1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 4 * Copyright (C) 2004-2006 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/mempool.h> 13 #include <linux/gfs2_ondisk.h> 14 #include <linux/bio.h> 15 #include <linux/fs.h> 16 #include <linux/list_sort.h> 17 #include <linux/blkdev.h> 18 19 #include "bmap.h" 20 #include "dir.h" 21 #include "gfs2.h" 22 #include "incore.h" 23 #include "inode.h" 24 #include "glock.h" 25 #include "glops.h" 26 #include "log.h" 27 #include "lops.h" 28 #include "meta_io.h" 29 #include "recovery.h" 30 #include "rgrp.h" 31 #include "trans.h" 32 #include "util.h" 33 #include "trace_gfs2.h" 34 35 /** 36 * gfs2_pin - Pin a buffer in memory 37 * @sdp: The superblock 38 * @bh: The buffer to be pinned 39 * 40 * The log lock must be held when calling this function 41 */ 42 void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh) 43 { 44 struct gfs2_bufdata *bd; 45 46 BUG_ON(!current->journal_info); 47 48 clear_buffer_dirty(bh); 49 if (test_set_buffer_pinned(bh)) 50 gfs2_assert_withdraw(sdp, 0); 51 if (!buffer_uptodate(bh)) 52 gfs2_io_error_bh_wd(sdp, bh); 53 bd = bh->b_private; 54 /* If this buffer is in the AIL and it has already been written 55 * to in-place disk block, remove it from the AIL. 56 */ 57 spin_lock(&sdp->sd_ail_lock); 58 if (bd->bd_tr) 59 list_move(&bd->bd_ail_st_list, &bd->bd_tr->tr_ail2_list); 60 spin_unlock(&sdp->sd_ail_lock); 61 get_bh(bh); 62 atomic_inc(&sdp->sd_log_pinned); 63 trace_gfs2_pin(bd, 1); 64 } 65 66 static bool buffer_is_rgrp(const struct gfs2_bufdata *bd) 67 { 68 return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP; 69 } 70 71 static void maybe_release_space(struct gfs2_bufdata *bd) 72 { 73 struct gfs2_glock *gl = bd->bd_gl; 74 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; 75 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl); 76 unsigned int index = bd->bd_bh->b_blocknr - gl->gl_name.ln_number; 77 struct gfs2_bitmap *bi = rgd->rd_bits + index; 78 79 if (bi->bi_clone == NULL) 80 return; 81 if (sdp->sd_args.ar_discard) 82 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL); 83 memcpy(bi->bi_clone + bi->bi_offset, 84 bd->bd_bh->b_data + bi->bi_offset, bi->bi_bytes); 85 clear_bit(GBF_FULL, &bi->bi_flags); 86 rgd->rd_free_clone = rgd->rd_free; 87 BUG_ON(rgd->rd_free_clone < rgd->rd_reserved); 88 rgd->rd_extfail_pt = rgd->rd_free; 89 } 90 91 /** 92 * gfs2_unpin - Unpin a buffer 93 * @sdp: the filesystem the buffer belongs to 94 * @bh: The buffer to unpin 95 * @ai: 96 * @flags: The inode dirty flags 97 * 98 */ 99 100 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh, 101 struct gfs2_trans *tr) 102 { 103 struct gfs2_bufdata *bd = bh->b_private; 104 105 BUG_ON(!buffer_uptodate(bh)); 106 BUG_ON(!buffer_pinned(bh)); 107 108 lock_buffer(bh); 109 mark_buffer_dirty(bh); 110 clear_buffer_pinned(bh); 111 112 if (buffer_is_rgrp(bd)) 113 maybe_release_space(bd); 114 115 spin_lock(&sdp->sd_ail_lock); 116 if (bd->bd_tr) { 117 list_del(&bd->bd_ail_st_list); 118 brelse(bh); 119 } else { 120 struct gfs2_glock *gl = bd->bd_gl; 121 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list); 122 atomic_inc(&gl->gl_ail_count); 123 } 124 bd->bd_tr = tr; 125 list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list); 126 spin_unlock(&sdp->sd_ail_lock); 127 128 clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags); 129 trace_gfs2_pin(bd, 0); 130 unlock_buffer(bh); 131 atomic_dec(&sdp->sd_log_pinned); 132 } 133 134 void gfs2_log_incr_head(struct gfs2_sbd *sdp) 135 { 136 BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) && 