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(struct kmem_cache *cachep,
 271                             void *foo)
 272 {
 273         struct dlmfs_inode_private *ip =
 274                 (struct dlmfs_inode_private *) foo;
 275
 276         ip->ip_dlm = NULL;
 277         ip->ip_parent = NULL;
 278
 279         inode_init_once(&ip->ip_vfs_inode);
 280 }
 281
 282 static struct inode *dlmfs_alloc_inode(struct super_block *sb)
 283 {
 284         struct dlmfs_inode_private *ip;
 285
 286         ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS);
 287         if (!ip)
 288                 return NULL;
 289
 290         return &ip->ip_vfs_inode;
 291 }
 292
 293 static void dlmfs_destroy_inode(struct inode *inode)
 294 {
 295         kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode));
 296 }
 297
 298 static void dlmfs_clear_inode(struct inode *inode)
 299 {
 300         int status;
 301         struct dlmfs_inode_private *ip;
 302
 303         if (!inode)
 304                 return;
 305
 306         mlog(0, "inode %lu\n", inode->i_ino);
 307
 308         ip = DLMFS_I(inode);
 309
 310         if (S_ISREG(inode->i_mode)) {
 311                 status = user_dlm_destroy_lock(&ip->ip_lockres);
 312                 if (status < 0)
 313                         mlog_errno(status);
 314                 iput(ip->ip_parent);
 315                 goto clear_fields;
 316         }
 317
 318         mlog(0, "we're a directory, ip->ip_dlm = 0x%p\n", ip->ip_dlm);
 319         /* we must be a directory. If required, lets unregister the
 320          * dlm context now. */
 321         if (ip->ip_dlm)
 322                 user_dlm_unregister_context(ip->ip_dlm);
 323 clear_fields:
 324         ip->ip_parent = NULL;
 325         ip->ip_dlm = NULL;
 326 }
 327
 328 static struct backing_dev_info dlmfs_backing_dev_info = {
 329         .ra_pages       = 0,    /* No readahead */
 330         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
 331 };
 332
 333 static struct inode *dlmfs_get_root_inode(struct super_block *sb)
 334 {
 335         struct inode *inode = new_inode(sb);
 336         int mode = S_IFDIR | 0755;
 337         struct dlmfs_inode_private *ip;
 338
 339         if (inode) {
 340                 ip = DLMFS_I(inode);
 341
 342                 inode->i_mode = mode;
 343                 inode->i_uid = current->fsuid;
 344                 inode->i_gid = current->fsgid;
 345                 inode->i_blocks = 0;
 346                 inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
 347                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 348                 inc_nlink(inode);
 349
 350                 inode->i_fop = &simple_dir_operations;
 351                 inode->i_op = &dlmfs_root_inode_operations;
 352         }
 353
 354         return inode;
 355 }
 356
 357 static struct inode *dlmfs_get_inode(struct inode *parent,
 358                                      struct dentry *dentry,
 359                                      int mode)
 360 {
 361         struct super_block *sb = parent->i_sb;
 362         struct inode * inode = new_inode(sb);
 363         struct dlmfs_inode_private *ip;
 364
 365         if (!inode)
 366                 return NULL;
 367
 368         inode->i_mode = mode;
 369         inode->i_uid = current->fsuid;
 370         inode->i_gid = current->fsgid;
 371         inode->i_blocks = 0;
 372         inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
 373         inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 374
 375         ip = DLMFS_I(inode);
 376         ip->ip_dlm = DLMFS_I(parent)->ip_dlm;
 377
 378         switch (mode & S_IFMT) {
 379         default:
 380                 /* for now we don't support anything other than
 381                  * directories and regular files. */
 382                 BUG();
 383                 break;
 384         case S_IFREG:
 385                 inode->i_op = &dlmfs_file_inode_operations;
 386                 inode->i_fop = &dlmfs_file_operations;
 387
 388                 i_size_write(inode,  DLM_LVB_LEN);
 389
 390                 user_dlm_lock_res_init(&ip->ip_lockres, dentry);
 391
 392                 /* released at clear_inode time, this insures that we
 393                  * get to drop the dlm reference on each lock *before*
 394                  * we call the unregister code for releasing parent
 395                  * directories. */
 396                 ip->ip_parent = igrab(parent);
 397                 BUG_ON(!ip->ip_parent);
 398                 break;
 399         case S_IFDIR:
 400                 inode->i_op = &dlmfs_dir_inode_operations;
 401                 inode->i_fop = &simple_dir_operations;
 402
 403                 /* directory inodes start off with i_nlink ==
 404                  * 2 (for "." entry) */
 405                 inc_nlink(inode);
 406                 break;
 407         }
 408
 409         if (parent->i_mode & S_ISGID) {
 410                 inode->i_gid = parent->i_gid;
 411                 if (S_ISDIR(mode))
 412                         inode->i_mode |= S_ISGID;
 413         }
 414
 415         return inode;
 416 }
 417
 418 /*
 419  * File creation. Allocate an inode, and we're done..
