| 1 | /* -*- mode: c; c-basic-offset: 8; -*- |
| 2 | * vim: noexpandtab sw=8 ts=8 sts=0: |
| 3 | * |
| 4 | * slot_map.c |
| 5 | * |
| 6 | * |
| 7 | * |
| 8 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. |
| 9 | * |
| 10 | * This program is free software; you can redistribute it and/or |
| 11 | * modify it under the terms of the GNU General Public |
| 12 | * License as published by the Free Software Foundation; either |
| 13 | * version 2 of the License, or (at your option) any later version. |
| 14 | * |
| 15 | * This program is distributed in the hope that it will be useful, |
| 16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 18 | * General Public License for more details. |
| 19 | * |
| 20 | * You should have received a copy of the GNU General Public |
| 21 | * License along with this program; if not, write to the |
| 22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, |
| 23 | * Boston, MA 021110-1307, USA. |
| 24 | */ |
| 25 | |
| 26 | #include <linux/types.h> |
| 27 | #include <linux/slab.h> |
| 28 | #include <linux/highmem.h> |
| 29 | |
| 30 | #include <cluster/masklog.h> |
| 31 | |
| 32 | #include "ocfs2.h" |
| 33 | |
| 34 | #include "dlmglue.h" |
| 35 | #include "extent_map.h" |
| 36 | #include "heartbeat.h" |
| 37 | #include "inode.h" |
| 38 | #include "slot_map.h" |
| 39 | #include "super.h" |
| 40 | #include "sysfile.h" |
| 41 | #include "ocfs2_trace.h" |
| 42 | |
| 43 | #include "buffer_head_io.h" |
| 44 | |
| 45 | |
| 46 | struct ocfs2_slot { |
| 47 | int sl_valid; |
| 48 | unsigned int sl_node_num; |
| 49 | }; |
| 50 | |
| 51 | struct ocfs2_slot_info { |
| 52 | int si_extended; |
| 53 | int si_slots_per_block; |
| 54 | struct inode *si_inode; |
| 55 | unsigned int si_blocks; |
| 56 | struct buffer_head **si_bh; |
| 57 | unsigned int si_num_slots; |
| 58 | struct ocfs2_slot *si_slots; |
| 59 | }; |
| 60 | |
| 61 | |
| 62 | static int __ocfs2_node_num_to_slot(struct ocfs2_slot_info *si, |
| 63 | unsigned int node_num); |
| 64 | |
| 65 | static void ocfs2_invalidate_slot(struct ocfs2_slot_info *si, |
| 66 | int slot_num) |
| 67 | { |
| 68 | BUG_ON((slot_num < 0) || (slot_num >= si->si_num_slots)); |
| 69 | si->si_slots[slot_num].sl_valid = 0; |
| 70 | } |
| 71 | |
| 72 | static void ocfs2_set_slot(struct ocfs2_slot_info *si, |
| 73 | int slot_num, unsigned int node_num) |
| 74 | { |
| 75 | BUG_ON((slot_num < 0) || (slot_num >= si->si_num_slots)); |
| 76 | |
| 77 | si->si_slots[slot_num].sl_valid = 1; |
| 78 | si->si_slots[slot_num].sl_node_num = node_num; |
| 79 | } |
| 80 | |
| 81 | /* This version is for the extended slot map */ |
| 82 | static void ocfs2_update_slot_info_extended(struct ocfs2_slot_info *si) |
| 83 | { |
| 84 | int b, i, slotno; |
| 85 | struct ocfs2_slot_map_extended *se; |
| 86 | |
| 87 | slotno = 0; |
