1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 4 * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved. 5 */ 6 7 #include <linux/sched.h> 8 #include <linux/slab.h> 9 #include <linux/spinlock.h> 10 #include <linux/completion.h> 11 #include <linux/buffer_head.h> 12 #include <linux/gfs2_ondisk.h> 13 #include <linux/crc32.h> 14 #include <linux/crc32c.h> 15 #include <linux/delay.h> 16 #include <linux/kthread.h> 17 #include <linux/freezer.h> 18 #include <linux/bio.h> 19 #include <linux/blkdev.h> 20 #include <linux/writeback.h> 21 #include <linux/list_sort.h> 22 23 #include "gfs2.h" 24 #include "incore.h" 25 #include "bmap.h" 26 #include "glock.h" 27 #include "log.h" 28 #include "lops.h" 29 #include "meta_io.h" 30 #include "util.h" 31 #include "dir.h" 32 #include "trace_gfs2.h" 33 34 static void gfs2_log_shutdown(struct gfs2_sbd *sdp); 35 36 /** 37 * gfs2_struct2blk - compute stuff 38 * @sdp: the filesystem 39 * @nstruct: the number of structures 40 * @ssize: the size of the structures 41 * 42 * Compute the number of log descriptor blocks needed to hold a certain number 43 * of structures of a certain size. 44 * 45 * Returns: the number of blocks needed (minimum is always 1) 46 */ 47 48 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct, 49 unsigned int ssize) 50 { 51 unsigned int blks; 52 unsigned int first, second; 53 54 blks = 1; 55 first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize; 56 57 if (nstruct > first) { 58 second = (sdp->sd_sb.sb_bsize - 59 sizeof(struct gfs2_meta_header)) / ssize; 60 blks += DIV_ROUND_UP(nstruct - first, second); 61 } 62 63 return blks; 64 } 65 66 /** 67 * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters 68 * @mapping: The associated mapping (maybe NULL) 69 * @bd: The gfs2_bufdata to remove 70 * 71 * The ail lock _must_ be held when calling this function 72 * 73 */ 74 75 static void gfs2_remove_from_ail(struct gfs2_bufdata *bd) 76 { 77 bd->bd_tr = NULL; 78 list_del_init(&bd->bd_ail_st_list); 79 list_del_init(&bd->bd_ail_gl_list); 80 atomic_dec(&bd->bd_gl->gl_ail_count); 81 brelse(bd->bd_bh); 82 } 83 84 /** 85 * gfs2_ail1_start_one - Start I/O on a part of the AIL 86 * @sdp: the filesystem 87 * @wbc: The writeback control structure 88 * @ai: The ail structure 89 * 90 */ 91 92 static int gfs2_ail1_start_one(struct gfs2_sbd *sdp, 93 struct writeback_control *wbc, 94 struct gfs2_trans *tr, 95 bool *withdraw) 96 __releases(&sdp->sd_ail_lock) 97 __acquires(&sdp->sd_ail_lock) 98 { 99 struct gfs2_glock *gl = NULL; 100 struct address_space *mapping; 101 struct gfs2_bufdata *bd, *s; 102 struct buffer_head *bh; 103 104 list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) { 105 bh = bd->bd_bh; 106 107 gfs2_assert(sdp, bd->bd_tr == tr); 108 109 if (!buffer_busy(bh)) { 110 if (!buffer_uptodate(bh) && 111 !test_and_set_bit(SDF_AIL1_IO_ERROR, 112 &sdp->sd_flags)) { 113 gfs2_io_error_bh(sdp, bh); 114 *withdraw = true; 115 } 116 list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list); 117 continue; 118 } 119 120 if (!buffer_dirty(bh)) 121 continue; 122 if (gl == bd->bd_gl) 123 continue; 124 gl = bd->bd_gl; 125 list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list); 126 mapping = bh->b_page->mapping; 127 if (!mapping) 128 continue; 129 spin_unlock(&sdp->sd_ail_lock); 130 generic_writepages(mapping, wbc); 131 spin_lock(&sdp->sd_ail_lock); 132 if (wbc->nr_to_write <= 0) 133 break; 134 return 1; 135 } 136 137 return 0; 138 } 139 140 141 /** 142 * gfs2_ail1_flush - start writeback of some ail1 entries 143 * @sdp: The super block 144 * @wbc: The writeback control structure 145 * 146 * Writes back some ail1 entries, according to the limits in the 147 * writeback control structure 148 */ 149 150 void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc) 151 { 152 struct list_head *head = &sdp->sd_ail1_list; 153 struct gfs2_trans *tr; 154 struct blk_plug plug; 155 bool withdraw = false; 156 157 trace_gfs2_ail_flush(sdp, wbc, 