fs/ocfs2/dlmfs/dlmfs.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* -*- mode: c; c-basic-offset: 8; -*-
   3  * vim: noexpandtab sw=8 ts=8 sts=0:
   4  *
   5  * dlmfs.c
   6  *
   7  * Code which implements the kernel side of a minimal userspace
   8  * interface to our DLM. This file handles the virtual file system
   9  * used for communication with userspace. Credit should go to ramfs,
  10  * which was a template for the fs side of this module.
  11  *
  12  * Copyright (C) 2003, 2004 Oracle.  All rights reserved.
  13  */
  14
  15 /* Simple VFS hooks based on: */
  16 /*
  17  * Resizable simple ram filesystem for Linux.
  18  *
  19  * Copyright (C) 2000 Linus Torvalds.
  20  *               2000 Transmeta Corp.
  21  */
  22
  23 #include <linux/module.h>
  24 #include <linux/fs.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/types.h>
  27 #include <linux/slab.h>
  28 #include <linux/highmem.h>
  29 #include <linux/init.h>
  30 #include <linux/string.h>
  31 #include <linux/backing-dev.h>
  32 #include <linux/poll.h>
  33
  34 #include <linux/uaccess.h>
  35
  36 #include "../stackglue.h"
  37 #include "userdlm.h"
  38
  39 #define MLOG_MASK_PREFIX ML_DLMFS
  40 #include "../cluster/masklog.h"
  41
  42
  43 static const struct super_operations dlmfs_ops;
  44 static const struct file_operations dlmfs_file_operations;
  45 static const struct inode_operations dlmfs_dir_inode_operations;
  46 static const struct inode_operations dlmfs_root_inode_operations;
  47 static const struct inode_operations dlmfs_file_inode_operations;
  48 static struct kmem_cache *dlmfs_inode_cache;
  49
  50 struct workqueue_struct *user_dlm_worker;
  51
  52
  53
  54 /*
  55  * These are the ABI capabilities of dlmfs.
  56  *
  57  * Over time, dlmfs has added some features that were not part of the
  58  * initial ABI.  Unfortunately, some of these features are not detectable
  59  * via standard usage.  For example, Linux's default poll always returns
  60  * EPOLLIN, so there is no way for a caller of poll(2) to know when dlmfs
  61  * added poll support.  Instead, we provide this list of new capabilities.
  62  *
  63  * Capabilities is a read-only attribute.  We do it as a module parameter
  64  * so we can discover it whether dlmfs is built in, loaded, or even not
  65  * loaded.
  66  *
  67  * The ABI features are local to this machine's dlmfs mount.  This is
  68  * distinct from the locking protocol, which is concerned with inter-node
  69  * interaction.
  70  *
  71  * Capabilities:
  72  * - bast       : EPOLLIN against the file descriptor of a held lock
  73  *                signifies a bast fired on the lock.
  74  */
  75 #define DLMFS_CAPABILITIES "bast stackglue"
  76 static int param_set_dlmfs_capabilities(const char *val,
  77                                         const struct kernel_param *kp)
  78 {
  79         printk(KERN_ERR "%s: readonly parameter\n", kp->name);
  80         return -EINVAL;
  81 }
  82 static int param_get_dlmfs_capabilities(char *buffer,
  83                                         const struct kernel_param *kp)
  84 {
  85         return strlcpy(buffer, DLMFS_CAPABILITIES,
  86                        strlen(DLMFS_CAPABILITIES) + 1);
  87 }
  88 module_param_call(capabilities, param_set_dlmfs_capabilities,
  89                   param_get_dlmfs_capabilities, NULL, 0444);
  90 MODULE_PARM_DESC(capabilities, DLMFS_CAPABILITIES);
  91
  92
  93 /*
  94  * decodes a set of open flags into a valid lock level and a set of flags.