| 88 | for (b = 0; b < si->si_blocks; b++) { |
| 89 | se = (struct ocfs2_slot_map_extended *)si->si_bh[b]->b_data; |
| 90 | for (i = 0; |
| 91 | (i < si->si_slots_per_block) && |
| 92 | (slotno < si->si_num_slots); |
| 93 | i++, slotno++) { |
| 94 | if (se->se_slots[i].es_valid) |
| 95 | ocfs2_set_slot(si, slotno, |
| 96 | le32_to_cpu(se->se_slots[i].es_node_num)); |
| 97 | else |
| 98 | ocfs2_invalidate_slot(si, slotno); |
| 99 | } |
| 100 | } |
| 101 | } |
| 102 | |
| 103 | /* |
| 104 | * Post the slot information on disk into our slot_info struct. |
| 105 | * Must be protected by osb_lock. |
| 106 | */ |
| 107 | static void ocfs2_update_slot_info_old(struct ocfs2_slot_info *si) |
| 108 | { |
| 109 | int i; |
| 110 | struct ocfs2_slot_map *sm; |
| 111 | |
| 112 | sm = (struct ocfs2_slot_map *)si->si_bh[0]->b_data; |
| 113 | |
| 114 | for (i = 0; i < si->si_num_slots; i++) { |
| 115 | if (le16_to_cpu(sm->sm_slots[i]) == (u16)OCFS2_INVALID_SLOT) |
| 116 | ocfs2_invalidate_slot(si, i); |
| 117 | else |
| 118 | ocfs2_set_slot(si, i, le16_to_cpu(sm->sm_slots[i])); |
| 119 | } |
| 120 | } |
| 121 | |
| 122 | static void ocfs2_update_slot_info(struct ocfs2_slot_info *si) |
| 123 | { |
| 124 | /* |
| 125 | * The slot data will have been refreshed when ocfs2_super_lock |
| 126 | * was taken. |
| 127 | */ |
| 128 | if (si->si_extended) |
| 129 | ocfs2_update_slot_info_extended(si); |
| 130 | else |
| 131 | ocfs2_update_slot_info_old(si); |
| 132 | } |
| 133 | |
| 134 | int ocfs2_refresh_slot_info(struct ocfs2_super *osb) |
| 135 | { |
| 136 | int ret; |
| 137 | struct ocfs2_slot_info *si = osb->slot_info; |
| 138 | |
| 139 | if (si == NULL) |
| 140 | return 0; |
| 141 | |
| 142 | BUG_ON(si->si_blocks == 0); |
| 143 | BUG_ON(si->si_bh == NULL); |
| 144 | |
| 145 | trace_ocfs2_refresh_slot_info(si->si_blocks); |
| 146 | |
| 147 | /* |
| 148 | * We pass -1 as blocknr because we expect all of si->si_bh to |
| 149 | * be !NULL. Thus, ocfs2_read_blocks() will ignore blocknr. If |
| 150 | * this is not true, the read of -1 (UINT64_MAX) will fail. |
| 151 | */ |
| 152 | ret = ocfs2_read_blocks(INODE_CACHE(si->si_inode), -1, si->si_blocks, |
| 153 | si->si_bh, OCFS2_BH_IGNORE_CACHE, NULL); |
| 154 | if (ret == 0) { |
| 155 | spin_lock(&osb->osb_lock); |
| 156 | ocfs2_update_slot_info(si); |
| 157 | spin_unlock(&osb->osb_lock); |
| 158 | } |
| 159 | |
| 160 | return ret; |
| 161 | } |
| 162 | |
| 163 | /* post the our slot info stuff into it's destination bh and write it |
| 164 | * out. */ |
| 165 | static void ocfs2_update_disk_slot_extended(struct ocfs2_slot_info *si, |
| 166 | int slot_num, |
| 167 | struct buffer_head **bh) |
| 168 | { |
| 169 | int blkind = slot_num / si->si_slots_per_block; |
| 170 | int slotno = slot_num % si->si_slots_per_block; |