1); 158 blk_start_plug(&plug); 159 spin_lock(&sdp->sd_ail_lock); 160 restart: 161 list_for_each_entry_reverse(tr, head, tr_list) { 162 if (wbc->nr_to_write <= 0) 163 break; 164 if (gfs2_ail1_start_one(sdp, wbc, tr, &withdraw) && 165 !gfs2_withdrawn(sdp)) 166 goto restart; 167 } 168 spin_unlock(&sdp->sd_ail_lock); 169 blk_finish_plug(&plug); 170 if (withdraw) 171 gfs2_lm_withdraw(sdp, NULL); 172 trace_gfs2_ail_flush(sdp, wbc, 0); 173 } 174 175 /** 176 * gfs2_ail1_start - start writeback of all ail1 entries 177 * @sdp: The superblock 178 */ 179 180 static void gfs2_ail1_start(struct gfs2_sbd *sdp) 181 { 182 struct writeback_control wbc = { 183 .sync_mode = WB_SYNC_NONE, 184 .nr_to_write = LONG_MAX, 185 .range_start = 0, 186 .range_end = LLONG_MAX, 187 }; 188 189 return gfs2_ail1_flush(sdp, &wbc); 190 } 191 192 /** 193 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced 194 * @sdp: the filesystem 195 * @ai: the AIL entry 196 * 197 */ 198 199 static void gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr, 200 bool *withdraw) 201 { 202 struct gfs2_bufdata *bd, *s; 203 struct buffer_head *bh; 204 205 list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, 206 bd_ail_st_list) { 207 bh = bd->bd_bh; 208 gfs2_assert(sdp, bd->bd_tr == tr); 209 if (buffer_busy(bh)) 210 continue; 211 if (!buffer_uptodate(bh) && 212 !test_and_set_bit(SDF_AIL1_IO_ERROR, &sdp->sd_flags)) { 213 gfs2_io_error_bh(sdp, bh); 214 *withdraw = true; 215 } 216 list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list); 217 } 218 } 219 220 /** 221 * gfs2_ail1_empty - Try to empty the ail1 lists 222 * @sdp: The superblock 223 * 224 * Tries to empty the ail1 lists, starting with the oldest first 225 */ 226 227 static int gfs2_ail1_empty(struct gfs2_sbd *sdp) 228 { 229 struct gfs2_trans *tr, *s; 230 int oldest_tr = 1; 231 int ret; 232 bool withdraw = false; 233 234 spin_lock(&sdp->sd_ail_lock); 235 list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) { 236 gfs2_ail1_empty_one(sdp, tr, &withdraw); 237 if (list_empty(&tr->tr_ail1_list) && oldest_tr) 238 list_move(&tr->tr_list, &sdp->sd_ail2_list); 239 else 240 oldest_tr = 0; 241 } 242 ret = list_empty(&sdp->sd_ail1_list); 243 spin_unlock(&sdp->sd_ail_lock); 244 245 if (withdraw) 246 gfs2_lm_withdraw(sdp, "fatal: I/O error(s)\n"); 247 248 return ret; 249 } 250 251 static void gfs2_ail1_wait(struct gfs2_sbd *sdp) 252 { 253 struct gfs2_trans *tr; 254 struct gfs2_bufdata *bd; 255 struct buffer_head *bh; 256 257 spin_lock(&sdp->sd_ail_lock); 258 list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) { 259 list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) { 260 bh = bd->bd_bh; 261 if (!buffer_locked(bh)) 262 continue; 263 get_bh(bh); 264 spin_unlock(&sdp->sd_ail_lock); 265 wait_on_buffer(bh); 266 brelse(bh); 267 return; 268 } 269 } 270 spin_unlock(&sdp->sd_ail_lock); 271 } 272 273 /** 274 * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced 275 * @sdp: the filesystem 276 * @ai: the AIL entry 277 * 278 */ 279 280 static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 281 { 282 struct list_head *head = &tr->tr_ail2_list; 283 struct gfs2_bufdata *bd; 284 285 while (!list_empty(head)) { 286 bd = list_entry(head->prev, struct gfs2_bufdata, 287 bd_ail_st_list); 288 gfs2_assert(sdp, bd->bd_tr == tr); 289 gfs2_remove_from_ail(bd); 290 } 291 } 292 293 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail) 294 { 295 struct gfs2_trans *tr, *safe; 296 unsigned int old_tail = sdp->sd_log_tail; 297 int wrap = (new_tail < old_tail); 298 int a, b, rm; 299 300 spin_lock(&sdp->sd_ail_lock); 301 302 list_for_each_entry_safe(tr, safe, &sdp->sd_ail2_list, tr_list) { 303 a = (old_tail <= tr->tr_first); 304 b = (tr->tr_first < new_tail); 305 rm = (wrap) ? (a || b) : (a && b); 306 if (!rm) 307 continue; 308 309 gfs2_ail2_empty_one(sdp, tr); 310 list_del(&tr->tr_list); 311 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list)); 312 gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list)); 313 kfree(tr); 314 } 315 316 spin_unlock(&sdp->sd_ail_lock); 317 } 318 319 /** 320 * gfs2_log_release - Release a given number of log blocks 321 * @sdp: The GFS2 superblock 322 * @blks: The number of blocks 323 * 324 */ 325 326 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks) 327 { 328 329 atomic_add(blks, &sdp->sd_log_blks_free); 330 trace_gfs2_log_blocks(sdp, blks); 331 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= 332 sdp->sd_jdesc->jd_blocks); 333 up_read(&sdp->sd_log_flush_lock); 334 } 335 336 /** 337 * gfs2_log_reserve - Make a log reservation 338 * @sdp: The GFS2 superblock 339 * @blks: The number of blocks to reserve 340 * 341 * Note that we never give out the last few blocks of the journal. Thats 342 * due to the fact that there is a small number of header blocks 343 * associated with each log flush. The exact number can't be known until 344 * flush time, so we ensure that we have just enough free blocks at all 345 * times to avoid running out during a log flush. 346 * 347 * We no longer flush the log here, instead we wake up logd to do that 348 * for us. To avoid the thundering herd and to ensure that we deal fairly 349 * with queued waiters, we use an exclusive wait. This means that when we 350 * get woken with enough journal space to get our reservation, we need to 351 * wake the next waiter on the list. 352 * 353 * Returns: errno 354 */ 355 356 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks) 357 { 358 int ret = 0; 359 unsigned reserved_blks = 7 * (4096 / sdp->sd_vfs->s_blocksize); 360 unsigned wanted = blks + reserved_blks; 361 DEFINE_WAIT(wait); 362 int did_wait = 0; 363 unsigned int free_blocks; 364 365 if (gfs2_assert_warn(sdp, blks) || 366 gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks)) 367 return -EINVAL; 368 atomic_add(blks, &sdp->sd_log_blks_needed); 369 retry: 370 free_blocks = atomic_read(&sdp->sd_log_blks_free); 371 if (unlikely(free_blocks <= wanted)) { 372 do { 373 prepare_to_wait_exclusive(&sdp->sd_log_waitq, &wait, 374 TASK_UNINTERRUPTIBLE); 375 wake_up(&sdp->sd_logd_waitq); 376 did_wait = 1; 377 if (atomic_read(&sdp->sd_log_blks_free) <= wanted) 378 io_schedule(); 379 free_blocks = atomic_read(&sdp->sd_log_blks_free); 380 } while(free_blocks <= wanted); 381 finish_wait(&sdp->sd_log_waitq, &wait); 382 } 383 atomic_inc(&sdp->sd_reserving_log); 384 if (atomic_cmpxchg(&sdp->sd_log_blks_free, free_blocks, 385 free_blocks - blks) != free_blocks) { 386 if (atomic_dec_and_test(&sdp->sd_reserving_log)) 387 wake_up(&sdp->sd_reserving_log_wait); 388 goto retry; 389 } 390 atomic_sub(blks, &sdp->sd_log_blks_needed); 391 trace_gfs2_log_blocks(sdp, -blks); 392 393 /* 394 * If we waited, then so might others, wake them up _after_ we get 395 * our share of the log. 396 */ 397 if (unlikely(did_wait)) 398 wake_up(&sdp->sd_log_waitq); 399 400 down_read(&sdp->sd_log_flush_lock); 401 if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) { 402 gfs2_log_release(sdp, blks); 403 ret = -EROFS; 404 } 405 if (atomic_dec_and_test(&sdp->sd_reserving_log)) 406 wake_up(&sdp->sd_reserving_log_wait); 407 return ret; 408 } 409 410 /** 411 * log_distance - Compute distance between two journal blocks 412 * @sdp: The GFS2 superblock 413 * @newer: The most recent journal block of the pair 414 * @older: The older journal block of the pair 415 * 416 * Compute the distance (in the journal direction) between two 417 * blocks in the journal 418 * 419 * Returns: the distance in blocks 420 */ 421 422 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer, 423 unsigned int older) 424 { 425 int dist; 426 427 dist = newer - older; 428 if (dist < 0) 429 dist += sdp->sd_jdesc->jd_blocks; 430 431 return dist; 432 } 433 434 /** 435 * calc_reserved - Calculate the number of blocks to reserve when 436 * refunding a transaction's unused buffers. 