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