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/backing-dev.h>
  46
  47 #include <asm/uaccess.h>
  48
  49
  50 #include "cluster/nodemanager.h"
  51 #include "cluster/heartbeat.h"
  52 #include "cluster/tcp.h"
  53
  54 #include "dlmapi.h"
  55
  56 #include "userdlm.h"
  57
  58 #include "dlmfsver.h"
  59
  60 #define MLOG_MASK_PREFIX ML_DLMFS
  61 #include "cluster/masklog.h"
  62
  63 #include "ocfs2_lockingver.h"
  64
  65 static const struct super_operations dlmfs_ops;
  66 static const struct file_operations dlmfs_file_operations;
  67 static const struct inode_operations dlmfs_dir_inode_operations;
  68 static const struct inode_operations dlmfs_root_inode_operations;
  69 static const struct inode_operations dlmfs_file_inode_operations;
  70 static struct kmem_cache *dlmfs_inode_cache;
  71
  72 struct workqueue_struct *user_dlm_worker;
  73
  74 /*
  75  * This is the userdlmfs locking protocol version.
  76  *
  77  * See fs/ocfs2/dlmglue.c for more details on locking versions.
  78  */
  79 static const struct dlm_protocol_version user_locking_protocol = {
  80         .pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR,
  81         .pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR,
  82 };
  83
  84 /*
  85  * decodes a set of open flags into a valid lock level and a set of flags.
  86  * returns < 0 if we have invalid flags
  87  * flags which mean something to us:
  88  * O_RDONLY -> PRMODE level
  89  * O_WRONLY -> EXMODE level
  90  *
  91  * O_NONBLOCK -> LKM_NOQUEUE
  92  */
  93 static int dlmfs_decode_open_flags(int open_flags,
  94                                    int *level,
  95                                    int *flags)
  96 {
  97         if (open_flags & (O_WRONLY|O_RDWR))
  98                 *level = LKM_EXMODE;
  99         else
 100                 *level = LKM_PRMODE;
 101
 102         *flags = 0;
 103         if (open_flags & O_NONBLOCK)
 104                 *flags |= LKM_NOQUEUE;
 105
 106         return 0;
 107 }
 108
 109 static int dlmfs_file_open(struct inode *inode,
 110                            struct file *file)
 111 {
 112         int status, level, flags;
 113         struct dlmfs_filp_private *fp = NULL;
 114         struct dlmfs_inode_private *ip;
 115
 116         if (S_ISDIR(inode->i_mode))
 117                 BUG();
 118
 119         mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino,
 120                 file->f_flags);
 121
 122         status = dlmfs_decode_open_flags(file->f_flags, &level, &flags);
 123         if (status < 0)
 124                 goto bail;
 125
 126         /* We don't want to honor O_APPEND at read/write time as it
 127          * doesn't make sense for LVB writes. */
 128         file->f_flags &= ~O_APPEND;
 129
 130         fp = kmalloc(sizeof(*fp), GFP_NOFS);
 131         if (!fp) {
 132                 status = -ENOMEM;
 133                 goto bail;
 134         }
 135         fp->fp_lock_level = level;
 136
 137         ip = DLMFS_I(inode);
 138
 139         status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags);
 140         if (status < 0) {
 141                 /* this is a strange error to return here but I want
 142                  * to be able userspace to be able to distinguish a
 143                  * valid lock request from one that simply couldn't be
 144                  * granted. */
 145                 if (flags & LKM_NOQUEUE && status == -EAGAIN)
 146                         status = -ETXTBSY;
 147                 kfree(fp);
 148                 goto bail;
 149         }
 150
 151         file->private_data = fp;
 152 bail:
 153         return status;
 154 }
 155
 156 static int dlmfs_file_release(struct inode *inode,
 157                               struct file *file)
 158 {
 159         int level, status;
 160         struct dlmfs_inode_private *ip = DLMFS_I(inode);
 161         struct dlmfs_filp_private *fp =
 162                 (struct dlmfs_filp_private *) file->private_data;
 163
 164         if (S_ISDIR(inode->i_mode))
 165                 BUG();
 166
 167         mlog(0, "close called on inode %lu\n", inode->i_ino);
 168
 169         status = 0;
 170         if (fp) {