| 171 | struct ocfs2_slot_map_extended *se; |
| 172 | |
| 173 | BUG_ON(blkind >= si->si_blocks); |
| 174 | |
| 175 | se = (struct ocfs2_slot_map_extended *)si->si_bh[blkind]->b_data; |
| 176 | se->se_slots[slotno].es_valid = si->si_slots[slot_num].sl_valid; |
| 177 | if (si->si_slots[slot_num].sl_valid) |
| 178 | se->se_slots[slotno].es_node_num = |
| 179 | cpu_to_le32(si->si_slots[slot_num].sl_node_num); |
| 180 | *bh = si->si_bh[blkind]; |
| 181 | } |
| 182 | |
| 183 | static void ocfs2_update_disk_slot_old(struct ocfs2_slot_info *si, |
| 184 | int slot_num, |
| 185 | struct buffer_head **bh) |
| 186 | { |
| 187 | int i; |
| 188 | struct ocfs2_slot_map *sm; |
| 189 | |
| 190 | sm = (struct ocfs2_slot_map *)si->si_bh[0]->b_data; |
| 191 | for (i = 0; i < si->si_num_slots; i++) { |
| 192 | if (si->si_slots[i].sl_valid) |
| 193 | sm->sm_slots[i] = |
| 194 | cpu_to_le16(si->si_slots[i].sl_node_num); |
| 195 | else |
| 196 | sm->sm_slots[i] = cpu_to_le16(OCFS2_INVALID_SLOT); |
| 197 | } |
| 198 | *bh = si->si_bh[0]; |
| 199 | } |
| 200 | |
| 201 | static int ocfs2_update_disk_slot(struct ocfs2_super *osb, |
| 202 | struct ocfs2_slot_info *si, |
| 203 | int slot_num) |
| 204 | { |
| 205 | int status; |
| 206 | struct buffer_head *bh; |
| 207 | |
| 208 | spin_lock(&osb->osb_lock); |
| 209 | if (si->si_extended) |
| 210 | ocfs2_update_disk_slot_extended(si, slot_num, &bh); |
| 211 | else |
| 212 | ocfs2_update_disk_slot_old(si, slot_num, &bh); |
| 213 | spin_unlock(&osb->osb_lock); |
| 214 | |
| 215 | status = ocfs2_write_block(osb, bh, INODE_CACHE(si->si_inode)); |
| 216 | if (status < 0) |
| 217 | mlog_errno(status); |
| 218 | |
| 219 | return status; |
| 220 | } |
| 221 | |
| 222 | /* |
| 223 | * Calculate how many bytes are needed by the slot map. Returns |
| 224 | * an error if the slot map file is too small. |
| 225 | */ |
| 226 | static int ocfs2_slot_map_physical_size(struct ocfs2_super *osb, |
| 227 | struct inode *inode, |
| 228 | unsigned long long *bytes) |
| 229 | { |
| 230 | unsigned long long bytes_needed; |
| 231 | |
| 232 | if (ocfs2_uses_extended_slot_map(osb)) { |
| 233 | bytes_needed = osb->max_slots * |
| 234 | sizeof(struct ocfs2_extended_slot); |
| 235 | } else { |
| 236 | bytes_needed = osb->max_slots * sizeof(__le16); |
| 237 | } |
| 238 | if (bytes_needed > i_size_read(inode)) { |
| 239 | mlog(ML_ERROR, |
| 240 | "Slot map file is too small! (size %llu, needed %llu)\n", |
| 241 | i_size_read(inode), bytes_needed); |
| 242 | return -ENOSPC; |
| 243 | } |
| 244 | |
| 245 | *bytes = bytes_needed; |
| 246 | return 0; |
| 247 | } |
| 248 | |
| 249 | /* try to find global node in the slot info. Returns -ENOENT |
| 250 | * if nothing is found. */ |
| 251 | static int __ocfs2_node_num_to_slot(struct ocfs2_slot_info *si, |
| 252 | unsigned int node_num) |
| 253 | { |