  95  * returns < 0 if we have invalid flags
  96  * flags which mean something to us:
  97  * O_RDONLY -> PRMODE level
  98  * O_WRONLY -> EXMODE level
  99  *
 100  * O_NONBLOCK -> NOQUEUE
 101  */
 102 static int dlmfs_decode_open_flags(int open_flags,
 103                                    int *level,
 104                                    int *flags)
 105 {
 106         if (open_flags & (O_WRONLY|O_RDWR))
 107                 *level = DLM_LOCK_EX;
 108         else
 109                 *level = DLM_LOCK_PR;
 110
 111         *flags = 0;
 112         if (open_flags & O_NONBLOCK)
 113                 *flags |= DLM_LKF_NOQUEUE;
 114
 115         return 0;
 116 }
 117
 118 static int dlmfs_file_open(struct inode *inode,
 119                            struct file *file)
 120 {
 121         int status, level, flags;
 122         struct dlmfs_filp_private *fp = NULL;
 123         struct dlmfs_inode_private *ip;
 124
 125         if (S_ISDIR(inode->i_mode))
 126                 BUG();
 127
 128         mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino,
 129                 file->f_flags);
 130
 131         status = dlmfs_decode_open_flags(file->f_flags, &level, &flags);
 132         if (status < 0)
 133                 goto bail;
 134
 135         /* We don't want to honor O_APPEND at read/write time as it
 136          * doesn't make sense for LVB writes. */
 137         file->f_flags &= ~O_APPEND;
 138
 139         fp = kmalloc(sizeof(*fp), GFP_NOFS);
 140         if (!fp) {
 141                 status = -ENOMEM;
 142                 goto bail;
 143         }
 144         fp->fp_lock_level = level;
 145
 146         ip = DLMFS_I(inode);
 147
 148         status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags);
 149         if (status < 0) {
 150                 /* this is a strange error to return here but I want
 151                  * to be able userspace to be able to distinguish a
 152                  * valid lock request from one that simply couldn't be
 153                  * granted. */
 154                 if (flags & DLM_LKF_NOQUEUE && status == -EAGAIN)
 155                         status = -ETXTBSY;
 156                 kfree(fp);
 157                 goto bail;
 158         }
 159
 160         file->private_data = fp;
 161 bail:
 162         return status;
 163 }
 164
 165 static int dlmfs_file_release(struct inode *inode,
 166                               struct file *file)
 167 {
 168         int level;
 169         struct dlmfs_inode_private *ip = DLMFS_I(inode);
 170         struct dlmfs_filp_private *fp = file->private_data;
 171
 172         if (S_ISDIR(inode->i_mode))
 173                 BUG();
 174
 175         mlog(0, "close called on inode %lu\n", inode->i_ino);
 176
 177         if (fp) {
 178                 level = fp->fp_lock_level;
 179                 if (level != DLM_LOCK_IV)
 180                         user_dlm_cluster_unlock(&ip->ip_lockres, level);
 181
 182                 kfree(fp);
 183                 file->private_data = NULL;
 184         }
 185
 186         return 0;
 187 }
 188
 189 /*
 190  * We do ->setattr() just to override size changes.  Our size is the size
 191  * of the LVB and nothing else.
 192  */
 193 static int dlmfs_file_setattr(struct dentry *dentry, struct iattr *attr)
 194 {
 195         int error;
 196         struct inode *inode = d_inode(dentry);
 197
 198         attr->ia_valid &= ~ATTR_SIZE;
 199         error = setattr_prepare(dentry, attr);
 200         if (error)
 201                 return error;
 202
 203         setattr_copy(inode, attr);
 204         mark_inode_dirty(inode);
 205         return 0;
 206 }
 207
 208 static __poll_t dlmfs_file_poll(struct file *file, poll_table *wait)
 209 {
 210         __poll_t event = 0;
 211         struct inode *inode = file_inode(file);
 212         struct dlmfs_inode_private *ip = DLMFS_I(inode);
 213
 214         poll_wait(file, &ip->ip_lockres.l_event, wait);
 215
 216         spin_lock(&ip->ip_lockres.l_lock);
 217         if (ip->ip_lockres.l_flags & USER_LOCK_BLOCKED)
 218                 event = EPOLLIN | EPOLLRDNORM;
 219         spin_unlock(&ip->ip_lockres.l_lock);
 220
 221         return event;
 222 }
 223
 224 static ssize_t dlmfs_file_read(struct file *filp,
 225                                char __user *buf,
 226                                size_t count,
 227                                loff_t *ppos)
 228 {
 229         int bytes_left;
 230         ssize_t got;
 231         char *lvb_buf;
 232         struct inode *inode = file_inode(filp);
 233
 234         mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
