1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * (C) 2001 Clemson University and The University of Chicago 4 * 5 * See COPYING in top-level directory. 6 */ 7 8 #include "protocol.h" 9 #include "orangefs-kernel.h" 10 #include "orangefs-bufmap.h" 11 12 #include <linux/parser.h> 13 #include <linux/hashtable.h> 14 #include <linux/seq_file.h> 15 16 /* a cache for orangefs-inode objects (i.e. orangefs inode private data) */ 17 static struct kmem_cache *orangefs_inode_cache; 18 19 /* list for storing orangefs specific superblocks in use */ 20 LIST_HEAD(orangefs_superblocks); 21 22 DEFINE_SPINLOCK(orangefs_superblocks_lock); 23 24 enum { 25 Opt_intr, 26 Opt_acl, 27 Opt_local_lock, 28 29 Opt_err 30 }; 31 32 static const match_table_t tokens = { 33 { Opt_acl, "acl" }, 34 { Opt_intr, "intr" }, 35 { Opt_local_lock, "local_lock" }, 36 { Opt_err, NULL } 37 }; 38 39 uint64_t orangefs_features; 40 41 static int orangefs_show_options(struct seq_file *m, struct dentry *root) 42 { 43 struct orangefs_sb_info_s *orangefs_sb = ORANGEFS_SB(root->d_sb); 44 45 if (root->d_sb->s_flags & SB_POSIXACL) 46 seq_puts(m, ",acl"); 47 if (orangefs_sb->flags & ORANGEFS_OPT_INTR) 48 seq_puts(m, ",intr"); 49 if (orangefs_sb->flags & ORANGEFS_OPT_LOCAL_LOCK) 50 seq_puts(m, ",local_lock"); 51 return 0; 52 } 53 54 static int parse_mount_options(struct super_block *sb, char *options, 55 int silent) 56 { 57 struct orangefs_sb_info_s *orangefs_sb = ORANGEFS_SB(sb); 58 substring_t args[MAX_OPT_ARGS]; 59 char *p; 60 61 /* 62 * Force any potential flags that might be set from the mount 63 * to zero, ie, initialize to unset. 64 */ 65 sb->s_flags &= ~SB_POSIXACL; 66 orangefs_sb->flags &= ~ORANGEFS_OPT_INTR; 67 orangefs_sb->flags &= ~ORANGEFS_OPT_LOCAL_LOCK; 68 69 while ((p = strsep(&options, ",")) != NULL) { 70 int token; 71 72 if (!*p) 73 continue; 74 75 token = match_token(p, tokens, args); 76 switch (token) { 77 case Opt_acl: 78 sb->s_flags |= SB_POSIXACL; 79 break; 80 case Opt_intr: 81 orangefs_sb->flags |= ORANGEFS_OPT_INTR; 82 break; 83 case Opt_local_lock: 84 orangefs_sb->flags |= ORANGEFS_OPT_LOCAL_LOCK; 85 break; 86 default: 87 goto fail; 88 } 89 } 90 91 return 0; 92 fail: 93 if (!silent) 94 gossip_err("Error: mount option [%s] is not supported.\n", p); 95 return -EINVAL; 96 } 97 98 static void orangefs_inode_cache_ctor(void *req) 99 { 100 struct orangefs_inode_s *orangefs_inode = req; 101 102 inode_init_once(&orangefs_inode->vfs_inode); 103 init_rwsem(&orangefs_inode->xattr_sem); 104 } 105 106 static struct inode *orangefs_alloc_inode(struct super_block *sb) 107 { 108 struct orangefs_inode_s *orangefs_inode; 109 110 orangefs_inode = alloc_inode_sb(sb, orangefs_inode_cache, GFP_KERNEL); 111 if (!orangefs_inode) 112 return NULL; 113 114 /* 115 * We want to clear everything except for rw_semaphore and the 116 * vfs_inode. 