Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1 | /* -*- mode: c; c-basic-offset: 8; -*- |
| 2 | * vim: noexpandtab sw=8 ts=8 sts=0: |
| 3 | * |
| 4 | * journal.c |
| 5 | * |
| 6 | * Defines functions of journalling api |
| 7 | * |
| 8 | * Copyright (C) 2003, 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/fs.h> |
| 27 | #include <linux/types.h> |
| 28 | #include <linux/slab.h> |
| 29 | #include <linux/highmem.h> |
| 30 | #include <linux/kthread.h> |
| 31 | |
| 32 | #define MLOG_MASK_PREFIX ML_JOURNAL |
| 33 | #include <cluster/masklog.h> |
| 34 | |
| 35 | #include "ocfs2.h" |
| 36 | |
| 37 | #include "alloc.h" |
| 38 | #include "dlmglue.h" |
| 39 | #include "extent_map.h" |
| 40 | #include "heartbeat.h" |
| 41 | #include "inode.h" |
| 42 | #include "journal.h" |
| 43 | #include "localalloc.h" |
| 44 | #include "namei.h" |
| 45 | #include "slot_map.h" |
| 46 | #include "super.h" |
| 47 | #include "vote.h" |
| 48 | #include "sysfile.h" |
| 49 | |
| 50 | #include "buffer_head_io.h" |
| 51 | |
Ingo Molnar | 34af946 | 2006-06-27 02:53:55 -0700 | [diff] [blame] | 52 | DEFINE_SPINLOCK(trans_inc_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 53 | |
| 54 | static int ocfs2_force_read_journal(struct inode *inode); |
| 55 | static int ocfs2_recover_node(struct ocfs2_super *osb, |
| 56 | int node_num); |
| 57 | static int __ocfs2_recovery_thread(void *arg); |
| 58 | static int ocfs2_commit_cache(struct ocfs2_super *osb); |
| 59 | static int ocfs2_wait_on_mount(struct ocfs2_super *osb); |
| 60 | static void ocfs2_handle_cleanup_locks(struct ocfs2_journal *journal, |
| 61 | struct ocfs2_journal_handle *handle); |
| 62 | static void ocfs2_commit_unstarted_handle(struct ocfs2_journal_handle *handle); |
| 63 | static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, |
| 64 | int dirty); |
| 65 | static int ocfs2_trylock_journal(struct ocfs2_super *osb, |
| 66 | int slot_num); |
| 67 | static int ocfs2_recover_orphans(struct ocfs2_super *osb, |
| 68 | int slot); |
| 69 | static int ocfs2_commit_thread(void *arg); |
| 70 | |
| 71 | static int ocfs2_commit_cache(struct ocfs2_super *osb) |
| 72 | { |
| 73 | int status = 0; |
| 74 | unsigned int flushed; |
| 75 | unsigned long old_id; |
| 76 | struct ocfs2_journal *journal = NULL; |
| 77 | |
| 78 | mlog_entry_void(); |
| 79 | |
| 80 | journal = osb->journal; |
| 81 | |
| 82 | /* Flush all pending commits and checkpoint the journal. */ |
| 83 | down_write(&journal->j_trans_barrier); |
| 84 | |
| 85 | if (atomic_read(&journal->j_num_trans) == 0) { |
| 86 | up_write(&journal->j_trans_barrier); |
| 87 | mlog(0, "No transactions for me to flush!\n"); |
| 88 | goto finally; |
| 89 | } |
| 90 | |
| 91 | journal_lock_updates(journal->j_journal); |
| 92 | status = journal_flush(journal->j_journal); |
| 93 | journal_unlock_updates(journal->j_journal); |
| 94 | if (status < 0) { |
| 95 | up_write(&journal->j_trans_barrier); |
| 96 | mlog_errno(status); |
| 97 | goto finally; |
| 98 | } |
| 99 | |
| 100 | old_id = ocfs2_inc_trans_id(journal); |
| 101 | |
| 102 | flushed = atomic_read(&journal->j_num_trans); |
| 103 | atomic_set(&journal->j_num_trans, 0); |
| 104 | up_write(&journal->j_trans_barrier); |
| 105 | |
| 106 | mlog(0, "commit_thread: flushed transaction %lu (%u handles)\n", |
| 107 | journal->j_trans_id, flushed); |
| 108 | |
| 109 | ocfs2_kick_vote_thread(osb); |
| 110 | wake_up(&journal->j_checkpointed); |
| 111 | finally: |
| 112 | mlog_exit(status); |
| 113 | return status; |
| 114 | } |
| 115 | |
| 116 | struct ocfs2_journal_handle *ocfs2_alloc_handle(struct ocfs2_super *osb) |
| 117 | { |
| 118 | struct ocfs2_journal_handle *retval = NULL; |
| 119 | |
Sunil Mushran | afae00ab | 2006-04-12 14:37:00 -0700 | [diff] [blame] | 120 | retval = kcalloc(1, sizeof(*retval), GFP_NOFS); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 121 | if (!retval) { |
| 122 | mlog(ML_ERROR, "Failed to allocate memory for journal " |
| 123 | "handle!\n"); |
| 124 | return NULL; |
| 125 | } |
| 126 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 127 | retval->num_locks = 0; |
| 128 | retval->k_handle = NULL; |
| 129 | |
| 130 | INIT_LIST_HEAD(&retval->locks); |
| 131 | INIT_LIST_HEAD(&retval->inode_list); |
| 132 | retval->journal = osb->journal; |
| 133 | |
| 134 | return retval; |
| 135 | } |
| 136 | |
| 137 | /* pass it NULL and it will allocate a new handle object for you. If |
| 138 | * you pass it a handle however, it may still return error, in which |
| 139 | * case it has free'd the passed handle for you. */ |
| 140 | struct ocfs2_journal_handle *ocfs2_start_trans(struct ocfs2_super *osb, |
| 141 | struct ocfs2_journal_handle *handle, |
| 142 | int max_buffs) |
| 143 | { |
| 144 | int ret; |
| 145 | journal_t *journal = osb->journal->j_journal; |
| 146 | |
| 147 | mlog_entry("(max_buffs = %d)\n", max_buffs); |
| 148 | |
Eric Sesterhenn / snakebyte | ebdec83 | 2006-01-27 10:32:52 +0100 | [diff] [blame] | 149 | BUG_ON(!osb || !osb->journal->j_journal); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 150 | |
| 151 | if (ocfs2_is_hard_readonly(osb)) { |
| 152 | ret = -EROFS; |
| 153 | goto done_free; |
| 154 | } |
| 155 | |
| 156 | BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); |
| 157 | BUG_ON(max_buffs <= 0); |
| 158 | |
| 159 | /* JBD might support this, but our journalling code doesn't yet. */ |
| 160 | if (journal_current_handle()) { |
| 161 | mlog(ML_ERROR, "Recursive transaction attempted!\n"); |
| 162 | BUG(); |
| 163 | } |
| 164 | |
| 165 | if (!handle) |
| 166 | handle = ocfs2_alloc_handle(osb); |
| 167 | if (!handle) { |
| 168 | ret = -ENOMEM; |
| 169 | mlog(ML_ERROR, "Failed to allocate memory for journal " |
| 170 | "handle!\n"); |
| 171 | goto done_free; |
| 172 | } |
| 173 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 174 | down_read(&osb->journal->j_trans_barrier); |
| 175 | |
| 176 | /* actually start the transaction now */ |
| 177 | handle->k_handle = journal_start(journal, max_buffs); |
| 178 | if (IS_ERR(handle->k_handle)) { |
| 179 | up_read(&osb->journal->j_trans_barrier); |
| 180 | |
| 181 | ret = PTR_ERR(handle->k_handle); |
| 182 | handle->k_handle = NULL; |
| 183 | mlog_errno(ret); |
| 184 | |
| 185 | if (is_journal_aborted(journal)) { |
| 186 | ocfs2_abort(osb->sb, "Detected aborted journal"); |
| 187 | ret = -EROFS; |
| 188 | } |
| 189 | goto done_free; |
| 190 | } |
| 191 | |
| 192 | atomic_inc(&(osb->journal->j_num_trans)); |
| 193 | handle->flags |= OCFS2_HANDLE_STARTED; |
| 194 | |
| 195 | mlog_exit_ptr(handle); |
| 196 | return handle; |
| 197 | |
| 198 | done_free: |
| 199 | if (handle) |
| 200 | ocfs2_commit_unstarted_handle(handle); /* will kfree handle */ |
| 201 | |
| 202 | mlog_exit(ret); |
| 203 | return ERR_PTR(ret); |
| 204 | } |
| 205 | |
| 206 | void ocfs2_handle_add_inode(struct ocfs2_journal_handle *handle, |
| 207 | struct inode *inode) |
| 208 | { |
| 209 | BUG_ON(!handle); |
| 210 | BUG_ON(!inode); |
| 211 | |
| 212 | atomic_inc(&inode->i_count); |
| 213 | |
| 214 | /* we're obviously changing it... */ |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 215 | mutex_lock(&inode->i_mutex); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 216 | |
| 217 | /* sanity check */ |
| 218 | BUG_ON(OCFS2_I(inode)->ip_handle); |
| 219 | BUG_ON(!list_empty(&OCFS2_I(inode)->ip_handle_list)); |
| 220 | |
| 221 | OCFS2_I(inode)->ip_handle = handle; |
Akinobu Mita | f116629 | 2006-06-26 00:24:46 -0700 | [diff] [blame] | 222 | list_move_tail(&(OCFS2_I(inode)->ip_handle_list), &(handle->inode_list)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 223 | } |
| 224 | |
| 225 | static void ocfs2_handle_unlock_inodes(struct ocfs2_journal_handle *handle) |
| 226 | { |
| 227 | struct list_head *p, *n; |
| 228 | struct inode *inode; |
| 229 | struct ocfs2_inode_info *oi; |
| 230 | |
| 231 | list_for_each_safe(p, n, &handle->inode_list) { |