 171                 level = fp->fp_lock_level;
 172                 if (level != LKM_IVMODE)
 173                         user_dlm_cluster_unlock(&ip->ip_lockres, level);
 174
 175                 kfree(fp);
 176                 file->private_data = NULL;
 177         }
 178
 179         return 0;
 180 }
 181
 182 static ssize_t dlmfs_file_read(struct file *filp,
 183                                char __user *buf,
 184                                size_t count,
 185                                loff_t *ppos)
 186 {
 187         int bytes_left;
 188         ssize_t readlen;
 189         char *lvb_buf;
 190         struct inode *inode = filp->f_path.dentry->d_inode;
 191
 192         mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
 193                 inode->i_ino, count, *ppos);
 194
 195         if (*ppos >= i_size_read(inode))
 196                 return 0;
 197
 198         if (!count)
 199                 return 0;
 200
 201         if (!access_ok(VERIFY_WRITE, buf, count))
 202                 return -EFAULT;
 203
 204         /* don't read past the lvb */
 205         if ((count + *ppos) > i_size_read(inode))
 206                 readlen = i_size_read(inode) - *ppos;
 207         else
 208                 readlen = count - *ppos;
 209
 210         lvb_buf = kmalloc(readlen, GFP_NOFS);
 211         if (!lvb_buf)
 212                 return -ENOMEM;
 213
 214         user_dlm_read_lvb(inode, lvb_buf, readlen);
 215         bytes_left = __copy_to_user(buf, lvb_buf, readlen);
 216         readlen -= bytes_left;
 217
 218         kfree(lvb_buf);
 219
 220         *ppos = *ppos + readlen;
 221
 222         mlog(0, "read %zd bytes\n", readlen);
 223         return readlen;
 224 }
 225
 226 static ssize_t dlmfs_file_write(struct file *filp,
 227                                 const char __user *buf,
 228                                 size_t count,
 229                                 loff_t *ppos)
 230 {
 231         int bytes_left;
 232         ssize_t writelen;
 233         char *lvb_buf;
 234         struct inode *inode = filp->f_path.dentry->d_inode;
 235
 236         mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
 237                 inode->i_ino, count, *ppos);
 238
 239         if (*ppos >= i_size_read(inode))
 240                 return -ENOSPC;
 241
 242         if (!count)
 243                 return 0;
 244
 245         if (!access_ok(VERIFY_READ, buf, count))
 246                 return -EFAULT;
 247
 248         /* don't write past the lvb */
 249         if ((count + *ppos) > i_size_read(inode))
 250                 writelen = i_size_read(inode) - *ppos;
 251         else
 252                 writelen = count - *ppos;
 253
 254         lvb_buf = kmalloc(writelen, GFP_NOFS);
 255         if (!lvb_buf)
 256                 return -ENOMEM;
 257
 258         bytes_left = copy_from_user(lvb_buf, buf, writelen);
 259         writelen -= bytes_left;
 260         if (writelen)
 261                 user_dlm_write_lvb(inode, lvb_buf, writelen);
 262
 263         kfree(lvb_buf);
 264
 265         *ppos = *ppos + writelen;
 266         mlog(0, "wrote %zd bytes\n", writelen);
 267         return writelen;
 268 }
 269
 270 static void dlmfs_init_once(void *foo)
 271 {
 272         struct dlmfs_inode_private *ip =
 273                 (struct dlmfs_inode_private *) foo;
 274
 275         ip->ip_dlm = NULL;
 276         ip->ip_parent = NULL;
 277
 278         inode_init_once(&ip->ip_vfs_inode);
 279 }
 280
 281 static struct inode *dlmfs_alloc_inode(struct super_block *sb)
 282 {
 283         struct dlmfs_inode_private *ip;
 284
 285         ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS);
 286         if (!ip)
 287                 return NULL;
 288
 289         return &ip->ip_vfs_inode;
 290 }
 291
 292 static void dlmfs_destroy_inode(struct inode *inode)
 293 {
 294         kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode));
 295 }
 296
 297 static void dlmfs_clear_inode(struct inode *inode)
 298 {
 299         int status;
 300         struct dlmfs_inode_private *ip;
 301
 302         if (!inode)
 303                 return;
 304
 305         mlog(0, "inode %lu\n", inode->i_ino);