117 */ 118 memset(&orangefs_inode->refn.khandle, 0, 16); 119 orangefs_inode->refn.fs_id = ORANGEFS_FS_ID_NULL; 120 orangefs_inode->last_failed_block_index_read = 0; 121 memset(orangefs_inode->link_target, 0, sizeof(orangefs_inode->link_target)); 122 123 gossip_debug(GOSSIP_SUPER_DEBUG, 124 "orangefs_alloc_inode: allocated %p\n", 125 &orangefs_inode->vfs_inode); 126 return &orangefs_inode->vfs_inode; 127 } 128 129 static void orangefs_free_inode(struct inode *inode) 130 { 131 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode); 132 struct orangefs_cached_xattr *cx; 133 struct hlist_node *tmp; 134 int i; 135 136 hash_for_each_safe(orangefs_inode->xattr_cache, i, tmp, cx, node) { 137 hlist_del(&cx->node); 138 kfree(cx); 139 } 140 141 kmem_cache_free(orangefs_inode_cache, orangefs_inode); 142 } 143 144 static void orangefs_destroy_inode(struct inode *inode) 145 { 146 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode); 147 148 gossip_debug(GOSSIP_SUPER_DEBUG, 149 "%s: deallocated %p destroying inode %pU\n", 150 __func__, orangefs_inode, get_khandle_from_ino(inode)); 151 } 152 153 static int orangefs_write_inode(struct inode *inode, 154 struct writeback_control *wbc) 155 { 156 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_write_inode\n"); 157 return orangefs_inode_setattr(inode); 158 } 159 160 /* 161 * NOTE: information filled in here is typically reflected in the 162 * output of the system command 'df' 163 */ 164 static int orangefs_statfs(struct dentry *dentry, struct kstatfs *buf) 165 { 166 int ret = -ENOMEM; 167 struct orangefs_kernel_op_s *new_op = NULL; 168 int flags = 0; 169 struct super_block *sb = NULL; 170 171 sb = dentry->d_sb; 172 173 gossip_debug(GOSSIP_SUPER_DEBUG, 174 "%s: called on sb %p (fs_id is %d)\n", 175 __func__, 176 sb, 177 (int)(ORANGEFS_SB(sb)->fs_id)); 178 179 new_op = op_alloc(ORANGEFS_VFS_OP_STATFS); 180 if (!new_op) 181 return ret; 182 new_op->upcall.req.statfs.fs_id = ORANGEFS_SB(sb)->fs_id; 183 184 if (ORANGEFS_SB(sb)->flags & ORANGEFS_OPT_INTR) 185 flags = ORANGEFS_OP_INTERRUPTIBLE; 186 187 ret = service_operation(new_op, "orangefs_statfs", flags); 188 189 if (new_op->downcall.status < 0) 190 goto out_op_release; 191 192 gossip_debug(GOSSIP_SUPER_DEBUG, 193 "%s: got %ld blocks available | " 194 "%ld blocks total | %ld block size | " 195 "%ld files total | %ld files avail\n", 196 __func__, 197 (long)new_op->downcall.resp.statfs.blocks_avail, 198 (long)new_op->downcall.resp.statfs.blocks_total, 199 (long)new_op->downcall.resp.statfs.block_size, 200 (long)new_op->downcall.resp.statfs.files_total, 201 (long)new_op->downcall.resp.statfs.files_avail); 202 203 buf->f_type = sb->s_magic; 204 memcpy(&buf->f_fsid, &ORANGEFS_SB(sb)->fs_id, sizeof(buf->f_fsid)); 205 buf->f_bsize = new_op->downcall.resp.statfs.block_size; 206 buf->f_namelen = ORANGEFS_NAME_MAX; 207 208 buf->f_blocks = (sector_t) new_op->downcall.resp.statfs.blocks_total; 209 buf->f_bfree = (sector_t) new_op->downcall.resp.statfs.blocks_avail; 210 buf->f_bavail = (sector_t) new_op->downcall.resp.statfs.blocks_avail; 211 buf->f_files = (sector_t) new_op->downcall.resp.statfs.files_total; 212 buf->f_ffree = (sector_t) new_op->downcall.resp.statfs.files_avail; 213 buf->f_frsize = 0; 214 215 out_op_release: 216 op_release(new_op); 217 gossip_debug(GOSSIP_SUPER_DEBUG, "%s: returning %d\n", __func__, ret); 218 return ret; 219 } 220 221 /* 222 * Remount as initiated by VFS layer. We just need