| 232 | oi = list_entry(p, struct ocfs2_inode_info, |
| 233 | ip_handle_list); |
| 234 | inode = &oi->vfs_inode; |
| 235 | |
| 236 | OCFS2_I(inode)->ip_handle = NULL; |
| 237 | list_del_init(&OCFS2_I(inode)->ip_handle_list); |
| 238 | |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 239 | mutex_unlock(&inode->i_mutex); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 240 | iput(inode); |
| 241 | } |
| 242 | } |
| 243 | |
| 244 | /* This is trivial so we do it out of the main commit |
| 245 | * paths. Beware, it can be called from start_trans too! */ |
| 246 | static void ocfs2_commit_unstarted_handle(struct ocfs2_journal_handle *handle) |
| 247 | { |
| 248 | mlog_entry_void(); |
| 249 | |
| 250 | BUG_ON(handle->flags & OCFS2_HANDLE_STARTED); |
| 251 | |
| 252 | ocfs2_handle_unlock_inodes(handle); |
| 253 | /* You are allowed to add journal locks before the transaction |
| 254 | * has started. */ |
| 255 | ocfs2_handle_cleanup_locks(handle->journal, handle); |
| 256 | |
| 257 | kfree(handle); |
| 258 | |
| 259 | mlog_exit_void(); |
| 260 | } |
| 261 | |
| 262 | void ocfs2_commit_trans(struct ocfs2_journal_handle *handle) |
| 263 | { |
| 264 | handle_t *jbd_handle; |
| 265 | int retval; |
| 266 | struct ocfs2_journal *journal = handle->journal; |
| 267 | |
| 268 | mlog_entry_void(); |
| 269 | |
| 270 | BUG_ON(!handle); |
| 271 | |
| 272 | if (!(handle->flags & OCFS2_HANDLE_STARTED)) { |
| 273 | ocfs2_commit_unstarted_handle(handle); |
| 274 | mlog_exit_void(); |
| 275 | return; |
| 276 | } |
| 277 | |
| 278 | /* release inode semaphores we took during this transaction */ |
| 279 | ocfs2_handle_unlock_inodes(handle); |
| 280 | |
| 281 | /* ocfs2_extend_trans may have had to call journal_restart |
| 282 | * which will always commit the transaction, but may return |
| 283 | * error for any number of reasons. If this is the case, we |
| 284 | * clear k_handle as it's not valid any more. */ |
| 285 | if (handle->k_handle) { |
| 286 | jbd_handle = handle->k_handle; |
| 287 | |
| 288 | if (handle->flags & OCFS2_HANDLE_SYNC) |
| 289 | jbd_handle->h_sync = 1; |
| 290 | else |
| 291 | jbd_handle->h_sync = 0; |
| 292 | |
| 293 | /* actually stop the transaction. if we've set h_sync, |
| 294 | * it'll have been committed when we return */ |
| 295 | retval = journal_stop(jbd_handle); |
| 296 | if (retval < 0) { |
| 297 | mlog_errno(retval); |
| 298 | mlog(ML_ERROR, "Could not commit transaction\n"); |
| 299 | BUG(); |
| 300 | } |
| 301 | |
| 302 | handle->k_handle = NULL; /* it's been free'd in journal_stop */ |
| 303 | } |
| 304 | |
| 305 | ocfs2_handle_cleanup_locks(journal, handle); |
| 306 | |
| 307 | up_read(&journal->j_trans_barrier); |
| 308 | |
| 309 | kfree(handle); |
| 310 | mlog_exit_void(); |
| 311 | } |
| 312 | |
| 313 | /* |
| 314 | * 'nblocks' is what you want to add to the current |
| 315 | * transaction. extend_trans will either extend the current handle by |
| 316 | * nblocks, or commit it and start a new one with nblocks credits. |
| 317 | * |
| 318 | * WARNING: This will not release any semaphores or disk locks taken |
| 319 | * during the transaction, so make sure they were taken *before* |
| 320 | * start_trans or we'll have ordering deadlocks. |
| 321 | * |
| 322 | * WARNING2: Note that we do *not* drop j_trans_barrier here. This is |
| 323 | * good because transaction ids haven't yet been recorded on the |
| 324 | * cluster locks associated with this handle. |
| 325 | */ |
| 326 | int ocfs2_extend_trans(struct ocfs2_journal_handle *handle, |
| 327 | int nblocks) |
| 328 | { |
| 329 | int status; |
| 330 | |
| 331 | BUG_ON(!handle); |
| 332 | BUG_ON(!(handle->flags & OCFS2_HANDLE_STARTED)); |
| 333 | BUG_ON(!nblocks); |
| 334 | |
| 335 | mlog_entry_void(); |
| 336 | |
| 337 | mlog(0, "Trying to extend transaction by %d blocks\n", nblocks); |
| 338 | |
| 339 | status = journal_extend(handle->k_handle, nblocks); |
| 340 | if (status < 0) { |
| 341 | mlog_errno(status); |
| 342 | goto bail; |
| 343 | } |
| 344 | |
| 345 | if (status > 0) { |
| 346 | mlog(0, "journal_extend failed, trying journal_restart\n"); |
| 347 | status = journal_restart(handle->k_handle, nblocks); |
| 348 | if (status < 0) { |
| 349 | handle->k_handle = NULL; |
| 350 | mlog_errno(status); |
| 351 | goto bail; |
| 352 | } |
Mark Fasheh | 01ddf1e | 2006-10-05 13:54:39 -0700 | [diff] [blame^] | 353 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 354 | |
| 355 | status = 0; |
| 356 | bail: |
| 357 | |
| 358 | mlog_exit(status); |
| 359 | return status; |
| 360 | } |
| 361 | |
| 362 | int ocfs2_journal_access(struct ocfs2_journal_handle *handle, |
| 363 | struct inode *inode, |
| 364 | struct buffer_head *bh, |
| 365 | int type) |
| 366 | { |
| 367 | int status; |
| 368 | |
| 369 | BUG_ON(!inode); |
| 370 | BUG_ON(!handle); |
| 371 | BUG_ON(!bh); |
| 372 | BUG_ON(!(handle->flags & OCFS2_HANDLE_STARTED)); |
| 373 | |
Badari Pulavarty | 205f87f | 2006-03-26 01:38:00 -0800 | [diff] [blame] | 374 | mlog_entry("bh->b_blocknr=%llu, type=%d (\"%s\"), bh->b_size = %zu\n", |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 375 | (unsigned long long)bh->b_blocknr, type, |
| 376 | (type == OCFS2_JOURNAL_ACCESS_CREATE) ? |
| 377 | "OCFS2_JOURNAL_ACCESS_CREATE" : |
| 378 | "OCFS2_JOURNAL_ACCESS_WRITE", |
| 379 | bh->b_size); |
| 380 | |
| 381 | /* we can safely remove this assertion after testing. */ |
| 382 | if (!buffer_uptodate(bh)) { |
| 383 | mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); |
| 384 | mlog(ML_ERROR, "b_blocknr=%llu\n", |
| 385 | (unsigned long long)bh->b_blocknr); |
| 386 | BUG(); |
| 387 | } |
| 388 | |
| 389 | /* Set the current transaction information on the inode so |
| 390 | * that the locking code knows whether it can drop it's locks |
| 391 | * on this inode or not. We're protected from the commit |
| 392 | * thread updating the current transaction id until |
| 393 | * ocfs2_commit_trans() because ocfs2_start_trans() took |
| 394 | * j_trans_barrier for us. */ |
| 395 | ocfs2_set_inode_lock_trans(OCFS2_SB(inode->i_sb)->journal, inode); |
| 396 | |
Mark Fasheh | 251b6ec | 2006-01-10 15:41:43 -0800 | [diff] [blame] | 397 | mutex_lock(&OCFS2_I(inode)->ip_io_mutex); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 398 | switch (type) { |
| 399 | case OCFS2_JOURNAL_ACCESS_CREATE: |
| 400 | case OCFS2_JOURNAL_ACCESS_WRITE: |
| 401 | status = journal_get_write_access(handle->k_handle, bh); |
| 402 | break; |
| 403 | |
| 404 | case OCFS2_JOURNAL_ACCESS_UNDO: |
| 405 | status = journal_get_undo_access(handle->k_handle, bh); |
| 406 | break; |
| 407 | |
| 408 | default: |
| 409 | status = -EINVAL; |
| 410 | mlog(ML_ERROR, "Uknown access type!\n"); |
| 411 | } |
Mark Fasheh | 251b6ec | 2006-01-10 15:41:43 -0800 | [diff] [blame] | 412 | mutex_unlock(&OCFS2_I(inode)->ip_io_mutex); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 413 | |
| 414 | if (status < 0) |
| 415 | mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", |
| 416 | status, type); |
| 417 | |
| 418 | mlog_exit(status); |
| 419 | return status; |
| 420 | } |
| 421 | |
| 422 | int ocfs2_journal_dirty(struct ocfs2_journal_handle *handle, |
| 423 | struct buffer_head *bh) |
| 424 | { |
| 425 | int status; |
| 426 | |
| 427 | BUG_ON(!(handle->flags & OCFS2_HANDLE_STARTED)); |
| 428 | |
| 429 | mlog_entry("(bh->b_blocknr=%llu)\n", |
| 430 | (unsigned long long)bh->b_blocknr); |
| 431 | |
| 432 | status = journal_dirty_metadata(handle->k_handle, bh); |
| 433 | if (status < 0) |
| 434 | mlog(ML_ERROR, "Could not dirty metadata buffer. " |
| 435 | "(bh->b_blocknr=%llu)\n", |
| 436 | (unsigned long long)bh->b_blocknr); |
| 437 | |
| 438 | mlog_exit(status); |
| 439 | return status; |
| 440 | } |
| 441 | |
| 442 | int ocfs2_journal_dirty_data(handle_t *handle, |
| 443 | struct buffer_head *bh) |
| 444 | { |