 306
 307         ip = DLMFS_I(inode);
 308
 309         if (S_ISREG(inode->i_mode)) {
 310                 status = user_dlm_destroy_lock(&ip->ip_lockres);
 311                 if (status < 0)
 312                         mlog_errno(status);
 313                 iput(ip->ip_parent);
 314                 goto clear_fields;
 315         }
 316
 317         mlog(0, "we're a directory, ip->ip_dlm = 0x%p\n", ip->ip_dlm);
 318         /* we must be a directory. If required, lets unregister the
 319          * dlm context now. */
 320         if (ip->ip_dlm)
 321                 user_dlm_unregister_context(ip->ip_dlm);
 322 clear_fields:
 323         ip->ip_parent = NULL;
 324         ip->ip_dlm = NULL;
 325 }
 326
 327 static struct backing_dev_info dlmfs_backing_dev_info = {
 328         .ra_pages       = 0,    /* No readahead */
 329         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
 330 };
 331
 332 static struct inode *dlmfs_get_root_inode(struct super_block *sb)
 333 {
 334         struct inode *inode = new_inode(sb);
 335         int mode = S_IFDIR | 0755;
 336         struct dlmfs_inode_private *ip;
 337
 338         if (inode) {
 339                 ip = DLMFS_I(inode);
 340
 341                 inode->i_mode = mode;
 342                 inode->i_uid = current->fsuid;
 343                 inode->i_gid = current->fsgid;
 344                 inode->i_blocks = 0;
 345                 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
 346                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 347                 inc_nlink(inode);
 348
 349                 inode->i_fop = &simple_dir_operations;
 350                 inode->i_op = &dlmfs_root_inode_operations;
 351         }
 352
 353         return inode;
 354 }
 355
 356 static struct inode *dlmfs_get_inode(struct inode *parent,
 357                                      struct dentry *dentry,
 358                                      int mode)
 359 {
 360         struct super_block *sb = parent->i_sb;
 361         struct inode * inode = new_inode(sb);
 362         struct dlmfs_inode_private *ip;
 363
 364         if (!inode)
 365                 return NULL;
 366
 367         inode->i_mode = mode;
 368         inode->i_uid = current->fsuid;
 369         inode->i_gid = current->fsgid;
 370         inode->i_blocks = 0;
 371         inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
 372         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 373
 374         ip = DLMFS_I(inode);
 375         ip->ip_dlm = DLMFS_I(parent)->ip_dlm;
 376
 377         switch (mode & S_IFMT) {
 378         default:
 379                 /* for now we don't support anything other than
 380                  * directories and regular files. */
 381                 BUG();
 382                 break;
 383         case S_IFREG:
 384                 inode->i_op = &dlmfs_file_inode_operations;
 385                 inode->i_fop = &dlmfs_file_operations;
 386
 387                 i_size_write(inode,  DLM_LVB_LEN);
 388
 389                 user_dlm_lock_res_init(&ip->ip_lockres, dentry);
 390
 391                 /* released at clear_inode time, this insures that we
 392                  * get to drop the dlm reference on each lock *before*
 393                  * we call the unregister code for releasing parent
 394                  * directories. */
 395                 ip->ip_parent = igrab(parent);
 396                 BUG_ON(!ip->ip_parent);
 397                 break;
 398         case S_IFDIR:
 399                 inode->i_op = &dlmfs_dir_inode_operations;
 400                 inode->i_fop = &simple_dir_operations;
 401
 402                 /* directory inodes start off with i_nlink ==
 403                  * 2 (for "." entry) */
 404                 inc_nlink(inode);
 405                 break;
 406         }
 407
 408         if (parent->i_mode & S_ISGID) {
 409                 inode->i_gid = parent->i_gid;
 410                 if (S_ISDIR(mode))
 411                         inode->i_mode |= S_ISGID;
 412         }
 413
 414         return inode;
 415 }
 416
 417 /*
 418  * File creation. Allocate an inode, and we're done..