| 254 | int i, ret = -ENOENT; |
| 255 | |
| 256 | for(i = 0; i < si->si_num_slots; i++) { |
| 257 | if (si->si_slots[i].sl_valid && |
| 258 | (node_num == si->si_slots[i].sl_node_num)) { |
| 259 | ret = i; |
| 260 | break; |
| 261 | } |
| 262 | } |
| 263 | |
| 264 | return ret; |
| 265 | } |
| 266 | |
| 267 | static int __ocfs2_find_empty_slot(struct ocfs2_slot_info *si, |
| 268 | int preferred) |
| 269 | { |
| 270 | int i, ret = -ENOSPC; |
| 271 | |
| 272 | if ((preferred >= 0) && (preferred < si->si_num_slots)) { |
| 273 | if (!si->si_slots[preferred].sl_valid) { |
| 274 | ret = preferred; |
| 275 | goto out; |
| 276 | } |
| 277 | } |
| 278 | |
| 279 | for(i = 0; i < si->si_num_slots; i++) { |
| 280 | if (!si->si_slots[i].sl_valid) { |
| 281 | ret = i; |
| 282 | break; |
| 283 | } |
| 284 | } |
| 285 | out: |
| 286 | return ret; |
| 287 | } |
| 288 | |
| 289 | int ocfs2_node_num_to_slot(struct ocfs2_super *osb, unsigned int node_num) |
| 290 | { |
| 291 | int slot; |
| 292 | struct ocfs2_slot_info *si = osb->slot_info; |
| 293 | |
| 294 | spin_lock(&osb->osb_lock); |
| 295 | slot = __ocfs2_node_num_to_slot(si, node_num); |
| 296 | spin_unlock(&osb->osb_lock); |
| 297 | |
| 298 | return slot; |
| 299 | } |
| 300 | |
| 301 | int ocfs2_slot_to_node_num_locked(struct ocfs2_super *osb, int slot_num, |
| 302 | unsigned int *node_num) |
| 303 | { |
| 304 | struct ocfs2_slot_info *si = osb->slot_info; |
| 305 | |
| 306 | assert_spin_locked(&osb->osb_lock); |
| 307 | |
| 308 | BUG_ON(slot_num < 0); |
| 309 | BUG_ON(slot_num > osb->max_slots); |
| 310 | |
| 311 | if (!si->si_slots[slot_num].sl_valid) |
| 312 | return -ENOENT; |
| 313 | |
| 314 | *node_num = si->si_slots[slot_num].sl_node_num; |
| 315 | return 0; |
| 316 | } |
| 317 | |
| 318 | static void __ocfs2_free_slot_info(struct ocfs2_slot_info *si) |
| 319 | { |
| 320 | unsigned int i; |
| 321 | |
| 322 | if (si == NULL) |
| 323 | return; |
| 324 | |
| 325 | if (si->si_inode) |
| 326 | iput(si->si_inode); |
| 327 | if (si->si_bh) { |
| 328 | for (i = 0; i < si->si_blocks; i++) { |
| 329 | if (si->si_bh[i]) { |
| 330 | brelse(si->si_bh[i]); |
| 331 | si->si_bh[i] = NULL; |
| 332 | } |
| 333 | } |
| 334 | kfree(si->si_bh); |
| 335 | } |
| 336 | |
| 337 | kfree(si); |
| 338 | } |
| 339 | |
| 340 | int ocfs2_clear_slot(struct ocfs2_super *osb, int slot_num) |
| 341 | { |
| 342 | struct ocfs2_slot_info *si = osb->slot_info; |
| 343 | |
| 344 | if (si == NULL) |
| 345 | return 0; |
| 346 | |
| 347 | spin_lock(&osb->osb_lock); |
| 348 | ocfs2_invalidate_slot(si, slot_num); |
| 349 | spin_unlock(&osb->osb_lock); |
| 350 | |
| 351 | return ocfs2_update_disk_slot(osb, osb->slot_info, slot_num); |
| 352 | } |
| 353 | |
| 354 | static int ocfs2_map_slot_buffers(struct ocfs2_super *osb, |
| 355 | struct ocfs2_slot_info *si) |