| 445 | int err = journal_dirty_data(handle, bh); |
| 446 | if (err) |
| 447 | mlog_errno(err); |
| 448 | /* TODO: When we can handle it, abort the handle and go RO on |
| 449 | * error here. */ |
| 450 | |
| 451 | return err; |
| 452 | } |
| 453 | |
| 454 | /* We always assume you're adding a metadata lock at level 'ex' */ |
| 455 | int ocfs2_handle_add_lock(struct ocfs2_journal_handle *handle, |
| 456 | struct inode *inode) |
| 457 | { |
| 458 | int status; |
| 459 | struct ocfs2_journal_lock *lock; |
| 460 | |
| 461 | BUG_ON(!inode); |
| 462 | |
| 463 | lock = kmem_cache_alloc(ocfs2_lock_cache, GFP_NOFS); |
| 464 | if (!lock) { |
| 465 | status = -ENOMEM; |
| 466 | mlog_errno(-ENOMEM); |
| 467 | goto bail; |
| 468 | } |
| 469 | |
| 470 | if (!igrab(inode)) |
| 471 | BUG(); |
| 472 | lock->jl_inode = inode; |
| 473 | |
| 474 | list_add_tail(&(lock->jl_lock_list), &(handle->locks)); |
| 475 | handle->num_locks++; |
| 476 | |
| 477 | status = 0; |
| 478 | bail: |
| 479 | mlog_exit(status); |
| 480 | return status; |
| 481 | } |
| 482 | |
| 483 | static void ocfs2_handle_cleanup_locks(struct ocfs2_journal *journal, |
| 484 | struct ocfs2_journal_handle *handle) |
| 485 | { |
| 486 | struct list_head *p, *n; |
| 487 | struct ocfs2_journal_lock *lock; |
| 488 | struct inode *inode; |
| 489 | |
| 490 | list_for_each_safe(p, n, &(handle->locks)) { |
| 491 | lock = list_entry(p, struct ocfs2_journal_lock, |
| 492 | jl_lock_list); |
| 493 | list_del(&lock->jl_lock_list); |
| 494 | handle->num_locks--; |
| 495 | |
| 496 | inode = lock->jl_inode; |
| 497 | ocfs2_meta_unlock(inode, 1); |
| 498 | if (atomic_read(&inode->i_count) == 1) |
| 499 | mlog(ML_ERROR, |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 500 | "Inode %llu, I'm doing a last iput for!", |
| 501 | (unsigned long long)OCFS2_I(inode)->ip_blkno); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 502 | iput(inode); |
| 503 | kmem_cache_free(ocfs2_lock_cache, lock); |
| 504 | } |
| 505 | } |
| 506 | |
| 507 | #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * 5) |
| 508 | |
| 509 | void ocfs2_set_journal_params(struct ocfs2_super *osb) |
| 510 | { |
| 511 | journal_t *journal = osb->journal->j_journal; |
| 512 | |
| 513 | spin_lock(&journal->j_state_lock); |
| 514 | journal->j_commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; |
| 515 | if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) |
| 516 | journal->j_flags |= JFS_BARRIER; |
| 517 | else |
| 518 | journal->j_flags &= ~JFS_BARRIER; |
| 519 | spin_unlock(&journal->j_state_lock); |
| 520 | } |
| 521 | |
| 522 | int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) |
| 523 | { |
| 524 | int status = -1; |
| 525 | struct inode *inode = NULL; /* the journal inode */ |
| 526 | journal_t *j_journal = NULL; |
| 527 | struct ocfs2_dinode *di = NULL; |
| 528 | struct buffer_head *bh = NULL; |
| 529 | struct ocfs2_super *osb; |
| 530 | int meta_lock = 0; |
| 531 | |
| 532 | mlog_entry_void(); |
| 533 | |
| 534 | BUG_ON(!journal); |
| 535 | |
| 536 | osb = journal->j_osb; |
| 537 | |
| 538 | /* already have the inode for our journal */ |
| 539 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, |
| 540 | osb->slot_num); |
| 541 | if (inode == NULL) { |
| 542 | status = -EACCES; |
| 543 | mlog_errno(status); |
| 544 | goto done; |
| 545 | } |
| 546 | if (is_bad_inode(inode)) { |
| 547 | mlog(ML_ERROR, "access error (bad inode)\n"); |
| 548 | iput(inode); |
| 549 | inode = NULL; |
| 550 | status = -EACCES; |
| 551 | goto done; |
| 552 | } |
| 553 | |
| 554 | SET_INODE_JOURNAL(inode); |
| 555 | OCFS2_I(inode)->ip_open_count++; |
| 556 | |
Mark Fasheh | 6eff579 | 2006-01-18 10:31:47 -0800 | [diff] [blame] | 557 | /* Skip recovery waits here - journal inode metadata never |
| 558 | * changes in a live cluster so it can be considered an |
| 559 | * exception to the rule. */ |
| 560 | status = ocfs2_meta_lock_full(inode, NULL, &bh, 1, |
| 561 | OCFS2_META_LOCK_RECOVERY); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 562 | if (status < 0) { |
| 563 | if (status != -ERESTARTSYS) |
| 564 | mlog(ML_ERROR, "Could not get lock on journal!\n"); |
| 565 | goto done; |
| 566 | } |
| 567 | |
| 568 | meta_lock = 1; |
| 569 | di = (struct ocfs2_dinode *)bh->b_data; |
| 570 | |
| 571 | if (inode->i_size < OCFS2_MIN_JOURNAL_SIZE) { |
| 572 | mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", |
| 573 | inode->i_size); |
| 574 | status = -EINVAL; |
| 575 | goto done; |
| 576 | } |
| 577 | |
| 578 | mlog(0, "inode->i_size = %lld\n", inode->i_size); |
Andrew Morton | 5515eff | 2006-03-26 01:37:53 -0800 | [diff] [blame] | 579 | mlog(0, "inode->i_blocks = %llu\n", |
| 580 | (unsigned long long)inode->i_blocks); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 581 | mlog(0, "inode->ip_clusters = %u\n", OCFS2_I(inode)->ip_clusters); |
| 582 | |
| 583 | /* call the kernels journal init function now */ |
| 584 | j_journal = journal_init_inode(inode); |
| 585 | if (j_journal == NULL) { |
| 586 | mlog(ML_ERROR, "Linux journal layer error\n"); |
| 587 | status = -EINVAL; |
| 588 | goto done; |
| 589 | } |
| 590 | |
| 591 | mlog(0, "Returned from journal_init_inode\n"); |
| 592 | mlog(0, "j_journal->j_maxlen = %u\n", j_journal->j_maxlen); |
| 593 | |
| 594 | *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & |
| 595 | OCFS2_JOURNAL_DIRTY_FL); |
| 596 | |
| 597 | journal->j_journal = j_journal; |
| 598 | journal->j_inode = inode; |
| 599 | journal->j_bh = bh; |
| 600 | |
| 601 | ocfs2_set_journal_params(osb); |
| 602 | |
| 603 | journal->j_state = OCFS2_JOURNAL_LOADED; |
| 604 | |
| 605 | status = 0; |
| 606 | done: |
| 607 | if (status < 0) { |
| 608 | if (meta_lock) |
| 609 | ocfs2_meta_unlock(inode, 1); |
| 610 | if (bh != NULL) |
| 611 | brelse(bh); |
| 612 | if (inode) { |
| 613 | OCFS2_I(inode)->ip_open_count--; |
| 614 | iput(inode); |
| 615 | } |
| 616 | } |
| 617 | |
| 618 | mlog_exit(status); |
| 619 | return status; |
| 620 | } |
| 621 | |
| 622 | static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, |
| 623 | int dirty) |
| 624 | { |
| 625 | int status; |
| 626 | unsigned int flags; |
| 627 | struct ocfs2_journal *journal = osb->journal; |
| 628 | struct buffer_head *bh = journal->j_bh; |
| 629 | struct ocfs2_dinode *fe; |
| 630 | |
| 631 | mlog_entry_void(); |
| 632 | |
| 633 | fe = (struct ocfs2_dinode *)bh->b_data; |
| 634 | if (!OCFS2_IS_VALID_DINODE(fe)) { |
| 635 | /* This is called from startup/shutdown which will |
| 636 | * handle the errors in a specific manner, so no need |
| 637 | * to call ocfs2_error() here. */ |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 638 | mlog(ML_ERROR, "Journal dinode %llu has invalid " |
| 639 | "signature: %.*s", (unsigned long long)fe->i_blkno, 7, |
| 640 | fe->i_signature); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 641 | status = -EIO; |
| 642 | goto out; |
| 643 | } |
| 644 | |
| 645 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); |
| 646 | if (dirty) |
| 647 | flags |= OCFS2_JOURNAL_DIRTY_FL; |
| 648 | else |
| 649 | flags &= ~OCFS2_JOURNAL_DIRTY_FL; |
| 650 | fe->id1.journal1.ij_flags = cpu_to_le32(flags); |
| 651 | |
| 652 | status = ocfs2_write_block(osb, bh, journal->j_inode); |
| 653 | if (status < 0) |
| 654 | mlog_errno(status); |
| 655 | |
| 656 | out: |
| 657 | mlog_exit(status); |
| 658 | return status; |
| 659 | } |
| 660 | |
| 661 | /* |
| 662 | * If the journal has been kmalloc'd it needs to be freed after this |
| 663 | * call. |
| 664 | */ |
| 665 | void ocfs2_journal_shutdown(struct ocfs2_super *osb) |
| 666 | { |
| 667 | struct ocfs2_journal *journal = NULL; |
| 668 | int status = 0; |
| 669 | struct inode *inode = NULL; |
| 670 | int num_running_trans = 0; |
| 671 | |
| 672 | mlog_entry_void(); |
| 673 | |
Eric Sesterhenn / snakebyte | ebdec83 | 2006-01-27 10:32:52 +0100 | [diff] [blame] | 674 | BUG_ON(!osb); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 675 | |