137 (sdp->sd_log_flush_head != sdp->sd_log_head)); 138 139 if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) 140 sdp->sd_log_flush_head = 0; 141 } 142 143 u64 gfs2_log_bmap(struct gfs2_jdesc *jd, unsigned int lblock) 144 { 145 struct gfs2_journal_extent *je; 146 147 list_for_each_entry(je, &jd->extent_list, list) { 148 if (lblock >= je->lblock && lblock < je->lblock + je->blocks) 149 return je->dblock + lblock - je->lblock; 150 } 151 152 return -1; 153 } 154 155 /** 156 * gfs2_end_log_write_bh - end log write of pagecache data with buffers 157 * @sdp: The superblock 158 * @bvec: The bio_vec 159 * @error: The i/o status 160 * 161 * This finds the relevant buffers and unlocks them and sets the 162 * error flag according to the status of the i/o request. This is 163 * used when the log is writing data which has an in-place version 164 * that is pinned in the pagecache. 165 */ 166 167 static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp, 168 struct bio_vec *bvec, 169 blk_status_t error) 170 { 171 struct buffer_head *bh, *next; 172 struct page *page = bvec->bv_page; 173 unsigned size; 174 175 bh = page_buffers(page); 176 size = bvec->bv_len; 177 while (bh_offset(bh) < bvec->bv_offset) 178 bh = bh->b_this_page; 179 do { 180 if (error) 181 mark_buffer_write_io_error(bh); 182 unlock_buffer(bh); 183 next = bh->b_this_page; 184 size -= bh->b_size; 185 brelse(bh); 186 bh = next; 187 } while(bh && size); 188 } 189 190 /** 191 * gfs2_end_log_write - end of i/o to the log 192 * @bio: The bio 193 * 194 * Each bio_vec contains either data from the pagecache or data 195 * relating to the log itself. Here we iterate over the bio_vec 196 * array, processing both kinds of data. 197 * 198 */ 199 200 static void gfs2_end_log_write(struct bio *bio) 201 { 202 struct gfs2_sbd *sdp = bio->bi_private; 203 struct bio_vec *bvec; 204 struct page *page; 205 struct bvec_iter_all iter_all; 206 207 if (bio->bi_status) { 208 if (!cmpxchg(&sdp->sd_log_error, 0, (int)bio->bi_status)) 209 fs_err(sdp, "Error %d writing to journal, jid=%u\n", 210 bio->bi_status, sdp->sd_jdesc->jd_jid); 211 gfs2_withdraw_delayed(sdp); 212 /* prevent more writes to the journal */ 213 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags); 214 wake_up(&sdp->sd_logd_waitq); 215 } 216 217 bio_for_each_segment_all(bvec, bio, iter_all) { 218 page = bvec->bv_page; 219 if (page_has_buffers(page)) 220 gfs2_end_log_write_bh(sdp, bvec, bio->bi_status); 221 else 222 mempool_free(page, gfs2_page_pool); 223 } 224 225 bio_put(bio); 226 if (atomic_dec_and_test(&sdp->sd_log_in_flight)) 227 wake_up(&sdp->sd_log_flush_wait); 228 } 229 230 /** 231 * gfs2_log_submit_bio - Submit any pending log bio 232 * @biop: Address of the bio pointer 233 * @opf: REQ_OP | op_flags 234 * 235 * Submit any pending part-built or full bio to the block device. If 236 * there is no pending bio, then this is a no-op. 237 */ 238 239 void gfs2_log_submit_bio(struct bio **biop, int opf) 240 { 241 struct bio *bio = *biop; 242 if (bio) { 243 struct gfs2_sbd *sdp = bio->bi_private; 244 atomic_inc(&sdp->sd_log_in_flight); 245 bio->bi_opf = opf; 246 submit_bio(bio); 247 *biop = NULL; 248 } 249 } 250 251 /** 252 * gfs2_log_alloc_bio - Allocate a bio 253 * @sdp: The super block 254 * @blkno: The device block number we want to write to 255 * @end_io: The bi_end_io callback 256 * 257 * Allocate a new bio, initialize it with the given parameters and return it. 258 * 259 * Returns: The newly allocated bio 260 */ 261 262 