to reparse the mount 223 * options, no need to signal pvfs2-client-core about it. 224 */ 225 static int orangefs_remount_fs(struct super_block *sb, int *flags, char *data) 226 { 227 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_remount_fs: called\n"); 228 return parse_mount_options(sb, data, 1); 229 } 230 231 /* 232 * Remount as initiated by pvfs2-client-core on restart. This is used to 233 * repopulate mount information left from previous pvfs2-client-core. 234 * 235 * the idea here is that given a valid superblock, we're 236 * re-initializing the user space client with the initial mount 237 * information specified when the super block was first initialized. 238 * this is very different than the first initialization/creation of a 239 * superblock. we use the special service_priority_operation to make 240 * sure that the mount gets ahead of any other pending operation that 241 * is waiting for servicing. this means that the pvfs2-client won't 242 * fail to start several times for all other pending operations before 243 * the client regains all of the mount information from us. 244 * NOTE: this function assumes that the request_mutex is already acquired! 245 */ 246 int orangefs_remount(struct orangefs_sb_info_s *orangefs_sb) 247 { 248 struct orangefs_kernel_op_s *new_op; 249 int ret = -EINVAL; 250 251 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_remount: called\n"); 252 253 new_op = op_alloc(ORANGEFS_VFS_OP_FS_MOUNT); 254 if (!new_op) 255 return -ENOMEM; 256 strscpy(new_op->upcall.req.fs_mount.orangefs_config_server, 257 orangefs_sb->devname); 258 259 gossip_debug(GOSSIP_SUPER_DEBUG, 260 "Attempting ORANGEFS Remount via host %s\n", 261 new_op->upcall.req.fs_mount.orangefs_config_server); 262 263 /* 264 * we assume that the calling function has already acquired the 265 * request_mutex to prevent other operations from bypassing 266 * this one 267 */ 268 ret = service_operation(new_op, "orangefs_remount", 269 ORANGEFS_OP_PRIORITY | ORANGEFS_OP_NO_MUTEX); 270 gossip_debug(GOSSIP_SUPER_DEBUG, 271 "orangefs_remount: mount got return value of %d\n", 272 ret); 273 if (ret == 0) { 274 /* 275 * store the id assigned to this sb -- it's just a 276 * short-lived mapping that the system interface uses 277 * to map this superblock to a particular mount entry 278 */ 279 orangefs_sb->id = new_op->downcall.resp.fs_mount.id; 280 orangefs_sb->mount_pending = 0; 281 } 282 283 op_release(new_op); 284 285 if (orangefs_userspace_version >= 20906) { 286 new_op = op_alloc(ORANGEFS_VFS_OP_FEATURES); 287 if (!new_op) 288 return -ENOMEM; 289 new_op->upcall.req.features.features = 0; 290 ret = service_operation(new_op, "orangefs_features", 291 ORANGEFS_OP_PRIORITY | ORANGEFS_OP_NO_MUTEX); 292 if (!ret) 293 orangefs_features = 294 new_op->downcall.resp.features.features; 295 else 296 orangefs_features = 0; 297 op_release(new_op); 298 } else { 299 orangefs_features = 0; 300 } 301 302 return ret; 303 } 304 305 int fsid_key_table_initialize(void) 306 { 307 return 0; 308 } 309 310 void fsid_key_table_finalize(void) 311 { 312 } 313 314 static const struct super_operations orangefs_s_ops = { 315 .alloc_inode = orangefs_alloc_inode, 316 .free_inode = orangefs_free_inode, 317 .destroy_inode = orangefs_destroy_inode, 318 .write_inode = orangefs_write_inode, 319 .drop_inode = generic_delete_inode, 320 .statfs = orangefs_statfs, 321 .remount_fs = orangefs_remount_fs, 322 .show_options = orangefs_show_options, 323 }; 324 325 static struct dentry *orangefs_fh_to_dentry(struct super_block *sb, 326 struct fid *fid, 327 int fh_len, 328 int fh_type) 329 { 330 struct orangefs_object_kref refn; 331 332 if (fh_len < 5 || fh_type > 2) 333 return NULL; 334 335 ORANGEFS_khandle_from(&(refn.khandle), fid->raw, 16); 336 refn.fs_id = (u32) fid->raw[4]; 337 gossip_debug(GOSSIP_SUPER_DEBUG, 338 "fh_to_dentry: handle %pU, fs_id %d\n", 339 &refn.khandle, 340 refn.fs_id); 341 342 return d_obtain_alias(orangefs_iget(sb, &refn)); 343 } 344 345 static int orangefs_encode_fh(struct inode *inode, 346 __u32 *fh, 347 int *max_len, 348 struct inode *parent) 349 { 350 int len = parent ? 10 : 5; 351 int type = 1; 352 struct orangefs_object_kref refn; 353 354 if (*max_len < len) { 355 gossip_err("fh buffer is too small for encoding\n"); 356 *max_len = len; 357 type = 255; 358 goto out; 359 } 360 361 refn = ORANGEFS_I(inode)->refn; 362 ORANGEFS_khandle_to(&refn.khandle, fh, 16); 363 fh[4] = refn.fs_id; 364 365 gossip_debug(GOSSIP_SUPER_DEBUG, 366 "Encoding fh: handle %pU, fsid %u\n", 367 &refn.khandle, 368 refn.fs_id); 369 370 371 if (parent) { 372 refn = ORANGEFS_I(parent)->refn; 373 ORANGEFS_khandle_to(&refn.khandle, (char *) fh + 20, 16); 374 fh[9] = refn.fs_id; 375 376 type = 2; 377 gossip_debug(GOSSIP_SUPER_DEBUG, 378 "Encoding parent: handle %pU, fsid %u\n", 379 &refn.khandle, 380 refn.fs_id); 381 } 382 *max_len = len; 383 384 out: 385 return type; 386 } 387 388 static const struct export_operations orangefs_export_ops = { 389 .encode_fh = orangefs_encode_fh, 390 .fh_to_dentry = orangefs_fh_to_dentry, 391 }; 392 393 static int orangefs_unmount(int id, __s32 fs_id, const char *devname) 394 { 395 struct orangefs_kernel_op_s *op; 396 int r; 397 op = op_alloc(ORANGEFS_VFS_OP_FS_UMOUNT); 398 if (!op) 399 return -ENOMEM; 400 op->upcall.req.fs_umount.id = id; 401 op->upcall.req.fs_umount.fs_id = fs_id; 402 strscpy(op->upcall.req.fs_umount.orangefs_config_server, devname); 403 r = service_operation(op, "orangefs_fs_umount", 0); 404 /* Not much to do about an error here. */ 405 if (r) 406 gossip_err("orangefs_unmount: service_operation %d\n", r); 407 op_release(op); 408 return r; 409 } 410 411 static int orangefs_fill_sb(struct super_block *sb, 412 struct orangefs_fs_mount_response *fs_mount, 413 void *data, int silent) 414 { 415 int ret; 416 struct inode *root; 417 struct dentry *root_dentry; 418 struct orangefs_object_kref root_object; 419 420 ORANGEFS_SB(sb)->sb = sb; 421 422 ORANGEFS_SB(sb)->root_khandle = fs_mount->root_khandle; 423 ORANGEFS_SB(sb)->fs_id = fs_mount->fs_id; 424 ORANGEFS_SB(sb)->id = fs_mount->id; 425 426 if (data) { 427 ret = parse_mount_options(sb, data, silent); 428 if (ret) 429 return ret; 430 } 431 432 /* Hang the xattr handlers off the superblock */ 433 sb->s_xattr = orangefs_xattr_handlers; 434 sb->s_magic = ORANGEFS_SUPER_MAGIC; 435 sb->s_op = &orangefs_s_ops; 436 sb->s_d_op = &orangefs_dentry_operations; 437 438 sb->s_blocksize = PAGE_SIZE; 439 sb->s_blocksize_bits = PAGE_SHIFT; 440 sb->s_maxbytes = MAX_LFS_FILESIZE; 441 442 ret = super_setup_bdi(sb); 443 if (ret) 444 return ret; 445 446 root_object.khandle = ORANGEFS_SB(sb)->root_khandle; 447 root_object.fs_id = ORANGEFS_SB(sb)->fs_id; 448 gossip_debug(GOSSIP_SUPER_DEBUG, 449 "get inode %pU, fsid %d\n", 450 &root_object.khandle, 451 root_object.fs_id); 452 453 root = orangefs_iget(sb, &root_object); 454 if (IS_ERR(root)) 455 return PTR_ERR(root); 456 457 gossip_debug(GOSSIP_SUPER_DEBUG, 458 "Allocated root inode [%p] with mode %x\n", 459 root, 460 root->i_mode); 461 462 /* allocates and places root dentry in dcache */ 463 root_dentry = d_make_root(root); 464 if (!root_dentry) 465 return -ENOMEM; 466 467 sb->s_export_op = &orangefs_export_ops; 468 sb->s_root = root_dentry; 469 return 0; 470 } 471 472 struct dentry *orangefs_mount(struct file_system_type *fst, 473 int flags, 474 const char *devname, 475 void *data) 476 { 477 int ret; 478 struct super_block *sb = ERR_PTR(-EINVAL); 479 struct orangefs_kernel_op_s *new_op; 480 struct dentry *d = ERR_PTR(-EINVAL); 481 482 gossip_debug(GOSSIP_SUPER_DEBUG, 483 "orangefs_mount: called with devname %s\n", 484 devname); 485 486 if (!devname) { 487 gossip_err("ERROR: device name not specified.\n"); 488 return ERR_PTR(-EINVAL); 489 } 490 491 new_op = op_alloc(ORANGEFS_VFS_OP_FS_MOUNT); 492 if (!new_op) 493 return ERR_PTR(-ENOMEM); 494 495 strscpy(new_op->upcall.req.fs_mount.orangefs_config_server, devname); 496 497 gossip_debug(GOSSIP_SUPER_DEBUG, 498 "Attempting ORANGEFS Mount via host %s\n", 499 new_op->upcall.req.fs_mount.orangefs_config_server); 500 501 ret = service_operation(new_op, "orangefs_mount", 0); 502 gossip_debug(GOSSIP_SUPER_DEBUG, 503 "orangefs_mount: mount got return value of %d\n", ret); 504 if (ret) 505 goto free_op; 506 507 if (new_op->downcall.resp.fs_mount.fs_id == ORANGEFS_FS_ID_NULL) { 508 gossip_err("ERROR: Retrieved null fs_id\n"); 509 ret = -EINVAL; 510 goto free_op; 511 } 512 513 sb = sget(fst, NULL, set_anon_super, flags, NULL); 514 515 if (IS_ERR(sb)) { 516 d = ERR_CAST(sb); 517 orangefs_unmount(new_op->downcall.resp.fs_mount.id, 518 new_op->downcall.resp.fs_mount.fs_id, devname); 519 goto free_op; 520 } 521 522 /* alloc and init our private orangefs sb info */ 523 sb->s_fs_info = kzalloc(sizeof(struct orangefs_sb_info_s), GFP_KERNEL); 524 if (!ORANGEFS_SB(sb)) { 525 d = ERR_PTR(-ENOMEM); 526 goto free_op; 527 } 528 529 ret = orangefs_fill_sb(sb, 530 &new_op->downcall.resp.fs_mount, data, 531 flags & SB_SILENT ? 