| 676 | journal = osb->journal; |
| 677 | if (!journal) |
| 678 | goto done; |
| 679 | |
| 680 | inode = journal->j_inode; |
| 681 | |
| 682 | if (journal->j_state != OCFS2_JOURNAL_LOADED) |
| 683 | goto done; |
| 684 | |
| 685 | /* need to inc inode use count as journal_destroy will iput. */ |
| 686 | if (!igrab(inode)) |
| 687 | BUG(); |
| 688 | |
| 689 | num_running_trans = atomic_read(&(osb->journal->j_num_trans)); |
| 690 | if (num_running_trans > 0) |
| 691 | mlog(0, "Shutting down journal: must wait on %d " |
| 692 | "running transactions!\n", |
| 693 | num_running_trans); |
| 694 | |
| 695 | /* Do a commit_cache here. It will flush our journal, *and* |
| 696 | * release any locks that are still held. |
| 697 | * set the SHUTDOWN flag and release the trans lock. |
| 698 | * the commit thread will take the trans lock for us below. */ |
| 699 | journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; |
| 700 | |
| 701 | /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not |
| 702 | * drop the trans_lock (which we want to hold until we |
| 703 | * completely destroy the journal. */ |
| 704 | if (osb->commit_task) { |
| 705 | /* Wait for the commit thread */ |
| 706 | mlog(0, "Waiting for ocfs2commit to exit....\n"); |
| 707 | kthread_stop(osb->commit_task); |
| 708 | osb->commit_task = NULL; |
| 709 | } |
| 710 | |
| 711 | BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); |
| 712 | |
| 713 | status = ocfs2_journal_toggle_dirty(osb, 0); |
| 714 | if (status < 0) |
| 715 | mlog_errno(status); |
| 716 | |
| 717 | /* Shutdown the kernel journal system */ |
| 718 | journal_destroy(journal->j_journal); |
| 719 | |
| 720 | OCFS2_I(inode)->ip_open_count--; |
| 721 | |
| 722 | /* unlock our journal */ |
| 723 | ocfs2_meta_unlock(inode, 1); |
| 724 | |
| 725 | brelse(journal->j_bh); |
| 726 | journal->j_bh = NULL; |
| 727 | |
| 728 | journal->j_state = OCFS2_JOURNAL_FREE; |
| 729 | |
| 730 | // up_write(&journal->j_trans_barrier); |
| 731 | done: |
| 732 | if (inode) |
| 733 | iput(inode); |
| 734 | mlog_exit_void(); |
| 735 | } |
| 736 | |
| 737 | static void ocfs2_clear_journal_error(struct super_block *sb, |
| 738 | journal_t *journal, |
| 739 | int slot) |
| 740 | { |
| 741 | int olderr; |
| 742 | |
| 743 | olderr = journal_errno(journal); |
| 744 | if (olderr) { |
| 745 | mlog(ML_ERROR, "File system error %d recorded in " |
| 746 | "journal %u.\n", olderr, slot); |
| 747 | mlog(ML_ERROR, "File system on device %s needs checking.\n", |
| 748 | sb->s_id); |
| 749 | |
| 750 | journal_ack_err(journal); |
| 751 | journal_clear_err(journal); |
| 752 | } |
| 753 | } |
| 754 | |
| 755 | int ocfs2_journal_load(struct ocfs2_journal *journal) |
| 756 | { |
| 757 | int status = 0; |
| 758 | struct ocfs2_super *osb; |
| 759 | |
| 760 | mlog_entry_void(); |
| 761 | |
| 762 | if (!journal) |
| 763 | BUG(); |
| 764 | |
| 765 | osb = journal->j_osb; |
| 766 | |
| 767 | status = journal_load(journal->j_journal); |
| 768 | if (status < 0) { |
| 769 | mlog(ML_ERROR, "Failed to load journal!\n"); |
| 770 | goto done; |
| 771 | } |
| 772 | |
| 773 | ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); |
| 774 | |
| 775 | status = ocfs2_journal_toggle_dirty(osb, 1); |
| 776 | if (status < 0) { |
| 777 | mlog_errno(status); |
| 778 | goto done; |
| 779 | } |
| 780 | |
| 781 | /* Launch the commit thread */ |
Mark Fasheh | 7842704 | 2006-05-04 12:03:26 -0700 | [diff] [blame] | 782 | osb->commit_task = kthread_run(ocfs2_commit_thread, osb, "ocfs2cmt"); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 783 | if (IS_ERR(osb->commit_task)) { |
| 784 | status = PTR_ERR(osb->commit_task); |
| 785 | osb->commit_task = NULL; |
| 786 | mlog(ML_ERROR, "unable to launch ocfs2commit thread, error=%d", |
| 787 | status); |
| 788 | goto done; |
| 789 | } |
| 790 | |
| 791 | done: |
| 792 | mlog_exit(status); |
| 793 | return status; |
| 794 | } |
| 795 | |
| 796 | |
| 797 | /* 'full' flag tells us whether we clear out all blocks or if we just |
| 798 | * mark the journal clean */ |
| 799 | int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) |
| 800 | { |
| 801 | int status; |
| 802 | |
| 803 | mlog_entry_void(); |
| 804 | |
Eric Sesterhenn / snakebyte | ebdec83 | 2006-01-27 10:32:52 +0100 | [diff] [blame] | 805 | BUG_ON(!journal); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 806 | |
| 807 | status = journal_wipe(journal->j_journal, full); |
| 808 | if (status < 0) { |
| 809 | mlog_errno(status); |
| 810 | goto bail; |
| 811 | } |
| 812 | |
| 813 | status = ocfs2_journal_toggle_dirty(journal->j_osb, 0); |
| 814 | if (status < 0) |
| 815 | mlog_errno(status); |
| 816 | |
| 817 | bail: |
| 818 | mlog_exit(status); |
| 819 | return status; |
| 820 | } |
| 821 | |
| 822 | /* |
| 823 | * JBD Might read a cached version of another nodes journal file. We |
| 824 | * don't want this as this file changes often and we get no |
| 825 | * notification on those changes. The only way to be sure that we've |
| 826 | * got the most up to date version of those blocks then is to force |
| 827 | * read them off disk. Just searching through the buffer cache won't |
| 828 | * work as there may be pages backing this file which are still marked |
| 829 | * up to date. We know things can't change on this file underneath us |
| 830 | * as we have the lock by now :) |
| 831 | */ |
| 832 | static int ocfs2_force_read_journal(struct inode *inode) |
| 833 | { |
| 834 | int status = 0; |
| 835 | int i, p_blocks; |
| 836 | u64 v_blkno, p_blkno; |
| 837 | #define CONCURRENT_JOURNAL_FILL 32 |
| 838 | struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL]; |
| 839 | |
| 840 | mlog_entry_void(); |
| 841 | |
| 842 | BUG_ON(inode->i_blocks != |
| 843 | ocfs2_align_bytes_to_sectors(i_size_read(inode))); |
| 844 | |
| 845 | memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL); |
| 846 | |
Andrew Morton | 5515eff | 2006-03-26 01:37:53 -0800 | [diff] [blame] | 847 | mlog(0, "Force reading %llu blocks\n", |
| 848 | (unsigned long long)(inode->i_blocks >> |
| 849 | (inode->i_sb->s_blocksize_bits - 9))); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 850 | |
| 851 | v_blkno = 0; |
| 852 | while (v_blkno < |
| 853 | (inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9))) { |
| 854 | |
| 855 | status = ocfs2_extent_map_get_blocks(inode, v_blkno, |
| 856 | 1, &p_blkno, |
| 857 | &p_blocks); |
| 858 | if (status < 0) { |
| 859 | mlog_errno(status); |
| 860 | goto bail; |
| 861 | } |
| 862 | |
| 863 | if (p_blocks > CONCURRENT_JOURNAL_FILL) |
| 864 | p_blocks = CONCURRENT_JOURNAL_FILL; |
| 865 | |
Mark Fasheh | dd4a2c2 | 2006-04-12 14:24:05 -0700 | [diff] [blame] | 866 | /* We are reading journal data which should not |
| 867 | * be put in the uptodate cache */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 868 | status = ocfs2_read_blocks(OCFS2_SB(inode->i_sb), |
| 869 | p_blkno, p_blocks, bhs, 0, |
Mark Fasheh | dd4a2c2 | 2006-04-12 14:24:05 -0700 | [diff] [blame] | 870 | NULL); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 871 | if (status < 0) { |
| 872 | mlog_errno(status); |
| 873 | goto bail; |
| 874 | } |
| 875 | |
| 876 | for(i = 0; i < p_blocks; i++) { |
| 877 | brelse(bhs[i]); |
| 878 | bhs[i] = NULL; |
| 879 | } |
| 880 | |
| 881 | v_blkno += p_blocks; |
| 882 | } |
| 883 | |
| 884 | bail: |
| 885 | for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++) |
| 886 | if (bhs[i]) |
| 887 | brelse(bhs[i]); |
| 888 | mlog_exit(status); |
| 889 | return status; |
| 890 | } |
| 891 | |
| 892 | struct ocfs2_la_recovery_item { |
| 893 | struct list_head lri_list; |
| 894 | int lri_slot; |
| 895 | struct ocfs2_dinode *lri_la_dinode; |
| 896 | struct ocfs2_dinode *lri_tl_dinode; |
| 897 | }; |
| 898 | |
| 899 | /* Does the second half of the recovery process. By this point, the |