437 * @sdp: The GFS2 superblock 438 * 439 * This is complex. We need to reserve room for all our currently used 440 * metadata buffers (e.g. normal file I/O rewriting file time stamps) and 441 * all our journaled data buffers for journaled files (e.g. files in the 442 * meta_fs like rindex, or files for which chattr +j was done.) 443 * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush 444 * will count it as free space (sd_log_blks_free) and corruption will follow. 445 * 446 * We can have metadata bufs and jdata bufs in the same journal. So each 447 * type gets its own log header, for which we need to reserve a block. 448 * In fact, each type has the potential for needing more than one header 449 * in cases where we have more buffers than will fit on a journal page. 450 * Metadata journal entries take up half the space of journaled buffer entries. 451 * Thus, metadata entries have buf_limit (502) and journaled buffers have 452 * databuf_limit (251) before they cause a wrap around. 453 * 454 * Also, we need to reserve blocks for revoke journal entries and one for an 455 * overall header for the lot. 456 * 457 * Returns: the number of blocks reserved 458 */ 459 static unsigned int calc_reserved(struct gfs2_sbd *sdp) 460 { 461 unsigned int reserved = 0; 462 unsigned int mbuf; 463 unsigned int dbuf; 464 struct gfs2_trans *tr = sdp->sd_log_tr; 465 466 if (tr) { 467 mbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm; 468 dbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm; 469 reserved = mbuf + dbuf; 470 /* Account for header blocks */ 471 reserved += DIV_ROUND_UP(mbuf, buf_limit(sdp)); 472 reserved += DIV_ROUND_UP(dbuf, databuf_limit(sdp)); 473 } 474 475 if (sdp->sd_log_commited_revoke > 0) 476 reserved += gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke, 477 sizeof(u64)); 478 /* One for the overall header */ 479 if (reserved) 480 reserved++; 481 return reserved; 482 } 483 484 static unsigned int current_tail(struct gfs2_sbd *sdp) 485 { 486 struct gfs2_trans *tr; 487 unsigned int tail; 488 489 spin_lock(&sdp->sd_ail_lock); 490 491 if (list_empty(&sdp->sd_ail1_list)) { 492 tail = sdp->sd_log_head; 493 } else { 494 tr = list_entry(sdp->sd_ail1_list.prev, struct gfs2_trans, 495 tr_list); 496 tail = tr->tr_first; 497 } 498 499 spin_unlock(&sdp->sd_ail_lock); 500 501 return tail; 502 } 503 504 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail) 505 { 506 unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail); 507 508 ail2_empty(sdp, new_tail); 509 510 atomic_add(dist, &sdp->sd_log_blks_free); 511 trace_gfs2_log_blocks(sdp, dist); 512 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= 513 sdp->sd_jdesc->jd_blocks); 514 515 sdp->sd_log_tail = new_tail; 516 } 517 518 519 static void log_flush_wait(struct gfs2_sbd *sdp) 520 { 521 DEFINE_WAIT(wait); 522 523 if (atomic_read(&sdp->sd_log_in_flight)) { 524 do { 525 prepare_to_wait(&sdp->sd_log_flush_wait, &wait, 526 TASK_UNINTERRUPTIBLE); 527 if (atomic_read(&sdp->sd_log_in_flight)) 528 io_schedule(); 529 } while(atomic_read(&sdp->sd_log_in_flight)); 530 finish_wait(&sdp->sd_log_flush_wait, &wait); 531 } 532 } 533 534 static int ip_cmp(void *priv, struct list_head *a, struct list_head *b) 535 { 536 struct gfs2_inode *ipa, *ipb; 537 538 ipa = list_entry(a, struct gfs2_inode, i_ordered); 539 ipb = list_entry(b, struct gfs2_inode, i_ordered); 540 541 if (ipa->i_no_addr < ipb->i_no_addr) 542 return -1; 543 if (ipa->i_no_addr > ipb->i_no_addr) 544 return 1; 545 return 0; 546 } 547 548 static void gfs2_ordered_write(struct gfs2_sbd *sdp) 549 { 550 struct gfs2_inode *ip; 551 LIST_HEAD(written); 552 553 spin_lock(&sdp->sd_ordered_lock); 554 list_sort(NULL, &sdp->sd_log_ordered, &ip_cmp); 555 while (!list_empty(&sdp->sd_log_ordered)) { 556 ip = list_entry(sdp->sd_log_ordered.next, struct gfs2_inode, i_ordered); 557 if (ip->i_inode.i_mapping->nrpages == 0) { 558 test_and_clear_bit(GIF_ORDERED, &ip->i_flags); 559 list_del(&ip->i_ordered); 560 continue; 561 } 562 list_move(&ip->i_ordered, &written); 