| 356 | { |
| 357 | int status = 0; |
| 358 | u64 blkno; |
| 359 | unsigned long long blocks, bytes = 0; |
| 360 | unsigned int i; |
| 361 | struct buffer_head *bh; |
| 362 | |
| 363 | status = ocfs2_slot_map_physical_size(osb, si->si_inode, &bytes); |
| 364 | if (status) |
| 365 | goto bail; |
| 366 | |
| 367 | blocks = ocfs2_blocks_for_bytes(si->si_inode->i_sb, bytes); |
| 368 | BUG_ON(blocks > UINT_MAX); |
| 369 | si->si_blocks = blocks; |
| 370 | if (!si->si_blocks) |
| 371 | goto bail; |
| 372 | |
| 373 | if (si->si_extended) |
| 374 | si->si_slots_per_block = |
| 375 | (osb->sb->s_blocksize / |
| 376 | sizeof(struct ocfs2_extended_slot)); |
| 377 | else |
| 378 | si->si_slots_per_block = osb->sb->s_blocksize / sizeof(__le16); |
| 379 | |
| 380 | /* The size checks above should ensure this */ |
| 381 | BUG_ON((osb->max_slots / si->si_slots_per_block) > blocks); |
| 382 | |
| 383 | trace_ocfs2_map_slot_buffers(bytes, si->si_blocks); |
| 384 | |
| 385 | si->si_bh = kzalloc(sizeof(struct buffer_head *) * si->si_blocks, |
| 386 | GFP_KERNEL); |
| 387 | if (!si->si_bh) { |
| 388 | status = -ENOMEM; |
| 389 | mlog_errno(status); |
| 390 | goto bail; |
| 391 | } |
| 392 | |
| 393 | for (i = 0; i < si->si_blocks; i++) { |
| 394 | status = ocfs2_extent_map_get_blocks(si->si_inode, i, |
| 395 | &blkno, NULL, NULL); |
| 396 | if (status < 0) { |
| 397 | mlog_errno(status); |
| 398 | goto bail; |
| 399 | } |
| 400 | |
| 401 | trace_ocfs2_map_slot_buffers_block((unsigned long long)blkno, i); |
| 402 | |
| 403 | bh = NULL; /* Acquire a fresh bh */ |
| 404 | status = ocfs2_read_blocks(INODE_CACHE(si->si_inode), blkno, |
| 405 | 1, &bh, OCFS2_BH_IGNORE_CACHE, NULL); |
| 406 | if (status < 0) { |
| 407 | mlog_errno(status); |
| 408 | goto bail; |
| 409 | } |
| 410 | |
| 411 | si->si_bh[i] = bh; |
| 412 | } |
| 413 | |
| 414 | bail: |
| 415 | return status; |
| 416 | } |
| 417 | |
| 418 | int ocfs2_init_slot_info(struct ocfs2_super *osb) |
| 419 | { |
| 420 | int status; |
| 421 | struct inode *inode = NULL; |
| 422 | struct ocfs2_slot_info *si; |
| 423 | |
| 424 | si = kzalloc(sizeof(struct ocfs2_slot_info) + |
| 425 | (sizeof(struct ocfs2_slot) * osb->max_slots), |
| 426 | GFP_KERNEL); |
| 427 | if (!si) { |
| 428 | status = -ENOMEM; |
| 429 | mlog_errno(status); |
| 430 | goto bail; |
| 431 | } |
| 432 | |
| 433 | si->si_extended = ocfs2_uses_extended_slot_map(osb); |
| 434 | si->si_num_slots = osb->max_slots; |
| 435 | si->si_slots = (struct ocfs2_slot *)((char *)si + |
| 436 | sizeof(struct ocfs2_slot_info)); |
| 437 | |
| 438 | inode = ocfs2_get_system_file_inode(osb, SLOT_MAP_SYSTEM_INODE, |
| 439 | OCFS2_INVALID_SLOT); |
| 440 | if (!inode) { |
| 441 | status = -EINVAL; |
| 442 | mlog_errno(status); |
| 443 | goto bail; |
| 444 | } |
| 445 | |
| 446 | si->si_inode = inode; |