| 900 | * node is marked clean and can actually be considered recovered, |
| 901 | * hence it's no longer in the recovery map, but there's still some |
| 902 | * cleanup we can do which shouldn't happen within the recovery thread |
| 903 | * as locking in that context becomes very difficult if we are to take |
| 904 | * recovering nodes into account. |
| 905 | * |
| 906 | * NOTE: This function can and will sleep on recovery of other nodes |
| 907 | * during cluster locking, just like any other ocfs2 process. |
| 908 | */ |
| 909 | void ocfs2_complete_recovery(void *data) |
| 910 | { |
| 911 | int ret; |
| 912 | struct ocfs2_super *osb = data; |
| 913 | struct ocfs2_journal *journal = osb->journal; |
| 914 | struct ocfs2_dinode *la_dinode, *tl_dinode; |
| 915 | struct ocfs2_la_recovery_item *item; |
| 916 | struct list_head *p, *n; |
| 917 | LIST_HEAD(tmp_la_list); |
| 918 | |
| 919 | mlog_entry_void(); |
| 920 | |
| 921 | mlog(0, "completing recovery from keventd\n"); |
| 922 | |
| 923 | spin_lock(&journal->j_lock); |
| 924 | list_splice_init(&journal->j_la_cleanups, &tmp_la_list); |
| 925 | spin_unlock(&journal->j_lock); |
| 926 | |
| 927 | list_for_each_safe(p, n, &tmp_la_list) { |
| 928 | item = list_entry(p, struct ocfs2_la_recovery_item, lri_list); |
| 929 | list_del_init(&item->lri_list); |
| 930 | |
| 931 | mlog(0, "Complete recovery for slot %d\n", item->lri_slot); |
| 932 | |
| 933 | la_dinode = item->lri_la_dinode; |
| 934 | if (la_dinode) { |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 935 | mlog(0, "Clean up local alloc %llu\n", |
| 936 | (unsigned long long)la_dinode->i_blkno); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 937 | |
| 938 | ret = ocfs2_complete_local_alloc_recovery(osb, |
| 939 | la_dinode); |
| 940 | if (ret < 0) |
| 941 | mlog_errno(ret); |
| 942 | |
| 943 | kfree(la_dinode); |
| 944 | } |
| 945 | |
| 946 | tl_dinode = item->lri_tl_dinode; |
| 947 | if (tl_dinode) { |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 948 | mlog(0, "Clean up truncate log %llu\n", |
| 949 | (unsigned long long)tl_dinode->i_blkno); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 950 | |
| 951 | ret = ocfs2_complete_truncate_log_recovery(osb, |
| 952 | tl_dinode); |
| 953 | if (ret < 0) |
| 954 | mlog_errno(ret); |
| 955 | |
| 956 | kfree(tl_dinode); |
| 957 | } |
| 958 | |
| 959 | ret = ocfs2_recover_orphans(osb, item->lri_slot); |
| 960 | if (ret < 0) |
| 961 | mlog_errno(ret); |
| 962 | |
| 963 | kfree(item); |
| 964 | } |
| 965 | |
| 966 | mlog(0, "Recovery completion\n"); |
| 967 | mlog_exit_void(); |
| 968 | } |
| 969 | |
| 970 | /* NOTE: This function always eats your references to la_dinode and |
| 971 | * tl_dinode, either manually on error, or by passing them to |
| 972 | * ocfs2_complete_recovery */ |
| 973 | static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, |
| 974 | int slot_num, |
| 975 | struct ocfs2_dinode *la_dinode, |
| 976 | struct ocfs2_dinode *tl_dinode) |
| 977 | { |
| 978 | struct ocfs2_la_recovery_item *item; |
| 979 | |
Sunil Mushran | afae00ab | 2006-04-12 14:37:00 -0700 | [diff] [blame] | 980 | item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 981 | if (!item) { |
| 982 | /* Though we wish to avoid it, we are in fact safe in |
| 983 | * skipping local alloc cleanup as fsck.ocfs2 is more |
| 984 | * than capable of reclaiming unused space. */ |
| 985 | if (la_dinode) |
| 986 | kfree(la_dinode); |
| 987 | |
| 988 | if (tl_dinode) |
| 989 | kfree(tl_dinode); |
| 990 | |
| 991 | mlog_errno(-ENOMEM); |
| 992 | return; |
| 993 | } |
| 994 | |
| 995 | INIT_LIST_HEAD(&item->lri_list); |
| 996 | item->lri_la_dinode = la_dinode; |
| 997 | item->lri_slot = slot_num; |
| 998 | item->lri_tl_dinode = tl_dinode; |
| 999 | |
| 1000 | spin_lock(&journal->j_lock); |
| 1001 | list_add_tail(&item->lri_list, &journal->j_la_cleanups); |
| 1002 | queue_work(ocfs2_wq, &journal->j_recovery_work); |
| 1003 | spin_unlock(&journal->j_lock); |
| 1004 | } |
| 1005 | |
| 1006 | /* Called by the mount code to queue recovery the last part of |
| 1007 | * recovery for it's own slot. */ |
| 1008 | void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) |
| 1009 | { |
| 1010 | struct ocfs2_journal *journal = osb->journal; |
| 1011 | |
| 1012 | if (osb->dirty) { |
| 1013 | /* No need to queue up our truncate_log as regular |
| 1014 | * cleanup will catch that. */ |
| 1015 | ocfs2_queue_recovery_completion(journal, |
| 1016 | osb->slot_num, |
| 1017 | osb->local_alloc_copy, |
| 1018 | NULL); |
| 1019 | ocfs2_schedule_truncate_log_flush(osb, 0); |
| 1020 | |
| 1021 | osb->local_alloc_copy = NULL; |
| 1022 | osb->dirty = 0; |
| 1023 | } |
| 1024 | } |
| 1025 | |
| 1026 | static int __ocfs2_recovery_thread(void *arg) |
| 1027 | { |
| 1028 | int status, node_num; |
| 1029 | struct ocfs2_super *osb = arg; |
| 1030 | |
| 1031 | mlog_entry_void(); |
| 1032 | |
| 1033 | status = ocfs2_wait_on_mount(osb); |
| 1034 | if (status < 0) { |
| 1035 | goto bail; |
| 1036 | } |
| 1037 | |
| 1038 | restart: |
| 1039 | status = ocfs2_super_lock(osb, 1); |
| 1040 | if (status < 0) { |
| 1041 | mlog_errno(status); |
| 1042 | goto bail; |
| 1043 | } |
| 1044 | |
| 1045 | while(!ocfs2_node_map_is_empty(osb, &osb->recovery_map)) { |
| 1046 | node_num = ocfs2_node_map_first_set_bit(osb, |
| 1047 | &osb->recovery_map); |
| 1048 | if (node_num == O2NM_INVALID_NODE_NUM) { |
| 1049 | mlog(0, "Out of nodes to recover.\n"); |
| 1050 | break; |
| 1051 | } |
| 1052 | |
| 1053 | status = ocfs2_recover_node(osb, node_num); |
| 1054 | if (status < 0) { |
| 1055 | mlog(ML_ERROR, |
| 1056 | "Error %d recovering node %d on device (%u,%u)!\n", |
| 1057 | status, node_num, |
| 1058 | MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); |
| 1059 | mlog(ML_ERROR, "Volume requires unmount.\n"); |
| 1060 | continue; |
| 1061 | } |
| 1062 | |
| 1063 | ocfs2_recovery_map_clear(osb, node_num); |
| 1064 | } |
| 1065 | ocfs2_super_unlock(osb, 1); |
| 1066 | |
| 1067 | /* We always run recovery on our own orphan dir - the dead |
| 1068 | * node(s) may have voted "no" on an inode delete earlier. A |
| 1069 | * revote is therefore required. */ |
| 1070 | ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, |
| 1071 | NULL); |
| 1072 | |
| 1073 | bail: |
Arjan van de Ven | c74ec2f | 2006-01-13 21:54:23 -0800 | [diff] [blame] | 1074 | mutex_lock(&osb->recovery_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1075 | if (!status && |
| 1076 | !ocfs2_node_map_is_empty(osb, &osb->recovery_map)) { |
Arjan van de Ven | c74ec2f | 2006-01-13 21:54:23 -0800 | [diff] [blame] | 1077 | mutex_unlock(&osb->recovery_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1078 | goto restart; |
| 1079 | } |
| 1080 | |
| 1081 | osb->recovery_thread_task = NULL; |
| 1082 | mb(); /* sync with ocfs2_recovery_thread_running */ |
| 1083 | wake_up(&osb->recovery_event); |
| 1084 | |
Arjan van de Ven | c74ec2f | 2006-01-13 21:54:23 -0800 | [diff] [blame] | 1085 | mutex_unlock(&osb->recovery_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1086 | |
| 1087 | mlog_exit(status); |
| 1088 | /* no one is callint kthread_stop() for us so the kthread() api |
| 1089 | * requires that we call do_exit(). And it isn't exported, but |
| 1090 | * complete_and_exit() seems to be a minimal wrapper around it. */ |
| 1091 | complete_and_exit(NULL, status); |
| 1092 | return status; |
| 1093 | } |
| 1094 | |
| 1095 | void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) |
| 1096 | { |
| 1097 | mlog_entry("(node_num=%d, osb->node_num = %d)\n", |
| 1098 | node_num, osb->node_num); |
| 1099 | |
Arjan van de Ven | c74ec2f | 2006-01-13 21:54:23 -0800 | [diff] [blame] | 1100 | mutex_lock(&osb->recovery_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1101 | if (osb->disable_recovery) |
| 1102 | goto out; |
| 1103 | |
| 1104 | /* People waiting on recovery will wait on |