563 spin_unlock(&sdp->sd_ordered_lock); 564 filemap_fdatawrite(ip->i_inode.i_mapping); 565 spin_lock(&sdp->sd_ordered_lock); 566 } 567 list_splice(&written, &sdp->sd_log_ordered); 568 spin_unlock(&sdp->sd_ordered_lock); 569 } 570 571 static void gfs2_ordered_wait(struct gfs2_sbd *sdp) 572 { 573 struct gfs2_inode *ip; 574 575 spin_lock(&sdp->sd_ordered_lock); 576 while (!list_empty(&sdp->sd_log_ordered)) { 577 ip = list_entry(sdp->sd_log_ordered.next, struct gfs2_inode, i_ordered); 578 list_del(&ip->i_ordered); 579 WARN_ON(!test_and_clear_bit(GIF_ORDERED, &ip->i_flags)); 580 if (ip->i_inode.i_mapping->nrpages == 0) 581 continue; 582 spin_unlock(&sdp->sd_ordered_lock); 583 filemap_fdatawait(ip->i_inode.i_mapping); 584 spin_lock(&sdp->sd_ordered_lock); 585 } 586 spin_unlock(&sdp->sd_ordered_lock); 587 } 588 589 void gfs2_ordered_del_inode(struct gfs2_inode *ip) 590 { 591 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 592 593 spin_lock(&sdp->sd_ordered_lock); 594 if (test_and_clear_bit(GIF_ORDERED, &ip->i_flags)) 595 list_del(&ip->i_ordered); 596 spin_unlock(&sdp->sd_ordered_lock); 597 } 598 599 void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd) 600 { 601 struct buffer_head *bh = bd->bd_bh; 602 struct gfs2_glock *gl = bd->bd_gl; 603 604 bh->b_private = NULL; 605 bd->bd_blkno = bh->b_blocknr; 606 gfs2_remove_from_ail(bd); /* drops ref on bh */ 607 bd->bd_bh = NULL; 608 sdp->sd_log_num_revoke++; 609 if (atomic_inc_return(&gl->gl_revokes) == 1) 610 gfs2_glock_hold(gl); 611 set_bit(GLF_LFLUSH, &gl->gl_flags); 612 list_add(&bd->bd_list, &sdp->sd_log_revokes); 613 } 614 615 void gfs2_glock_remove_revoke(struct gfs2_glock *gl) 616 { 617 if (atomic_dec_return(&gl->gl_revokes) == 0) { 618 clear_bit(GLF_LFLUSH, &gl->gl_flags); 619 gfs2_glock_queue_put(gl); 620 } 621 } 622 623 void gfs2_write_revokes(struct gfs2_sbd *sdp) 624 { 625 struct gfs2_trans *tr; 626 struct gfs2_bufdata *bd, *tmp; 627 int have_revokes = 0; 628 int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64); 629 630 gfs2_ail1_empty(sdp); 631 spin_lock(&sdp->sd_ail_lock); 632 list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) { 633 list_for_each_entry(bd, &tr->tr_ail2_list, bd_ail_st_list) { 634 if (list_empty(&bd->bd_list)) { 635 have_revokes = 1; 636 goto done; 637 } 638 } 639 } 640 done: 641 spin_unlock(&sdp->sd_ail_lock); 642 if (have_revokes == 0) 643 return; 644 while (sdp->sd_log_num_revoke > max_revokes) 645 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64); 646 max_revokes -= sdp->sd_log_num_revoke; 647 if (!sdp->sd_log_num_revoke) { 648 atomic_dec(&sdp->sd_log_blks_free); 649 /* If no blocks have been reserved, we need to also 650 * reserve a block for the header */ 651 if (!sdp->sd_log_blks_reserved) 652 atomic_dec(&sdp->sd_log_blks_free); 653 } 654 gfs2_log_lock(sdp); 655 spin_lock(&sdp->sd_ail_lock); 656 list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) { 657 list_for_each_entry_safe(bd, tmp, &tr->tr_ail2_list, bd_ail_st_list) { 658 if (max_revokes == 0) 659 goto out_of_blocks; 660 if (!list_empty(&bd->bd_list)) 661 continue; 662 gfs2_add_revoke(sdp, bd); 663 max_revokes--; 664 } 665 } 666 out_of_blocks: 667 spin_unlock(&sdp->sd_ail_lock); 668 gfs2_log_unlock(sdp); 669 670 if (!sdp->sd_log_num_revoke) { 671 atomic_inc(&sdp->sd_log_blks_free); 672 if (!sdp->sd_log_blks_reserved) 673 atomic_inc(&sdp->sd_log_blks_free); 674 } 675 } 676 677 /** 678 * gfs2_write_log_header - Write a journal log header buffer at lblock 679 * @sdp: The GFS2 superblock 680 * @jd: journal descriptor of the journal to which we are writing 681 * @seq: sequence number 682 * @tail: tail of the log 683 * @lblock: value for lh_blkno (block number relative to start of journal) 684 * @flags: log header flags GFS2_LOG_HEAD_* 685 * @op_flags: flags to pass to the bio 686 * 687 * Returns: the initialized log buffer descriptor 688 */ 689 690 void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd, 691 u64 seq, u32 tail, u32 lblock, u32 flags, 692 int op_flags) 693 { 694 struct gfs2_log_header *lh; 695 u32 hash, crc; 696 struct page *page; 697 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; 698 struct timespec64 tv; 699 struct super_block *sb = sdp->sd_vfs; 700 u64 dblock; 701 702 if (gfs2_withdrawn(sdp)) 703 goto out; 704 705 page = mempool_alloc(gfs2_page_pool, GFP_NOIO); 706 lh = page_address(page); 707 clear_page(lh); 708 709 lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC); 710 lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH); 711 lh->lh_header.