 419  */
 420 /* SMP-safe */
 421 static int dlmfs_mkdir(struct inode * dir,
 422                        struct dentry * dentry,
 423                        int mode)
 424 {
 425         int status;
 426         struct inode *inode = NULL;
 427         struct qstr *domain = &dentry->d_name;
 428         struct dlmfs_inode_private *ip;
 429         struct dlm_ctxt *dlm;
 430         struct dlm_protocol_version proto = user_locking_protocol;
 431
 432         mlog(0, "mkdir %.*s\n", domain->len, domain->name);
 433
 434         /* verify that we have a proper domain */
 435         if (domain->len >= O2NM_MAX_NAME_LEN) {
 436                 status = -EINVAL;
 437                 mlog(ML_ERROR, "invalid domain name for directory.\n");
 438                 goto bail;
 439         }
 440
 441         inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR);
 442         if (!inode) {
 443                 status = -ENOMEM;
 444                 mlog_errno(status);
 445                 goto bail;
 446         }
 447
 448         ip = DLMFS_I(inode);
 449
 450         dlm = user_dlm_register_context(domain, &proto);
 451         if (IS_ERR(dlm)) {
 452                 status = PTR_ERR(dlm);
 453                 mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n",
 454                      status, domain->len, domain->name);
 455                 goto bail;
 456         }
 457         ip->ip_dlm = dlm;
 458
 459         inc_nlink(dir);
 460         d_instantiate(dentry, inode);
 461         dget(dentry);   /* Extra count - pin the dentry in core */
 462
 463         status = 0;
 464 bail:
 465         if (status < 0)
 466                 iput(inode);
 467         return status;
 468 }
 469
 470 static int dlmfs_create(struct inode *dir,
 471                         struct dentry *dentry,
 472                         int mode,
 473                         struct nameidata *nd)
 474 {
 475         int status = 0;
 476         struct inode *inode;
 477         struct qstr *name = &dentry->d_name;
 478
 479         mlog(0, "create %.*s\n", name->len, name->name);
 480
 481         /* verify name is valid and doesn't contain any dlm reserved
 482          * characters */
 483         if (name->len >= USER_DLM_LOCK_ID_MAX_LEN ||
 484             name->name[0] == '$') {
 485                 status = -EINVAL;
 486                 mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len,
 487                      name->name);
 488                 goto bail;
 489         }
 490
 491         inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG);
 492         if (!inode) {
 493                 status = -ENOMEM;
 494                 mlog_errno(status);
 495                 goto bail;
 496         }
 497
 498         d_instantiate(dentry, inode);
 499         dget(dentry);   /* Extra count - pin the dentry in core */
 500 bail:
 501         return status;
 502 }
 503
 504 static int dlmfs_unlink(struct inode *dir,
 505                         struct dentry *dentry)
 506 {
 507         int status;
 508         struct inode *inode = dentry->d_inode;
 509
 510         mlog(0, "unlink inode %lu\n", inode->i_ino);
 511
 512         /* if there are no current holders, or none that are waiting
 513          * to acquire a lock, this basically destroys our lockres. */
 514         status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres);
 515         if (status < 0) {
 516                 mlog(ML_ERROR, "unlink %.*s, error %d from destroy\n",
 517                      dentry->d_name.len, dentry->d_name.name, status);
 518                 goto bail;
 519         }
 520         status = simple_unlink(dir, dentry);
 521 bail:
 522         return status;
 523 }
 524
 525 static int dlmfs_fill_super(struct super_block * sb,
 526                             void * data,
 527                             int silent)
 528 {
 529         struct inode * inode;
 530         struct dentry * root;
 531
 532         sb->s_maxbytes = MAX_LFS_FILESIZE;
 533         sb->s_blocksize = PAGE_CACHE_SIZE;
 534         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 535         sb->s_magic = DLMFS_MAGIC;