| 1105 | * the recovery map to empty. */ |
| 1106 | if (!ocfs2_recovery_map_set(osb, node_num)) |
| 1107 | mlog(0, "node %d already be in recovery.\n", node_num); |
| 1108 | |
| 1109 | mlog(0, "starting recovery thread...\n"); |
| 1110 | |
| 1111 | if (osb->recovery_thread_task) |
| 1112 | goto out; |
| 1113 | |
| 1114 | osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, |
Mark Fasheh | 7842704 | 2006-05-04 12:03:26 -0700 | [diff] [blame] | 1115 | "ocfs2rec"); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1116 | if (IS_ERR(osb->recovery_thread_task)) { |
| 1117 | mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); |
| 1118 | osb->recovery_thread_task = NULL; |
| 1119 | } |
| 1120 | |
| 1121 | out: |
Arjan van de Ven | c74ec2f | 2006-01-13 21:54:23 -0800 | [diff] [blame] | 1122 | mutex_unlock(&osb->recovery_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1123 | wake_up(&osb->recovery_event); |
| 1124 | |
| 1125 | mlog_exit_void(); |
| 1126 | } |
| 1127 | |
| 1128 | /* Does the actual journal replay and marks the journal inode as |
| 1129 | * clean. Will only replay if the journal inode is marked dirty. */ |
| 1130 | static int ocfs2_replay_journal(struct ocfs2_super *osb, |
| 1131 | int node_num, |
| 1132 | int slot_num) |
| 1133 | { |
| 1134 | int status; |
| 1135 | int got_lock = 0; |
| 1136 | unsigned int flags; |
| 1137 | struct inode *inode = NULL; |
| 1138 | struct ocfs2_dinode *fe; |
| 1139 | journal_t *journal = NULL; |
| 1140 | struct buffer_head *bh = NULL; |
| 1141 | |
| 1142 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, |
| 1143 | slot_num); |
| 1144 | if (inode == NULL) { |
| 1145 | status = -EACCES; |
| 1146 | mlog_errno(status); |
| 1147 | goto done; |
| 1148 | } |
| 1149 | if (is_bad_inode(inode)) { |
| 1150 | status = -EACCES; |
| 1151 | iput(inode); |
| 1152 | inode = NULL; |
| 1153 | mlog_errno(status); |
| 1154 | goto done; |
| 1155 | } |
| 1156 | SET_INODE_JOURNAL(inode); |
| 1157 | |
| 1158 | status = ocfs2_meta_lock_full(inode, NULL, &bh, 1, |
| 1159 | OCFS2_META_LOCK_RECOVERY); |
| 1160 | if (status < 0) { |
| 1161 | mlog(0, "status returned from ocfs2_meta_lock=%d\n", status); |
| 1162 | if (status != -ERESTARTSYS) |
| 1163 | mlog(ML_ERROR, "Could not lock journal!\n"); |
| 1164 | goto done; |
| 1165 | } |
| 1166 | got_lock = 1; |
| 1167 | |
| 1168 | fe = (struct ocfs2_dinode *) bh->b_data; |
| 1169 | |
| 1170 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); |
| 1171 | |
| 1172 | if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { |
| 1173 | mlog(0, "No recovery required for node %d\n", node_num); |
| 1174 | goto done; |
| 1175 | } |
| 1176 | |
| 1177 | mlog(ML_NOTICE, "Recovering node %d from slot %d on device (%u,%u)\n", |
| 1178 | node_num, slot_num, |
| 1179 | MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); |
| 1180 | |
| 1181 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); |
| 1182 | |
| 1183 | status = ocfs2_force_read_journal(inode); |
| 1184 | if (status < 0) { |
| 1185 | mlog_errno(status); |
| 1186 | goto done; |
| 1187 | } |
| 1188 | |
| 1189 | mlog(0, "calling journal_init_inode\n"); |
| 1190 | journal = journal_init_inode(inode); |
| 1191 | if (journal == NULL) { |
| 1192 | mlog(ML_ERROR, "Linux journal layer error\n"); |
| 1193 | status = -EIO; |
| 1194 | goto done; |
| 1195 | } |
| 1196 | |
| 1197 | status = journal_load(journal); |
| 1198 | if (status < 0) { |
| 1199 | mlog_errno(status); |
| 1200 | if (!igrab(inode)) |
| 1201 | BUG(); |
| 1202 | journal_destroy(journal); |
| 1203 | goto done; |
| 1204 | } |
| 1205 | |
| 1206 | ocfs2_clear_journal_error(osb->sb, journal, slot_num); |
| 1207 | |
| 1208 | /* wipe the journal */ |
| 1209 | mlog(0, "flushing the journal.\n"); |
| 1210 | journal_lock_updates(journal); |
| 1211 | status = journal_flush(journal); |
| 1212 | journal_unlock_updates(journal); |
| 1213 | if (status < 0) |
| 1214 | mlog_errno(status); |
| 1215 | |
| 1216 | /* This will mark the node clean */ |
| 1217 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); |
| 1218 | flags &= ~OCFS2_JOURNAL_DIRTY_FL; |
| 1219 | fe->id1.journal1.ij_flags = cpu_to_le32(flags); |
| 1220 | |
| 1221 | status = ocfs2_write_block(osb, bh, inode); |
| 1222 | if (status < 0) |
| 1223 | mlog_errno(status); |
| 1224 | |
| 1225 | if (!igrab(inode)) |
| 1226 | BUG(); |
| 1227 | |
| 1228 | journal_destroy(journal); |
| 1229 | |
| 1230 | done: |
| 1231 | /* drop the lock on this nodes journal */ |
| 1232 | if (got_lock) |
| 1233 | ocfs2_meta_unlock(inode, 1); |
| 1234 | |
| 1235 | if (inode) |
| 1236 | iput(inode); |
| 1237 | |
| 1238 | if (bh) |
| 1239 | brelse(bh); |
| 1240 | |
| 1241 | mlog_exit(status); |
| 1242 | return status; |
| 1243 | } |
| 1244 | |
| 1245 | /* |
| 1246 | * Do the most important parts of node recovery: |
| 1247 | * - Replay it's journal |
| 1248 | * - Stamp a clean local allocator file |
| 1249 | * - Stamp a clean truncate log |
| 1250 | * - Mark the node clean |
| 1251 | * |
| 1252 | * If this function completes without error, a node in OCFS2 can be |
| 1253 | * said to have been safely recovered. As a result, failure during the |
| 1254 | * second part of a nodes recovery process (local alloc recovery) is |
| 1255 | * far less concerning. |
| 1256 | */ |
| 1257 | static int ocfs2_recover_node(struct ocfs2_super *osb, |
| 1258 | int node_num) |
| 1259 | { |
| 1260 | int status = 0; |
| 1261 | int slot_num; |
| 1262 | struct ocfs2_slot_info *si = osb->slot_info; |
| 1263 | struct ocfs2_dinode *la_copy = NULL; |
| 1264 | struct ocfs2_dinode *tl_copy = NULL; |
| 1265 | |
| 1266 | mlog_entry("(node_num=%d, osb->node_num = %d)\n", |
| 1267 | node_num, osb->node_num); |
| 1268 | |
| 1269 | mlog(0, "checking node %d\n", node_num); |
| 1270 | |
| 1271 | /* Should not ever be called to recover ourselves -- in that |
| 1272 | * case we should've called ocfs2_journal_load instead. */ |
Eric Sesterhenn / snakebyte | ebdec83 | 2006-01-27 10:32:52 +0100 | [diff] [blame] | 1273 | BUG_ON(osb->node_num == node_num); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1274 | |
| 1275 | slot_num = ocfs2_node_num_to_slot(si, node_num); |
| 1276 | if (slot_num == OCFS2_INVALID_SLOT) { |
| 1277 | status = 0; |
| 1278 | mlog(0, "no slot for this node, so no recovery required.\n"); |
| 1279 | goto done; |
| 1280 | } |
| 1281 | |
| 1282 | mlog(0, "node %d was using slot %d\n", node_num, slot_num); |
| 1283 | |
| 1284 | status = ocfs2_replay_journal(osb, node_num, slot_num); |
| 1285 | if (status < 0) { |
| 1286 | mlog_errno(status); |
| 1287 | goto done; |
| 1288 | } |
| 1289 | |
| 1290 | /* Stamp a clean local alloc file AFTER recovering the journal... */ |
| 1291 | status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); |
| 1292 | if (status < 0) { |
| 1293 | mlog_errno(status); |
| 1294 | goto done; |
| 1295 | } |
| 1296 | |
| 1297 | /* An error from begin_truncate_log_recovery is not |
| 1298 | * serious enough to warrant halting the rest of |
| 1299 | * recovery. */ |
| 1300 | status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); |
| 1301 | if (status < 0) |
| 1302 | mlog_errno(status); |
| 1303 | |
| 1304 | /* Likewise, this would be a strange but ultimately not so |
| 1305 | * harmful place to get an error... */ |
| 1306 | ocfs2_clear_slot(si, slot_num); |
| 1307 | status = ocfs2_update_disk_slots(osb, si); |
| 1308 | if (status < 0) |
| 1309 | mlog_errno(status); |
| 1310 | |
| 1311 | /* This will kfree the memory pointed to by la_copy and tl_copy */ |
| 1312 | ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, |
| 1313 | tl_copy); |
| 1314 | |
| 1315 | status = 0; |
| 1316 | done: |
| 1317 | |
| 1318 | mlog_exit(status); |
| 1319 | return status; |
| 1320 | } |
| 1321 | |
| 1322 | /* Test node liveness by trylocking his journal. If we get the lock, |
| 1323 | * we drop it here. Return 0 if we got the lock, -EAGAIN if node is |