__pad0 = cpu_to_be64(0); 712 lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH); 713 lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid); 714 lh->lh_sequence = cpu_to_be64(seq); 715 lh->lh_flags = cpu_to_be32(flags); 716 lh->lh_tail = cpu_to_be32(tail); 717 lh->lh_blkno = cpu_to_be32(lblock); 718 hash = ~crc32(~0, lh, LH_V1_SIZE); 719 lh->lh_hash = cpu_to_be32(hash); 720 721 ktime_get_coarse_real_ts64(&tv); 722 lh->lh_nsec = cpu_to_be32(tv.tv_nsec); 723 lh->lh_sec = cpu_to_be64(tv.tv_sec); 724 if (!list_empty(&jd->extent_list)) 725 dblock = gfs2_log_bmap(jd, lblock); 726 else { 727 int ret = gfs2_lblk_to_dblk(jd->jd_inode, lblock, &dblock); 728 if (gfs2_assert_withdraw(sdp, ret == 0)) 729 return; 730 } 731 lh->lh_addr = cpu_to_be64(dblock); 732 lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr); 733 734 /* We may only write local statfs, quota, etc., when writing to our 735 own journal. The values are left 0 when recovering a journal 736 different from our own. */ 737 if (!(flags & GFS2_LOG_HEAD_RECOVERY)) { 738 lh->lh_statfs_addr = 739 cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr); 740 lh->lh_quota_addr = 741 cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr); 742 743 spin_lock(&sdp->sd_statfs_spin); 744 lh->lh_local_total = cpu_to_be64(l_sc->sc_total); 745 lh->lh_local_free = cpu_to_be64(l_sc->sc_free); 746 lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes); 747 spin_unlock(&sdp->sd_statfs_spin); 748 } 749 750 BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE); 751 752 crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4, 753 sb->s_blocksize - LH_V1_SIZE - 4); 754 lh->lh_crc = cpu_to_be32(crc); 755 756 gfs2_log_write(sdp, page, sb->s_blocksize, 0, dblock); 757 gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE | op_flags); 758 out: 759 log_flush_wait(sdp); 760 } 761 762 /** 763 * log_write_header - Get and initialize a journal header buffer 764 * @sdp: The GFS2 superblock 765 * @flags: The log header flags, including log header origin 766 * 767 * Returns: the initialized log buffer descriptor 768 */ 769 770 static void log_write_header(struct gfs2_sbd *sdp, u32 flags) 771 { 772 unsigned int tail; 773 int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC; 774 enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state); 775 776 gfs2_assert_withdraw(sdp, (state != SFS_FROZEN)); 777 tail = current_tail(sdp); 778 779 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) { 780 gfs2_ordered_wait(sdp); 781 log_flush_wait(sdp); 782 op_flags = REQ_SYNC | REQ_META | REQ_PRIO; 783 } 784 sdp->sd_log_idle = (tail == sdp->sd_log_flush_head); 785 gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++, tail, 786 sdp->sd_log_flush_head, flags, op_flags); 787 gfs2_log_incr_head(sdp); 788 789 if (sdp->sd_log_tail != tail) 790 log_pull_tail(sdp, tail); 791 } 792 793 /** 794 * gfs2_log_flush - flush incore transaction(s) 795 * @sdp: the filesystem 796 * @gl: The glock structure to flush. If NULL, flush the whole incore log 797 * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags 798 * 799 */ 800 801 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags) 802 { 803 struct gfs2_trans *tr; 804 enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state); 805 806 down_write(&sdp->sd_log_flush_lock); 807 808 /* Log might have been flushed while we waited for