 235                 inode->i_ino, count, *ppos);
 236
 237         if (*ppos >= i_size_read(inode))
 238                 return 0;
 239
 240         /* don't read past the lvb */
 241         if (count > i_size_read(inode) - *ppos)
 242                 count = i_size_read(inode) - *ppos;
 243
 244         if (!count)
 245                 return 0;
 246
 247         lvb_buf = kmalloc(count, GFP_NOFS);
 248         if (!lvb_buf)
 249                 return -ENOMEM;
 250
 251         got = user_dlm_read_lvb(inode, lvb_buf, count);
 252         if (got) {
 253                 BUG_ON(got != count);
 254                 bytes_left = copy_to_user(buf, lvb_buf, count);
 255                 count -= bytes_left;
 256         } else
 257                 count = 0;
 258
 259         kfree(lvb_buf);
 260
 261         *ppos = *ppos + count;
 262
 263         mlog(0, "read %zu bytes\n", count);
 264         return count;
 265 }
 266
 267 static ssize_t dlmfs_file_write(struct file *filp,
 268                                 const char __user *buf,
 269                                 size_t count,
 270                                 loff_t *ppos)
 271 {
 272         int bytes_left;
 273         char *lvb_buf;
 274         struct inode *inode = file_inode(filp);
 275
 276         mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
 277                 inode->i_ino, count, *ppos);
 278
 279         if (*ppos >= i_size_read(inode))
 280                 return -ENOSPC;
 281
 282         /* don't write past the lvb */
 283         if (count > i_size_read(inode) - *ppos)
 284                 count = i_size_read(inode) - *ppos;
 285
 286         if (!count)
 287                 return 0;
 288
 289         lvb_buf = kmalloc(count, GFP_NOFS);
 290         if (!lvb_buf)
 291                 return -ENOMEM;
 292
 293         bytes_left = copy_from_user(lvb_buf, buf, count);
 294         count -= bytes_left;
 295         if (count)
 296                 user_dlm_write_lvb(inode, lvb_buf, count);
 297
 298         kfree(lvb_buf);
 299
 300         *ppos = *ppos + count;
 301         mlog(0, "wrote %zu bytes\n", count);
 302         return count;
 303 }
 304
 305 static void dlmfs_init_once(void *foo)
 306 {
 307         struct dlmfs_inode_private *ip =
 308                 (struct dlmfs_inode_private *) foo;
 309
 310         ip->ip_conn = NULL;
 311         ip->ip_parent = NULL;
 312
 313         inode_init_once(&ip->ip_vfs_inode);
 314 }
 315
 316 static struct inode *dlmfs_alloc_inode(struct super_block *sb)
 317 {
 318         struct dlmfs_inode_private *ip;
 319
 320         ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS);
 321         if (!ip)
 322                 return NULL;
 323
 324         return &ip->ip_vfs_inode;
 325 }
 326
 327 static void dlmfs_free_inode(struct inode *inode)
 328 {
 329         kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode));
 330 }
 331
 332 static void dlmfs_evict_inode(struct inode *inode)
 333 {
 334         int status;
 335         struct dlmfs_inode_private *ip;
 336
 337         clear_inode(inode);
 338
 339         mlog(0, "inode %lu\n", inode->i_ino);
 340
 341         ip = DLMFS_I(inode);
 342
 343         if (S_ISREG(inode->i_mode)) {
 344                 status = user_dlm_destroy_lock(&ip->ip_lockres);
 345                 if (status < 0)
 346                         mlog_errno(status);
 347                 iput(ip->ip_parent);
 348                 goto clear_fields;
 349         }
 350
 351         mlog(0, "we're a directory, ip->ip_conn = 0x%p\n", ip->ip_conn);
 352         /* we must be a directory. If required, lets unregister the
 353          * dlm context now. */
 354         if (ip->ip_conn)
 355                 user_dlm_unregister(ip->ip_conn);
 356 clear_fields:
 357         ip->ip_parent = NULL;
 358         ip->ip_conn = NULL;
 359 }
 360
 361 static struct inode *dlmfs_get_root_inode(struct super_block *sb)
 362 {
 363         struct inode *inode = new_inode(sb);
 364         umode_t mode = S_IFDIR | 0755;
 365
 366         if (inode) {
 367                 inode->i_ino = get_next_ino();
 368                 inode_init_owner(inode, NULL, mode);
 369                 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
 370                 inc_nlink(inode);
 371
 372                 inode->i_fop = &simple_dir_operations;
 373                 inode->i_op = &dlmfs_root_inode_operations;
 374         }
 375
 376         return inode;
 377 }
 378
 379 static struct inode *dlmfs_get_inode(struct inode *parent,
 380                                      struct dentry *dentry,
 381                                      umode_t mode)