| 1324 | * still alive (we couldn't get the lock) and < 0 on error. */ |
| 1325 | static int ocfs2_trylock_journal(struct ocfs2_super *osb, |
| 1326 | int slot_num) |
| 1327 | { |
| 1328 | int status, flags; |
| 1329 | struct inode *inode = NULL; |
| 1330 | |
| 1331 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, |
| 1332 | slot_num); |
| 1333 | if (inode == NULL) { |
| 1334 | mlog(ML_ERROR, "access error\n"); |
| 1335 | status = -EACCES; |
| 1336 | goto bail; |
| 1337 | } |
| 1338 | if (is_bad_inode(inode)) { |
| 1339 | mlog(ML_ERROR, "access error (bad inode)\n"); |
| 1340 | iput(inode); |
| 1341 | inode = NULL; |
| 1342 | status = -EACCES; |
| 1343 | goto bail; |
| 1344 | } |
| 1345 | SET_INODE_JOURNAL(inode); |
| 1346 | |
| 1347 | flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; |
| 1348 | status = ocfs2_meta_lock_full(inode, NULL, NULL, 1, flags); |
| 1349 | if (status < 0) { |
| 1350 | if (status != -EAGAIN) |
| 1351 | mlog_errno(status); |
| 1352 | goto bail; |
| 1353 | } |
| 1354 | |
| 1355 | ocfs2_meta_unlock(inode, 1); |
| 1356 | bail: |
| 1357 | if (inode) |
| 1358 | iput(inode); |
| 1359 | |
| 1360 | return status; |
| 1361 | } |
| 1362 | |
| 1363 | /* Call this underneath ocfs2_super_lock. It also assumes that the |
| 1364 | * slot info struct has been updated from disk. */ |
| 1365 | int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) |
| 1366 | { |
| 1367 | int status, i, node_num; |
| 1368 | struct ocfs2_slot_info *si = osb->slot_info; |
| 1369 | |
| 1370 | /* This is called with the super block cluster lock, so we |
| 1371 | * know that the slot map can't change underneath us. */ |
| 1372 | |
| 1373 | spin_lock(&si->si_lock); |
| 1374 | for(i = 0; i < si->si_num_slots; i++) { |
| 1375 | if (i == osb->slot_num) |
| 1376 | continue; |
| 1377 | if (ocfs2_is_empty_slot(si, i)) |
| 1378 | continue; |
| 1379 | |
| 1380 | node_num = si->si_global_node_nums[i]; |
| 1381 | if (ocfs2_node_map_test_bit(osb, &osb->recovery_map, node_num)) |
| 1382 | continue; |
| 1383 | spin_unlock(&si->si_lock); |
| 1384 | |
| 1385 | /* Ok, we have a slot occupied by another node which |
| 1386 | * is not in the recovery map. We trylock his journal |
| 1387 | * file here to test if he's alive. */ |
| 1388 | status = ocfs2_trylock_journal(osb, i); |
| 1389 | if (!status) { |
| 1390 | /* Since we're called from mount, we know that |
| 1391 | * the recovery thread can't race us on |
| 1392 | * setting / checking the recovery bits. */ |
| 1393 | ocfs2_recovery_thread(osb, node_num); |
| 1394 | } else if ((status < 0) && (status != -EAGAIN)) { |
| 1395 | mlog_errno(status); |
| 1396 | goto bail; |
| 1397 | } |
| 1398 | |
| 1399 | spin_lock(&si->si_lock); |
| 1400 | } |
| 1401 | spin_unlock(&si->si_lock); |
| 1402 | |
| 1403 | status = 0; |
| 1404 | bail: |
| 1405 | mlog_exit(status); |
| 1406 | return status; |
| 1407 | } |
| 1408 | |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 1409 | static int ocfs2_queue_orphans(struct ocfs2_super *osb, |
| 1410 | int slot, |
| 1411 | struct inode **head) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1412 | { |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 1413 | int status; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1414 | struct inode *orphan_dir_inode = NULL; |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 1415 | struct inode *iter; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1416 | unsigned long offset, blk, local; |
| 1417 | struct buffer_head *bh = NULL; |
| 1418 | struct ocfs2_dir_entry *de; |
| 1419 | struct super_block *sb = osb->sb; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1420 | |
| 1421 | orphan_dir_inode = ocfs2_get_system_file_inode(osb, |
| 1422 | ORPHAN_DIR_SYSTEM_INODE, |
| 1423 | slot); |
| 1424 | if (!orphan_dir_inode) { |
| 1425 | status = -ENOENT; |
| 1426 | mlog_errno(status); |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 1427 | return status; |
| 1428 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1429 | |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 1430 | mutex_lock(&orphan_dir_inode->i_mutex); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1431 | status = ocfs2_meta_lock(orphan_dir_inode, NULL, NULL, 0); |
| 1432 | if (status < 0) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1433 | mlog_errno(status); |
| 1434 | goto out; |
| 1435 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1436 | |
| 1437 | offset = 0; |
| 1438 | iter = NULL; |
| 1439 | while(offset < i_size_read(orphan_dir_inode)) { |
| 1440 | blk = offset >> sb->s_blocksize_bits; |
| 1441 | |
| 1442 | bh = ocfs2_bread(orphan_dir_inode, blk, &status, 0); |
| 1443 | if (!bh) |
| 1444 | status = -EINVAL; |
| 1445 | if (status < 0) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1446 | if (bh) |
| 1447 | brelse(bh); |
| 1448 | mlog_errno(status); |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 1449 | goto out_unlock; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1450 | } |
| 1451 | |
| 1452 | local = 0; |
| 1453 | while(offset < i_size_read(orphan_dir_inode) |
| 1454 | && local < sb->s_blocksize) { |
| 1455 | de = (struct ocfs2_dir_entry *) (bh->b_data + local); |
| 1456 | |
| 1457 | if (!ocfs2_check_dir_entry(orphan_dir_inode, |
| 1458 | de, bh, local)) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1459 | status = -EINVAL; |
| 1460 | mlog_errno(status); |
| 1461 | brelse(bh); |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 1462 | goto out_unlock; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1463 | } |
| 1464 | |
| 1465 | local += le16_to_cpu(de->rec_len); |
| 1466 | offset += le16_to_cpu(de->rec_len); |
| 1467 | |
| 1468 | /* I guess we silently fail on no inode? */ |
| 1469 | if (!le64_to_cpu(de->inode)) |
| 1470 | continue; |
| 1471 | if (de->file_type > OCFS2_FT_MAX) { |
| 1472 | mlog(ML_ERROR, |
| 1473 | "block %llu contains invalid de: " |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 1474 | "inode = %llu, rec_len = %u, " |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1475 | "name_len = %u, file_type = %u, " |
| 1476 | "name='%.*s'\n", |
| 1477 | (unsigned long long)bh->b_blocknr, |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 1478 | (unsigned long long)le64_to_cpu(de->inode), |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1479 | le16_to_cpu(de->rec_len), |
| 1480 | de->name_len, |
| 1481 | de->file_type, |
| 1482 | de->name_len, |
| 1483 | de->name); |
| 1484 | continue; |
| 1485 | } |
| 1486 | if (de->name_len == 1 && !strncmp(".", de->name, 1)) |
| 1487 | continue; |
| 1488 | if (de->name_len == 2 && !strncmp("..", de->name, 2)) |
| 1489 | continue; |
| 1490 | |
Mark Fasheh | 24c19ef | 2006-09-22 17:28:19 -0700 | [diff] [blame] | 1491 | iter = ocfs2_iget(osb, le64_to_cpu(de->inode), |
| 1492 | OCFS2_FI_FLAG_NOLOCK); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1493 | if (IS_ERR(iter)) |
| 1494 | continue; |
| 1495 | |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 1496 | mlog(0, "queue orphan %llu\n", |
| 1497 | (unsigned long long)OCFS2_I(iter)->ip_blkno); |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 1498 | /* No locking is required for the next_orphan |
| 1499 | * queue as there is only ever a single |
| 1500 | * process doing orphan recovery. */ |
| 1501 | OCFS2_I(iter)->ip_next_orphan = *head; |
| 1502 | *head = iter; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1503 | } |
| 1504 | brelse(bh); |
| 1505 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1506 | |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 1507 | out_unlock: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1508 | ocfs2_meta_unlock(orphan_dir_inode, 0); |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 1509 | out: |