 420  */
 421 /* SMP-safe */
 422 static int dlmfs_mkdir(struct inode * dir,
 423                        struct dentry * dentry,
 424                        int mode)
 425 {
 426         int status;
 427         struct inode *inode = NULL;
 428         struct qstr *domain = &dentry->d_name;
 429         struct dlmfs_inode_private *ip;
 430         struct dlm_ctxt *dlm;
 431         struct dlm_protocol_version proto = user_locking_protocol;
 432
 433         mlog(0, "mkdir %.*s\n", domain->len, domain->name);
 434
 435         /* verify that we have a proper domain */
 436         if (domain->len >= O2NM_MAX_NAME_LEN) {
 437                 status = -EINVAL;
 438                 mlog(ML_ERROR, "invalid domain name for directory.\n");
 439                 goto bail;
 440         }
 441
 442         inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR);
 443         if (!inode) {
 444                 status = -ENOMEM;
 445                 mlog_errno(status);
 446                 goto bail;
 447         }
 448
 449         ip = DLMFS_I(inode);
 450
 451         dlm = user_dlm_register_context(domain, &proto);
 452         if (IS_ERR(dlm)) {
 453                 status = PTR_ERR(dlm);
 454                 mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n",
 455                      status, domain->len, domain->name);
 456                 goto bail;
 457         }
 458         ip->ip_dlm = dlm;
 459
 460         inc_nlink(dir);
 461         d_instantiate(dentry, inode);
 462         dget(dentry);   /* Extra count - pin the dentry in core */
 463
 464         status = 0;
 465 bail:
 466         if (status < 0)
 467                 iput(inode);
 468         return status;
 469 }
 470
 471 static int dlmfs_create(struct inode *dir,
 472                         struct dentry *dentry,
 473                         int mode,
 474                         struct nameidata *nd)
 475 {
 476         int status = 0;
 477         struct inode *inode;
 478         struct qstr *name = &dentry->d_name;
 479
 480         mlog(0, "create %.*s\n", name->len, name->name);
 481
 482         /* verify name is valid and doesn't contain any dlm reserved
 483          * characters */
 484         if (name->len >= USER_DLM_LOCK_ID_MAX_LEN ||
 485             name->name[0] == '$') {
 486                 status = -EINVAL;
 487                 mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len,
 488                      name->name);
 489                 goto bail;
 490         }
 491
 492         inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG);
 493         if (!inode) {
 494                 status = -ENOMEM;
 495                 mlog_errno(status);
 496                 goto bail;
 497         }
 498
 499         d_instantiate(dentry, inode);
 500         dget(dentry);   /* Extra count - pin the dentry in core */
 501 bail:
 502         return status;
 503 }
 504
 505 static int dlmfs_unlink(struct inode *dir,
 506                         struct dentry *dentry)
 507 {
 508         int status;
 509         struct inode *inode = dentry->d_inode;
 510
 511         mlog(0, "unlink inode %lu\n", inode->i_ino);
 512
 513         /* if there are no current holders, or none that are waiting
 514          * to acquire a lock, this basically destroys our lockres. */
 515         status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres);
 516         if (status < 0) {
 517                 mlog(ML_ERROR, "unlink %.*s, error %d from destroy\n",
 518                      dentry->d_name.len, dentry->d_name.name, status);
 519                 goto bail;
 520         }
 521         status = simple_unlink(dir, dentry);
 522 bail:
 523         return status;
 524 }
 525
 526 static int dlmfs_fill_super(struct super_block * sb,
 527                             void * data,
 528                             int silent)
 529 {
 530         struct inode * inode;
 531         struct dentry * root;
 532
 533         sb->s_maxbytes = MAX_LFS_FILESIZE;
 534         sb->s_blocksize = PAGE_CACHE_SIZE;
 535         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 536         sb->s_magic = DLMFS_MAGIC;