 382 {
 383         struct super_block *sb = parent->i_sb;
 384         struct inode * inode = new_inode(sb);
 385         struct dlmfs_inode_private *ip;
 386
 387         if (!inode)
 388                 return NULL;
 389
 390         inode->i_ino = get_next_ino();
 391         inode_init_owner(inode, parent, mode);
 392         inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
 393
 394         ip = DLMFS_I(inode);
 395         ip->ip_conn = DLMFS_I(parent)->ip_conn;
 396
 397         switch (mode & S_IFMT) {
 398         default:
 399                 /* for now we don't support anything other than
 400                  * directories and regular files. */
 401                 BUG();
 402                 break;
 403         case S_IFREG:
 404                 inode->i_op = &dlmfs_file_inode_operations;
 405                 inode->i_fop = &dlmfs_file_operations;
 406
 407                 i_size_write(inode,  DLM_LVB_LEN);
 408
 409                 user_dlm_lock_res_init(&ip->ip_lockres, dentry);
 410
 411                 /* released at clear_inode time, this insures that we
 412                  * get to drop the dlm reference on each lock *before*
 413                  * we call the unregister code for releasing parent
 414                  * directories. */
 415                 ip->ip_parent = igrab(parent);
 416                 BUG_ON(!ip->ip_parent);
 417                 break;
 418         case S_IFDIR:
 419                 inode->i_op = &dlmfs_dir_inode_operations;
 420                 inode->i_fop = &simple_dir_operations;
 421
 422                 /* directory inodes start off with i_nlink ==
 423                  * 2 (for "." entry) */
 424                 inc_nlink(inode);
 425                 break;
 426         }
 427         return inode;
 428 }
 429
 430 /*
 431  * File creation. Allocate an inode, and we're done..
 432  */
 433 /* SMP-safe */
 434 static int dlmfs_mkdir(struct inode * dir,
 435                        struct dentry * dentry,
 436                        umode_t mode)
 437 {
 438         int status;
 439         struct inode *inode = NULL;
 440         const struct qstr *domain = &dentry->d_name;
 441         struct dlmfs_inode_private *ip;
 442         struct ocfs2_cluster_connection *conn;
 443
 444         mlog(0, "mkdir %.*s\n", domain->len, domain->name);
 445
 446         /* verify that we have a proper domain */
 447         if (domain->len >= GROUP_NAME_MAX) {
 448                 status = -EINVAL;
 449                 mlog(ML_ERROR, "invalid domain name for directory.\n");
 450                 goto bail;
 451         }
 452
 453         inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR);
 454         if (!inode) {
 455                 status = -ENOMEM;
 456                 mlog_errno(status);
 457                 goto bail;
 458         }
 459
 460         ip = DLMFS_I(inode);
 461
 462         conn = user_dlm_register(domain);
 463         if (IS_ERR(conn)) {
 464                 status = PTR_ERR(conn);
 465                 mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n",
 466                      status, domain->len, domain->name);
 467                 goto bail;
 468         }
 469         ip->ip_conn = conn;
 470
 471         inc_nlink(dir);
 472         d_instantiate(dentry, inode);
 473         dget(dentry);   /* Extra count - pin the dentry in core */
 474
 475         status = 0;
 476 bail:
 477         if (status < 0)
 478                 iput(inode);
 479         return status;
 480 }
 481
 482 static int dlmfs_create(struct inode *dir,
 483                         struct dentry *dentry,
 484                         umode_t mode,
 485                         bool excl)
 486 {
 487         int status = 0;
 488         struct inode *inode;
 489         const struct qstr *name = &dentry->d_name;
 490
 491         mlog(0, "create %.*s\n", name->len, name->name);
 492
 493         /* verify name is valid and doesn't contain any dlm reserved
 494          * characters */
 495         if (name->len >= USER_DLM_LOCK_ID_MAX_LEN ||
 496             name->name[0] == '$') {
 497                 status = -EINVAL;
 498                 mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len,
 499                      name->name);
 500                 goto bail;
 501         }
 502
 503         inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG);
 504         if (!inode) {
 505                 status = -ENOMEM;
 506                 mlog_errno(status);
 507                 goto bail;
 508         }
 509
 510         d_instantiate(dentry, inode);