the flush lock */ 809 if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) { 810 up_write(&sdp->sd_log_flush_lock); 811 return; 812 } 813 trace_gfs2_log_flush(sdp, 1, flags); 814 815 if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN) 816 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags); 817 818 sdp->sd_log_flush_head = sdp->sd_log_head; 819 tr = sdp->sd_log_tr; 820 if (tr) { 821 sdp->sd_log_tr = NULL; 822 INIT_LIST_HEAD(&tr->tr_ail1_list); 823 INIT_LIST_HEAD(&tr->tr_ail2_list); 824 tr->tr_first = sdp->sd_log_flush_head; 825 if (unlikely (state == SFS_FROZEN)) 826 gfs2_assert_withdraw(sdp, !tr->tr_num_buf_new && !tr->tr_num_databuf_new); 827 } 828 829 if (unlikely(state == SFS_FROZEN)) 830 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke); 831 gfs2_assert_withdraw(sdp, 832 sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke); 833 834 gfs2_ordered_write(sdp); 835 lops_before_commit(sdp, tr); 836 gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE); 837 838 if (sdp->sd_log_head != sdp->sd_log_flush_head) { 839 log_flush_wait(sdp); 840 log_write_header(sdp, flags); 841 } else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){ 842 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */ 843 trace_gfs2_log_blocks(sdp, -1); 844 log_write_header(sdp, flags); 845 } 846 lops_after_commit(sdp, tr); 847 848 gfs2_log_lock(sdp); 849 sdp->sd_log_head = sdp->sd_log_flush_head; 850 sdp->sd_log_blks_reserved = 0; 851 sdp->sd_log_commited_revoke = 0; 852 853 spin_lock(&sdp->sd_ail_lock); 854 if (tr && !list_empty(&tr->tr_ail1_list)) { 855 list_add(&tr->tr_list, &sdp->sd_ail1_list); 856 tr = NULL; 857 } 858 spin_unlock(&sdp->sd_ail_lock); 859 gfs2_log_unlock(sdp); 860 861 if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) { 862 if (!sdp->sd_log_idle) { 863 for (;;) { 864 gfs2_ail1_start(sdp); 865 gfs2_ail1_wait(sdp); 866 if (gfs2_ail1_empty(sdp)) 867 break; 868 } 869 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */ 870 trace_gfs2_log_blocks(sdp, -1); 871 log_write_header(sdp, flags); 872 sdp->sd_log_head = sdp->sd_log_flush_head; 873 } 874 if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN | 875 GFS2_LOG_HEAD_FLUSH_FREEZE)) 876 gfs2_log_shutdown(sdp); 877 if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE) 878 atomic_set(&sdp->sd_freeze_state, SFS_FROZEN); 879 } 880 881 trace_gfs2_log_flush(sdp, 0, flags); 882 up_write(&sdp->sd_log_flush_lock); 883 884 kfree(tr); 885 } 886 887 /** 888 * gfs2_merge_trans - Merge a new transaction into a cached transaction 889 * @old: Original transaction to be expanded 890 * @new: New transaction to be merged 891 */ 892 893 static void gfs2_merge_trans(struct gfs2_trans *old, struct gfs2_trans *new) 894 { 895 WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags)); 896 897 old->tr_num_buf_new += new->tr_num_buf_new; 898 old->tr_num_databuf_new += new->tr_num_databuf_new; 899 old->tr_num_buf_rm += new->tr_num_buf_rm; 900 old->tr_num_databuf_rm += new->tr_num_databuf_rm; 901 old->tr_num_revoke += new->tr_num_revoke; 902 903 list_splice_tail_init(&new->tr_databuf, &old->tr_databuf); 904 list_splice_tail_init(&new->tr_buf, &old->tr_buf); 905 } 906 907 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 908 { 909 unsigned int reserved; 910 unsigned int unused; 911 unsigned int maxres; 912 913 gfs2_log_lock(sdp); 914 915 if (sdp->sd_log_tr) { 916 gfs2_merge_trans(sdp->sd_log_tr, tr); 917 } else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) { 918 gfs2_assert_withdraw(sdp, test_bit(TR_ALLOCED, &tr->tr_flags)); 919 sdp->sd_log_tr = tr; 920 set_bit(TR_ATTACHED, &tr->tr_flags); 921 } 922 923 sdp->sd_log_commited_revoke += tr->tr_num_revoke; 924 reserved = calc_reserved(sdp); 925 maxres = sdp->sd_log_blks_reserved + tr->tr_reserved; 926 gfs2_assert_withdraw(sdp, maxres >= reserved); 927 unused = maxres - reserved; 928 atomic_add(unused, &sdp->sd_log_blks_free); 929 trace_gfs2_log_blocks(sdp, unused); 930 gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= 931 sdp->sd_jdesc->jd_blocks); 932 sdp->sd_log_blks_reserved = reserved; 933 934 gfs2_log_unlock(sdp); 935 } 936 937 /** 938 * gfs2_log_commit - Commit a transaction to the log 939 * @sdp: the filesystem 940 * @tr: the transaction 941 * 942 * We wake up gfs2_logd if the number of pinned blocks exceed thresh1 943 * or the total number of used blocks (pinned blocks plus AIL blocks) 944 * is greater than thresh2. 