 537         sb->s_op = &dlmfs_ops;
 538         inode = dlmfs_get_root_inode(sb);
 539         if (!inode)
 540                 return -ENOMEM;
 541
 542         root = d_alloc_root(inode);
 543         if (!root) {
 544                 iput(inode);
 545                 return -ENOMEM;
 546         }
 547         sb->s_root = root;
 548         return 0;
 549 }
 550
 551 static const struct file_operations dlmfs_file_operations = {
 552         .open           = dlmfs_file_open,
 553         .release        = dlmfs_file_release,
 554         .read           = dlmfs_file_read,
 555         .write          = dlmfs_file_write,
 556 };
 557
 558 static const struct inode_operations dlmfs_dir_inode_operations = {
 559         .create         = dlmfs_create,
 560         .lookup         = simple_lookup,
 561         .unlink         = dlmfs_unlink,
 562 };
 563
 564 /* this way we can restrict mkdir to only the toplevel of the fs. */
 565 static const struct inode_operations dlmfs_root_inode_operations = {
 566         .lookup         = simple_lookup,
 567         .mkdir          = dlmfs_mkdir,
 568         .rmdir          = simple_rmdir,
 569 };
 570
 571 static const struct super_operations dlmfs_ops = {
 572         .statfs         = simple_statfs,
 573         .alloc_inode    = dlmfs_alloc_inode,
 574         .destroy_inode  = dlmfs_destroy_inode,
 575         .clear_inode    = dlmfs_clear_inode,
 576         .drop_inode     = generic_delete_inode,
 577 };
 578
 579 static const struct inode_operations dlmfs_file_inode_operations = {
 580         .getattr        = simple_getattr,
 581 };
 582
 583 static int dlmfs_get_sb(struct file_system_type *fs_type,
 584         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
 585 {
 586         return get_sb_nodev(fs_type, flags, data, dlmfs_fill_super, mnt);
 587 }
 588
 589 static struct file_system_type dlmfs_fs_type = {
 590         .owner          = THIS_MODULE,
 591         .name           = "ocfs2_dlmfs",
 592         .get_sb         = dlmfs_get_sb,
 593         .kill_sb        = kill_litter_super,
 594 };
 595
 596 static int __init init_dlmfs_fs(void)
 597 {
 598         int status;
 599         int cleanup_inode = 0, cleanup_worker = 0;
 600
 601         dlmfs_print_version();
 602
 603         status = bdi_init(&dlmfs_backing_dev_info);
 604         if (status)
 605                 return status;
 606
 607         dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache",
 608                                 sizeof(struct dlmfs_inode_private),
 609                                 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 610                                         SLAB_MEM_SPREAD),
 611                                 dlmfs_init_once);
 612         if (!dlmfs_inode_cache)
 613                 goto bail;
 614         cleanup_inode = 1;
 615
 616         user_dlm_worker = create_singlethread_workqueue("user_dlm");
 617         if (!user_dlm_worker) {
 618                 status = -ENOMEM;
 619                 goto bail;
 620         }
 621         cleanup_worker = 1;
 622
 623         status = register_filesystem(&dlmfs_fs_type);
 624 bail:
 625         if (status) {
 626                 if (cleanup_inode)
 627                         kmem_cache_destroy(dlmfs_inode_cache);
 628                 if (cleanup_worker)
 629                         destroy_workqueue(user_dlm_worker);
 630                 bdi_destroy(&dlmfs_backing_dev_info);
 631         } else
 632                 printk("OCFS2 User DLM kernel interface loaded\n");
 633         return status;
 634 }
 635
 636 static void __exit exit_dlmfs_fs(void)
 637 {
 638         unregister_filesystem(&dlmfs_fs_type);
 639
 640         flush_workqueue(user_dlm_worker);
 641         destroy_workqueue(user_dlm_worker);
 642
 643         kmem_cache_destroy(dlmfs_inode_cache);
 644
 645         bdi_destroy(&dlmfs_backing_dev_info);
 646 }
 647
 648 MODULE_AUTHOR("Oracle");
 649 MODULE_LICENSE("GPL");
 650
 651 module_init(init_dlmfs_fs)
 652 module_exit(exit_dlmfs_fs)