 511         dget(dentry);   /* Extra count - pin the dentry in core */
 512 bail:
 513         return status;
 514 }
 515
 516 static int dlmfs_unlink(struct inode *dir,
 517                         struct dentry *dentry)
 518 {
 519         int status;
 520         struct inode *inode = d_inode(dentry);
 521
 522         mlog(0, "unlink inode %lu\n", inode->i_ino);
 523
 524         /* if there are no current holders, or none that are waiting
 525          * to acquire a lock, this basically destroys our lockres. */
 526         status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres);
 527         if (status < 0) {
 528                 mlog(ML_ERROR, "unlink %pd, error %d from destroy\n",
 529                      dentry, status);
 530                 goto bail;
 531         }
 532         status = simple_unlink(dir, dentry);
 533 bail:
 534         return status;
 535 }
 536
 537 static int dlmfs_fill_super(struct super_block * sb,
 538                             void * data,
 539                             int silent)
 540 {
 541         sb->s_maxbytes = MAX_LFS_FILESIZE;
 542         sb->s_blocksize = PAGE_SIZE;
 543         sb->s_blocksize_bits = PAGE_SHIFT;
 544         sb->s_magic = DLMFS_MAGIC;
 545         sb->s_op = &dlmfs_ops;
 546         sb->s_root = d_make_root(dlmfs_get_root_inode(sb));
 547         if (!sb->s_root)
 548                 return -ENOMEM;
 549         return 0;
 550 }
 551
 552 static const struct file_operations dlmfs_file_operations = {
 553         .open           = dlmfs_file_open,
 554         .release        = dlmfs_file_release,
 555         .poll           = dlmfs_file_poll,
 556         .read           = dlmfs_file_read,
 557         .write          = dlmfs_file_write,
 558         .llseek         = default_llseek,
 559 };
 560
 561 static const struct inode_operations dlmfs_dir_inode_operations = {
 562         .create         = dlmfs_create,
 563         .lookup         = simple_lookup,
 564         .unlink         = dlmfs_unlink,
 565 };
 566
 567 /* this way we can restrict mkdir to only the toplevel of the fs. */
 568 static const struct inode_operations dlmfs_root_inode_operations = {
 569         .lookup         = simple_lookup,
 570         .mkdir          = dlmfs_mkdir,
 571         .rmdir          = simple_rmdir,
 572 };
 573
 574 static const struct super_operations dlmfs_ops = {
 575         .statfs         = simple_statfs,
 576         .alloc_inode    = dlmfs_alloc_inode,
 577         .free_inode     = dlmfs_free_inode,
 578         .evict_inode    = dlmfs_evict_inode,
 579         .drop_inode     = generic_delete_inode,
 580 };
 581
 582 static const struct inode_operations dlmfs_file_inode_operations = {
 583         .getattr        = simple_getattr,
 584         .setattr        = dlmfs_file_setattr,
 585 };
 586
 587 static struct dentry *dlmfs_mount(struct file_system_type *fs_type,
 588         int flags, const char *dev_name, void *data)
 589 {
 590         return mount_nodev(fs_type, flags, data, dlmfs_fill_super);
 591 }
 592
 593 static struct file_system_type dlmfs_fs_type = {
 594         .owner          = THIS_MODULE,
 595         .name           = "ocfs2_dlmfs",
 596         .mount          = dlmfs_mount,
 597         .kill_sb        = kill_litter_super,
 598 };
 599 MODULE_ALIAS_FS("ocfs2_dlmfs");
 600
 601 static int __init init_dlmfs_fs(void)
 602 {
 603         int status;
 604         int cleanup_inode = 0, cleanup_worker = 0;
 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|SLAB_ACCOUNT),
 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 = alloc_workqueue("user_dlm", WQ_MEM_RECLAIM, 0);
 618         if (!user_dlm_worker) {
 619                 status = -ENOMEM;
 620                 goto bail;
 621         }
 622         cleanup_worker = 1;
 623
 624         user_dlm_set_locking_protocol();
 625         status = register_filesystem(&dlmfs_fs_type);
 626 bail:
 627         if (status) {
 628                 if (cleanup_inode)
 629                         kmem_cache_destroy(dlmfs_inode_cache);
 630                 if (cleanup_worker)
 631                         destroy_workqueue(user_dlm_worker);
 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         destroy_workqueue(user_dlm_worker);
 642
 643         /*
 644          * Make sure all delayed rcu free inodes are flushed before we
 645          * destroy cache.
 646          */
 647         rcu_barrier();
 648         kmem_cache_destroy(dlmfs_inode_cache);
 649
 650 }
 651
 652 MODULE_AUTHOR("Oracle");
 653 MODULE_LICENSE("GPL");
 654 MODULE_DESCRIPTION("OCFS2 DLM-Filesystem");
 655
 656 module_init(init_dlmfs_fs)
 657 module_exit(exit_dlmfs_fs)