| 1510 | mutex_unlock(&orphan_dir_inode->i_mutex); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1511 | iput(orphan_dir_inode); |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 1512 | return status; |
| 1513 | } |
| 1514 | |
| 1515 | static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, |
| 1516 | int slot) |
| 1517 | { |
| 1518 | int ret; |
| 1519 | |
| 1520 | spin_lock(&osb->osb_lock); |
| 1521 | ret = !osb->osb_orphan_wipes[slot]; |
| 1522 | spin_unlock(&osb->osb_lock); |
| 1523 | return ret; |
| 1524 | } |
| 1525 | |
| 1526 | static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, |
| 1527 | int slot) |
| 1528 | { |
| 1529 | spin_lock(&osb->osb_lock); |
| 1530 | /* Mark ourselves such that new processes in delete_inode() |
| 1531 | * know to quit early. */ |
| 1532 | ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); |
| 1533 | while (osb->osb_orphan_wipes[slot]) { |
| 1534 | /* If any processes are already in the middle of an |
| 1535 | * orphan wipe on this dir, then we need to wait for |
| 1536 | * them. */ |
| 1537 | spin_unlock(&osb->osb_lock); |
| 1538 | wait_event_interruptible(osb->osb_wipe_event, |
| 1539 | ocfs2_orphan_recovery_can_continue(osb, slot)); |
| 1540 | spin_lock(&osb->osb_lock); |
| 1541 | } |
| 1542 | spin_unlock(&osb->osb_lock); |
| 1543 | } |
| 1544 | |
| 1545 | static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, |
| 1546 | int slot) |
| 1547 | { |
| 1548 | ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); |
| 1549 | } |
| 1550 | |
| 1551 | /* |
| 1552 | * Orphan recovery. Each mounted node has it's own orphan dir which we |
| 1553 | * must run during recovery. Our strategy here is to build a list of |
| 1554 | * the inodes in the orphan dir and iget/iput them. The VFS does |
| 1555 | * (most) of the rest of the work. |
| 1556 | * |
| 1557 | * Orphan recovery can happen at any time, not just mount so we have a |
| 1558 | * couple of extra considerations. |
| 1559 | * |
| 1560 | * - We grab as many inodes as we can under the orphan dir lock - |
| 1561 | * doing iget() outside the orphan dir risks getting a reference on |
| 1562 | * an invalid inode. |
| 1563 | * - We must be sure not to deadlock with other processes on the |
| 1564 | * system wanting to run delete_inode(). This can happen when they go |
| 1565 | * to lock the orphan dir and the orphan recovery process attempts to |
| 1566 | * iget() inside the orphan dir lock. This can be avoided by |
| 1567 | * advertising our state to ocfs2_delete_inode(). |
| 1568 | */ |
| 1569 | static int ocfs2_recover_orphans(struct ocfs2_super *osb, |
| 1570 | int slot) |
| 1571 | { |
| 1572 | int ret = 0; |
| 1573 | struct inode *inode = NULL; |
| 1574 | struct inode *iter; |
| 1575 | struct ocfs2_inode_info *oi; |
| 1576 | |
| 1577 | mlog(0, "Recover inodes from orphan dir in slot %d\n", slot); |
| 1578 | |
| 1579 | ocfs2_mark_recovering_orphan_dir(osb, slot); |
| 1580 | ret = ocfs2_queue_orphans(osb, slot, &inode); |
| 1581 | ocfs2_clear_recovering_orphan_dir(osb, slot); |
| 1582 | |
| 1583 | /* Error here should be noted, but we want to continue with as |
| 1584 | * many queued inodes as we've got. */ |
| 1585 | if (ret) |
| 1586 | mlog_errno(ret); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1587 | |
| 1588 | while (inode) { |
| 1589 | oi = OCFS2_I(inode); |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 1590 | mlog(0, "iput orphan %llu\n", (unsigned long long)oi->ip_blkno); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1591 | |
| 1592 | iter = oi->ip_next_orphan; |
| 1593 | |
| 1594 | spin_lock(&oi->ip_lock); |
| 1595 | /* Delete voting may have set these on the assumption |
| 1596 | * that the other node would wipe them successfully. |
| 1597 | * If they are still in the node's orphan dir, we need |
| 1598 | * to reset that state. */ |
| 1599 | oi->ip_flags &= ~(OCFS2_INODE_DELETED|OCFS2_INODE_SKIP_DELETE); |
| 1600 | |
| 1601 | /* Set the proper information to get us going into |
| 1602 | * ocfs2_delete_inode. */ |
| 1603 | oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; |
| 1604 | oi->ip_orphaned_slot = slot; |
| 1605 | spin_unlock(&oi->ip_lock); |
| 1606 | |
| 1607 | iput(inode); |
| 1608 | |
| 1609 | inode = iter; |
| 1610 | } |
| 1611 | |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 1612 | return ret; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1613 | } |
| 1614 | |
| 1615 | static int ocfs2_wait_on_mount(struct ocfs2_super *osb) |
| 1616 | { |
| 1617 | /* This check is good because ocfs2 will wait on our recovery |
| 1618 | * thread before changing it to something other than MOUNTED |
| 1619 | * or DISABLED. */ |
| 1620 | wait_event(osb->osb_mount_event, |
| 1621 | atomic_read(&osb->vol_state) == VOLUME_MOUNTED || |
| 1622 | atomic_read(&osb->vol_state) == VOLUME_DISABLED); |
| 1623 | |
| 1624 | /* If there's an error on mount, then we may never get to the |
| 1625 | * MOUNTED flag, but this is set right before |
| 1626 | * dismount_volume() so we can trust it. */ |
| 1627 | if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { |
| 1628 | mlog(0, "mount error, exiting!\n"); |
| 1629 | return -EBUSY; |
| 1630 | } |
| 1631 | |
| 1632 | return 0; |
| 1633 | } |
| 1634 | |
| 1635 | static int ocfs2_commit_thread(void *arg) |
| 1636 | { |
| 1637 | int status; |
| 1638 | struct ocfs2_super *osb = arg; |
| 1639 | struct ocfs2_journal *journal = osb->journal; |
| 1640 | |
| 1641 | /* we can trust j_num_trans here because _should_stop() is only set in |
| 1642 | * shutdown and nobody other than ourselves should be able to start |
| 1643 | * transactions. committing on shutdown might take a few iterations |
| 1644 | * as final transactions put deleted inodes on the list */ |
| 1645 | while (!(kthread_should_stop() && |
| 1646 | atomic_read(&journal->j_num_trans) == 0)) { |
| 1647 | |
Mark Fasheh | 745ae8ba | 2006-02-09 13:23:39 -0800 | [diff] [blame] | 1648 | wait_event_interruptible(osb->checkpoint_event, |
| 1649 | atomic_read(&journal->j_num_trans) |
| 1650 | || kthread_should_stop()); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1651 | |
| 1652 | status = ocfs2_commit_cache(osb); |
| 1653 | if (status < 0) |
| 1654 | mlog_errno(status); |
| 1655 | |
| 1656 | if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ |
| 1657 | mlog(ML_KTHREAD, |
| 1658 | "commit_thread: %u transactions pending on " |
| 1659 | "shutdown\n", |
| 1660 | atomic_read(&journal->j_num_trans)); |
| 1661 | } |
| 1662 | } |
| 1663 | |
| 1664 | return 0; |
| 1665 | } |
| 1666 | |
| 1667 | /* Look for a dirty journal without taking any cluster locks. Used for |
| 1668 | * hard readonly access to determine whether the file system journals |
| 1669 | * require recovery. */ |
| 1670 | int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) |
| 1671 | { |
| 1672 | int ret = 0; |
| 1673 | unsigned int slot; |
| 1674 | struct buffer_head *di_bh; |
| 1675 | struct ocfs2_dinode *di; |
| 1676 | struct inode *journal = NULL; |
| 1677 | |
| 1678 | for(slot = 0; slot < osb->max_slots; slot++) { |
| 1679 | journal = ocfs2_get_system_file_inode(osb, |
| 1680 | JOURNAL_SYSTEM_INODE, |
| 1681 | slot); |
| 1682 | if (!journal || is_bad_inode(journal)) { |
| 1683 | ret = -EACCES; |
| 1684 | mlog_errno(ret); |
| 1685 | goto out; |
| 1686 | } |
| 1687 | |
| 1688 | di_bh = NULL; |
| 1689 | ret = ocfs2_read_block(osb, OCFS2_I(journal)->ip_blkno, &di_bh, |
| 1690 | 0, journal); |
| 1691 | if (ret < 0) { |
| 1692 | mlog_errno(ret); |
| 1693 | goto out; |
| 1694 | } |
| 1695 | |
| 1696 | di = (struct ocfs2_dinode *) di_bh->b_data; |
| 1697 | |
| 1698 | if (le32_to_cpu(di->id1.journal1.ij_flags) & |
| 1699 | OCFS2_JOURNAL_DIRTY_FL) |
| 1700 | ret = -EROFS; |
| 1701 | |
| 1702 | brelse(di_bh); |
| 1703 | if (ret) |
| 1704 | break; |
| 1705 | } |
| 1706 | |
| 1707 | out: |
| 1708 | if (journal) |
| 1709 | iput(journal); |
| 1710 | |
| 1711 | return ret; |
| 1712 | } |