945 * 946 * At mount time thresh1 is 1/3rd of journal size, thresh2 is 2/3rd of 947 * journal size. 948 * 949 * Returns: errno 950 */ 951 952 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 953 { 954 log_refund(sdp, tr); 955 956 if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) || 957 ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) > 958 atomic_read(&sdp->sd_log_thresh2))) 959 wake_up(&sdp->sd_logd_waitq); 960 } 961 962 /** 963 * gfs2_log_shutdown - write a shutdown header into a journal 964 * @sdp: the filesystem 965 * 966 */ 967 968 static void gfs2_log_shutdown(struct gfs2_sbd *sdp) 969 { 970 gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved); 971 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke); 972 gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list)); 973 974 sdp->sd_log_flush_head = sdp->sd_log_head; 975 976 log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN); 977 978 gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail); 979 gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list)); 980 981 sdp->sd_log_head = sdp->sd_log_flush_head; 982 sdp->sd_log_tail = sdp->sd_log_head; 983 } 984 985 static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp) 986 { 987 return (atomic_read(&sdp->sd_log_pinned) + 988 atomic_read(&sdp->sd_log_blks_needed) >= 989 atomic_read(&sdp->sd_log_thresh1)); 990 } 991 992 static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp) 993 { 994 unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free); 995 996 if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags)) 997 return 1; 998 999 return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >= 1000 atomic_read(&sdp->sd_log_thresh2); 1001 } 1002 1003 /** 1004 * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks 1005 * @sdp: Pointer to GFS2 superblock 1006 * 1007 * Also, periodically check to make sure that we're using the most recent 1008 * journal index. 1009 */ 1010 1011 int gfs2_logd(void *data) 1012 { 1013 struct gfs2_sbd *sdp = data; 1014 unsigned long t = 1; 1015 DEFINE_WAIT(wait); 1016 bool did_flush; 1017 1018 while (!kthread_should_stop()) { 1019 1020 /* Check for errors writing to the journal */ 1021 if (sdp->sd_log_error) { 1022 gfs2_lm_withdraw(sdp, 1023 "GFS2: fsid=%s: error %d: " 1024 "withdrawing the file system to " 1025 "prevent further damage.\n", 1026 sdp->sd_fsname, sdp->sd_log_error); 1027 } 1028 1029 did_flush = false; 1030 if (gfs2_jrnl_flush_reqd(sdp) || t == 0) { 1031 gfs2_ail1_empty(sdp); 1032 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL | 1033 GFS2_LFC_LOGD_JFLUSH_REQD); 1034 did_flush = true; 1035 } 1036 1037 if (gfs2_ail_flush_reqd(sdp)) { 1038 gfs2_ail1_start(sdp); 1039 gfs2_ail1_wait(sdp); 1040 gfs2_ail1_empty(sdp); 1041 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL | 1042 GFS2_LFC_LOGD_AIL_FLUSH_REQD); 1043 did_flush = true; 1044 } 1045 1046 if (!gfs2_ail_flush_reqd(sdp) || did_flush) 1047 wake_up(&sdp->sd_log_waitq); 1048 1049 t = gfs2_tune_get(sdp, gt_logd_secs) * HZ; 1050 1051 try_to_freeze(); 1052 1053 do { 1054 prepare_to_wait(&sdp->sd_logd_waitq, &wait, 1055 TASK_INTERRUPTIBLE); 1056 if (!gfs2_ail_flush_reqd(sdp) && 1057 !gfs2_jrnl_flush_reqd(sdp) && 1058 !kthread_should_stop()) 1059 t = schedule_timeout(t); 1060 } while(t && !gfs2_ail_flush_reqd(sdp) && 1061 !gfs2_jrnl_flush_reqd(sdp) && 1062 !kthread_should_stop()); 1063 finish_wait(&sdp->sd_logd_waitq, &wait); 1064 } 1065 1066 return 0; 1067 } 1068 1069