| 447 | status = ocfs2_map_slot_buffers(osb, si); |
| 448 | if (status < 0) { |
| 449 | mlog_errno(status); |
| 450 | goto bail; |
| 451 | } |
| 452 | |
| 453 | osb->slot_info = (struct ocfs2_slot_info *)si; |
| 454 | bail: |
| 455 | if (status < 0 && si) |
| 456 | __ocfs2_free_slot_info(si); |
| 457 | |
| 458 | return status; |
| 459 | } |
| 460 | |
| 461 | void ocfs2_free_slot_info(struct ocfs2_super *osb) |
| 462 | { |
| 463 | struct ocfs2_slot_info *si = osb->slot_info; |
| 464 | |
| 465 | osb->slot_info = NULL; |
| 466 | __ocfs2_free_slot_info(si); |
| 467 | } |
| 468 | |
| 469 | int ocfs2_find_slot(struct ocfs2_super *osb) |
| 470 | { |
| 471 | int status; |
| 472 | int slot; |
| 473 | struct ocfs2_slot_info *si; |
| 474 | |
| 475 | si = osb->slot_info; |
| 476 | |
| 477 | spin_lock(&osb->osb_lock); |
| 478 | ocfs2_update_slot_info(si); |
| 479 | |
| 480 | /* search for ourselves first and take the slot if it already |
| 481 | * exists. Perhaps we need to mark this in a variable for our |
| 482 | * own journal recovery? Possibly not, though we certainly |
| 483 | * need to warn to the user */ |
| 484 | slot = __ocfs2_node_num_to_slot(si, osb->node_num); |
| 485 | if (slot < 0) { |
| 486 | /* if no slot yet, then just take 1st available |
| 487 | * one. */ |
| 488 | slot = __ocfs2_find_empty_slot(si, osb->preferred_slot); |
| 489 | if (slot < 0) { |
| 490 | spin_unlock(&osb->osb_lock); |
| 491 | mlog(ML_ERROR, "no free slots available!\n"); |
| 492 | status = -EINVAL; |
| 493 | goto bail; |
| 494 | } |
| 495 | } else |
| 496 | printk(KERN_INFO "ocfs2: Slot %d on device (%s) was already " |
| 497 | "allocated to this node!\n", slot, osb->dev_str); |
| 498 | |
| 499 | ocfs2_set_slot(si, slot, osb->node_num); |
| 500 | osb->slot_num = slot; |
| 501 | spin_unlock(&osb->osb_lock); |
| 502 | |
| 503 | trace_ocfs2_find_slot(osb->slot_num); |
| 504 | |
| 505 | status = ocfs2_update_disk_slot(osb, si, osb->slot_num); |
| 506 | if (status < 0) |
| 507 | mlog_errno(status); |
| 508 | |
| 509 | bail: |
| 510 | return status; |
| 511 | } |
| 512 | |
| 513 | void ocfs2_put_slot(struct ocfs2_super *osb) |
| 514 | { |
| 515 | int status, slot_num; |
| 516 | struct ocfs2_slot_info *si = osb->slot_info; |
| 517 | |
| 518 | if (!si) |
| 519 | return; |
| 520 | |
| 521 | spin_lock(&osb->osb_lock); |
| 522 | ocfs2_update_slot_info(si); |
| 523 | |
| 524 | slot_num = osb->slot_num; |
| 525 | ocfs2_invalidate_slot(si, osb->slot_num); |
| 526 | osb->slot_num = OCFS2_INVALID_SLOT; |
| 527 | spin_unlock(&osb->osb_lock); |
| 528 | |
| 529 | status = ocfs2_update_disk_slot(osb, si, slot_num); |
| 530 | if (status < 0) { |
| 531 | mlog_errno(status); |
| 532 | goto bail; |
| 533 | } |
| 534 | |
| 535 | bail: |
| 536 | ocfs2_free_slot_info(osb); |
| 537 | } |
| 538 | |