 536         sb->s_op = &dlmfs_ops;
 537         inode = dlmfs_get_root_inode(sb);
 538         if (!inode)
 539                 return -ENOMEM;
 540
 541         root = d_alloc_root(inode);
 542         if (!root) {
 543                 iput(inode);
 544                 return -ENOMEM;
 545         }
 546         sb->s_root = root;
 547         return 0;
 548 }
 549
 550 static const struct file_operations dlmfs_file_operations = {
 551         .open           = dlmfs_file_open,
 552         .release        = dlmfs_file_release,
 553         .read           = dlmfs_file_read,
 554         .write          = dlmfs_file_write,
 555 };
 556
 557 static const struct inode_operations dlmfs_dir_inode_operations = {
 558         .create         = dlmfs_create,
 559         .lookup         = simple_lookup,
 560         .unlink         = dlmfs_unlink,
 561 };
 562
 563 /* this way we can restrict mkdir to only the toplevel of the fs. */
 564 static const struct inode_operations dlmfs_root_inode_operations = {
 565         .lookup         = simple_lookup,
 566         .mkdir          = dlmfs_mkdir,
 567         .rmdir          = simple_rmdir,
 568 };
 569
 570 static const struct super_operations dlmfs_ops = {
 571         .statfs         = simple_statfs,
 572         .alloc_inode    = dlmfs_alloc_inode,
 573         .destroy_inode  = dlmfs_destroy_inode,
 574         .clear_inode    = dlmfs_clear_inode,
 575         .drop_inode     = generic_delete_inode,
 576 };
 577
 578 static const struct inode_operations dlmfs_file_inode_operations = {
 579         .getattr        = simple_getattr,
 580 };
 581
 582 static int dlmfs_get_sb(struct file_system_type *fs_type,
 583         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
 584 {
 585         return get_sb_nodev(fs_type, flags, data, dlmfs_fill_super, mnt);
 586 }
 587
 588 static struct file_system_type dlmfs_fs_type = {
 589         .owner          = THIS_MODULE,
 590         .name           = "ocfs2_dlmfs",
 591         .get_sb         = dlmfs_get_sb,
 592         .kill_sb        = kill_litter_super,
 593 };
 594
 595 static int __init init_dlmfs_fs(void)
 596 {
 597         int status;
 598         int cleanup_inode = 0, cleanup_worker = 0;
 599
 600         dlmfs_print_version();
 601
 602         status = bdi_init(&dlmfs_backing_dev_info);
 603         if (status)
 604                 return status;
 605
 606         dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache",
 607                                 sizeof(struct dlmfs_inode_private),
 608                                 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 609                                         SLAB_MEM_SPREAD),
 610                                 dlmfs_init_once);
 611         if (!dlmfs_inode_cache) {
 612                 status = -ENOMEM;
 613                 goto bail;
 614         }
 615         cleanup_inode = 1;
 616
 617         user_dlm_worker = create_singlethread_workqueue("user_dlm");
 618         if (!user_dlm_worker) {
 619                 status = -ENOMEM;
 620                 goto bail;
 621         }
 622         cleanup_worker = 1;
 623
 624         status = register_filesystem(&dlmfs_fs_type);
 625 bail:
 626         if (status) {
 627                 if (cleanup_inode)
 628                         kmem_cache_destroy(dlmfs_inode_cache);
 629                 if (cleanup_worker)
 630                         destroy_workqueue(user_dlm_worker);
 631                 bdi_destroy(&dlmfs_backing_dev_info);
 632         } else
 633                 printk("OCFS2 User DLM kernel interface loaded\n");
 634         return status;
 635 }
 636
 637 static void __exit exit_dlmfs_fs(void)
 638 {
 639         unregister_filesystem(&dlmfs_fs_type);
 640
 641         flush_workqueue(user_dlm_worker);
 642         destroy_workqueue(user_dlm_worker);
 643
 644         kmem_cache_destroy(dlmfs_inode_cache);
 645
 646         bdi_destroy(&dlmfs_backing_dev_info);
 647 }
 648
 649 MODULE_AUTHOR("Oracle");
 650 MODULE_LICENSE("GPL");
 651
 652 module_init(init_dlmfs_fs)
 653 module_exit(exit_dlmfs_fs)