1 : 0); 532 533 if (ret) { 534 d = ERR_PTR(ret); 535 goto free_sb_and_op; 536 } 537 538 /* 539 * on successful mount, store the devname and data 540 * used 541 */ 542 strscpy(ORANGEFS_SB(sb)->devname, devname); 543 544 545 /* mount_pending must be cleared */ 546 ORANGEFS_SB(sb)->mount_pending = 0; 547 548 /* 549 * finally, add this sb to our list of known orangefs 550 * sb's 551 */ 552 gossip_debug(GOSSIP_SUPER_DEBUG, 553 "Adding SB %p to orangefs superblocks\n", 554 ORANGEFS_SB(sb)); 555 spin_lock(&orangefs_superblocks_lock); 556 list_add_tail(&ORANGEFS_SB(sb)->list, &orangefs_superblocks); 557 spin_unlock(&orangefs_superblocks_lock); 558 op_release(new_op); 559 560 /* Must be removed from the list now. */ 561 ORANGEFS_SB(sb)->no_list = 0; 562 563 if (orangefs_userspace_version >= 20906) { 564 new_op = op_alloc(ORANGEFS_VFS_OP_FEATURES); 565 if (!new_op) 566 return ERR_PTR(-ENOMEM); 567 new_op->upcall.req.features.features = 0; 568 ret = service_operation(new_op, "orangefs_features", 0); 569 orangefs_features = new_op->downcall.resp.features.features; 570 op_release(new_op); 571 } else { 572 orangefs_features = 0; 573 } 574 575 return dget(sb->s_root); 576 577 free_sb_and_op: 578 /* Will call orangefs_kill_sb with sb not in list. */ 579 ORANGEFS_SB(sb)->no_list = 1; 580 /* ORANGEFS_VFS_OP_FS_UMOUNT is done by orangefs_kill_sb. */ 581 deactivate_locked_super(sb); 582 free_op: 583 gossip_err("orangefs_mount: mount request failed with %d\n", ret); 584 if (ret == -EINVAL) { 585 gossip_err("Ensure that all orangefs-servers have the same FS configuration files\n"); 586 gossip_err("Look at pvfs2-client-core log file (typically /tmp/pvfs2-client.log) for more details\n"); 587 } 588 589 op_release(new_op); 590 591 return d; 592 } 593 594 void orangefs_kill_sb(struct super_block *sb) 595 { 596 int r; 597 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_kill_sb: called\n"); 598 599 /* provided sb cleanup */ 600 kill_anon_super(sb); 601 602 if (!ORANGEFS_SB(sb)) { 603 mutex_lock(&orangefs_request_mutex); 604 mutex_unlock(&orangefs_request_mutex); 605 return; 606 } 607 /* 608 * issue the unmount to userspace to tell it to remove the 609 * dynamic mount info it has for this superblock 610 */ 611 r = orangefs_unmount(ORANGEFS_SB(sb)->id, ORANGEFS_SB(sb)->fs_id, 612 ORANGEFS_SB(sb)->devname); 613 if (!r) 614 ORANGEFS_SB(sb)->mount_pending = 1; 615 616 if (!ORANGEFS_SB(sb)->no_list) { 617 /* remove the sb from our list of orangefs specific sb's */ 618 spin_lock(&orangefs_superblocks_lock); 619 /* not list_del_init */ 620 __list_del_entry(&ORANGEFS_SB(sb)->list); 621 ORANGEFS_SB(sb)->list.prev = NULL; 622 spin_unlock(&orangefs_superblocks_lock); 623 } 624 625 /* 626 * make sure that ORANGEFS_DEV_REMOUNT_ALL loop that might've seen us 627 * gets completed before we free the dang thing. 628 */ 629 mutex_lock(&orangefs_request_mutex); 630 mutex_unlock(&orangefs_request_mutex); 631 632 /* free the orangefs superblock private data */ 633 kfree(ORANGEFS_SB(sb)); 634 } 635 636 int orangefs_inode_cache_initialize(void) 637 { 638 orangefs_inode_cache = kmem_cache_create_usercopy( 639 "orangefs_inode_cache", 640 sizeof(struct orangefs_inode_s), 641 0, 642 0, 643 offsetof(struct orangefs_inode_s, 644 link_target), 645 sizeof_field(struct orangefs_inode_s, 646 link_target), 647 orangefs_inode_cache_ctor); 648 649 if (!orangefs_inode_cache) { 650 gossip_err("Cannot create orangefs_inode_cache\n"); 651 return -ENOMEM; 652 } 653 return 0; 654 } 655 656 int orangefs_inode_cache_finalize(void) 657 { 658 kmem_cache_destroy(orangefs_inode_cache); 659 return 0; 660 } 661