 0;
 468         struct inode *inode;
 469         struct qstr *name = &dentry->d_name;
 470
 471         mlog(0, "create %.*s\n", name->len, name->name);
 472
 473         /* verify name is valid and doesn't contain any dlm reserved
 474          * characters */
 475         if (name->len >= USER_DLM_LOCK_ID_MAX_LEN ||
 476             name->name[0] == '$') {
 477                 status = -EINVAL;
 478                 mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len,
 479                      name->name);
 480                 goto bail;
 481         }
 482
 483         inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG);
 484         if (!inode) {
 485                 status = -ENOMEM;
 486                 mlog_errno(status);
 487                 goto bail;
 488         }
 489
 490         d_instantiate(dentry, inode);
 491         dget(dentry);   /* Extra count - pin the dentry in core */
 492 bail:
 493         return status;
 494 }
 495
 496 static int dlmfs_unlink(struct inode *dir,
 497                         struct dentry *dentry)
 498 {
 499         int status;
 500         struct inode *inode = dentry->d_inode;
 501
 502         mlog(0, "unlink inode %lu\n", inode->i_ino);
 503
 504         /* if there are no current holders, or none that are waiting
 505          * to acquire a lock, this basically destroys our lockres. */
 506         status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres);
 507         if (status < 0) {
 508                 mlog(ML_ERROR, "unlink %.*s, error %d from destroy\n",
 509                      dentry->d_name.len, dentry->d_name.name, status);
 510                 goto bail;
 511         }
 512         status = simple_unlink(dir, dentry);
 513 bail:
 514         return status;
 515 }
 516
 517 static int dlmfs_fill_super(struct super_block * sb,
 518                             void * data,
 519                             int silent)
 520 {
 521         struct inode * inode;
 522         struct dentry * root;
 523
 524         sb->s_maxbytes = MAX_LFS_FILESIZE;
 525         sb->s_blocksize = PAGE_CACHE_SIZE;
 526         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 527         sb->s_magic = DLMFS_MAGIC;
 528         sb->s_op = &dlmfs_ops;
 529         inode = dlmfs_get_root_inode(sb);
 530         if (!inode)
 531                 return -ENOMEM;
 532
 533         root = d_alloc_root(inode);
 534         if (!root) {
 535                 iput(inode);
 536                 return -ENOMEM;
 537         }
 538         sb->s_root = root;
 539         return 0;
 540 }
 541
 542 static const struct file_operations dlmfs_file_operations = {
 543         .open           = dlmfs_file_open,
 544         .release        = dlmfs_file_release,
 545         .read           = dlmfs_file_read,
 546         .write          = dlmfs_file_write,
 547 };
 548
 549 static const struct inode_operations dlmfs_dir_inode_operations = {
 550         .create         = dlmfs_create,
 551         .lookup         = simple_lookup,
 552         .unlink         = dlmfs_unlink,
 553 };
 554
 555 /* this way we can restrict mkdir to only the toplevel of the fs. */
 556 static const struct inode_operations dlmfs_root_inode_operations = {
 557         .lookup         = simple_lookup,
 558         .mkdir          = dlmfs_mkdir,
 559         .rmdir          = simple_rmdir,
 560 };
 561
 562 static const struct super_operations dlmfs_ops = {
 563         .statfs         = simple_statfs,
 564         .alloc_inode    = dlmfs_alloc_inode,
 565         .destroy_inode  = dlmfs_destroy_inode,
 566         .clear_inode    = dlmfs_clear_inode,
 567         .drop_inode     = generic_delete_inode,
 568 };
 569
 570 static const struct inode_operations dlmfs_file_inode_operations = {
 571         .getattr        = simple_getattr,
 572 };
 573
 574 static int dlmfs_get_sb(struct file_system_type *fs_type,
 575         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
 576 {
 577         return get_sb_nodev(fs_type, flags, data, dlmfs_fill_super, mnt);
 578 }
 579
 580 static struct file_system_type dlmfs_fs_type = {
 581         .owner          = THIS_MODULE,
 582         .name           = "ocfs2_dlmfs",
 583         .get_sb         = dlmfs_get_sb,
 584         .kill_sb        = kill_litter_super,
 585 };
 586
 587 static int __init init_dlmfs_fs(void)
 588 {
 589         int status;
 590         int cleanup_inode = 0, cleanup_worker = 0;
 591
 592         dlmfs_print_version();
 593
 594         dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache",
 595                                 sizeof(struct dlmfs_inode_private),
 596                                 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 597                                         SLAB_MEM_SPREAD),
 598                                 dlmfs_init_once, NULL);
 599         if (!dlmfs_inode_cache)
 600                 return -ENOMEM;
 601         cleanup_inode = 1;
 602
 603         user_dlm_worker = create_singlethread_workqueue("user_dlm");
 604         if (!user_dlm_worker) {
 605                 status = -ENOMEM;
 606                 goto bail;
 607         }
 608         cleanup_worker = 1;
 609
 610         status = register_filesystem(&dlmfs_fs_type);
 611 bail:
 612         if (status) {
 613                 if (cleanup_inode)
 614                         kmem_cache_destroy(dlmfs_inode_cache);
 615                 if (cleanup_worker)
 616                         destroy_workqueue(user_dlm_worker);
 617         } else
 618                 printk("OCFS2 User DLM kernel interface loaded\n");
 619         return status;
 620 }
 621
 622 static void __exit exit_dlmfs_fs(void)
 623 {
 624         unregister_filesystem(&dlmfs_fs_type);
 625
 626         flush_workqueue(user_dlm_worker);
 627         destroy_workqueue(user_dlm_worker);
 628
 629         kmem_cache_destroy(dlmfs_inode_cache);
 630 }
 631
 632 MODULE_AUTHOR("Oracle");
 633 MODULE_LICENSE("GPL");
 634
 635 module_init(init_dlmfs_fs)
 636 module_exit(exit_dlmfs_fs)