static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno, 263 bio_end_io_t *end_io) 264 { 265 struct super_block *sb = sdp->sd_vfs; 266 struct bio *bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES); 267 268 bio->bi_iter.bi_sector = blkno << sdp->sd_fsb2bb_shift; 269 bio_set_dev(bio, sb->s_bdev); 270 bio->bi_end_io = end_io; 271 bio->bi_private = sdp; 272 273 return bio; 274 } 275 276 /** 277 * gfs2_log_get_bio - Get cached log bio, or allocate a new one 278 * @sdp: The super block 279 * @blkno: The device block number we want to write to 280 * @bio: The bio to get or allocate 281 * @op: REQ_OP 282 * @end_io: The bi_end_io callback 283 * @flush: Always flush the current bio and allocate a new one? 284 * 285 * If there is a cached bio, then if the next block number is sequential 286 * with the previous one, return it, otherwise flush the bio to the 287 * device. If there is no cached bio, or we just flushed it, then 288 * allocate a new one. 289 * 290 * Returns: The bio to use for log writes 291 */ 292 293 static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno, 294 struct bio **biop, int op, 295 bio_end_io_t *end_io, bool flush) 296 { 297 struct bio *bio = *biop; 298 299 if (bio) { 300 u64 nblk; 301 302 nblk = bio_end_sector(bio); 303 nblk >>= sdp->sd_fsb2bb_shift; 304 if (blkno == nblk && !flush) 305 return bio; 306 gfs2_log_submit_bio(biop, op); 307 } 308 309 *biop = gfs2_log_alloc_bio(sdp, blkno, end_io); 310 return *biop; 311 } 312 313 /** 314 * gfs2_log_write - write to log 315 * @sdp: the filesystem 316 * @page: the page to write 317 * @size: the size of the data to write 318 * @offset: the offset within the page 319 * @blkno: block number of the log entry 320 * 321 * Try and add the page segment to the current bio. If that fails, 322 * submit the current bio to the device and create a new one, and 323 * then add the page segment to that. 324 */ 325 326 void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page, 327 unsigned size, unsigned offset, u64 blkno) 328 { 329 struct bio *bio; 330 int ret; 331 332 bio = gfs2_log_get_bio(sdp, blkno, &sdp->sd_log_bio, REQ_OP_WRITE, 333 gfs2_end_log_write, false); 334 ret = bio_add_page(bio, page, size, offset); 335 if (ret == 0) { 336 bio = gfs2_log_get_bio(sdp, blkno, &sdp->sd_log_bio, 337 REQ_OP_WRITE, gfs2_end_log_write, true); 338 ret = bio_add_page(bio, page, size, offset); 339 WARN_ON(ret == 0); 340 } 341 } 342 343 /** 344 * gfs2_log_write_bh - write a buffer's content to the log 345 * @sdp: The super block 346 * @bh: The buffer pointing to the in-place location 347 * 348 * This writes the content of the buffer to the next available location 349 * in the log. The buffer will be unlocked once the i/o to the log has 350 * completed. 351 */ 352 353 static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh) 354 { 355 u64 dblock; 356 357 dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head); 358 gfs2_log_incr_head(sdp); 359 gfs2_log_write(sdp, bh->b_page, bh->b_size, bh_offset(bh), dblock); 360 } 361 362 /** 363 * gfs2_log_write_page - write one block stored in a page, into the log 364 * @sdp: The superblock 365 * @page: The struct page 366 * 367 * This writes the first block-sized part of the page into the log. Note 368 * that the page must have been allocated from the gfs2_page_pool mempool 369 * and that after this has been called, ownership has been transferred and 370 * the page may be freed at any time. 371 */ 372 373 void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page) 374 { 375 struct super_block *sb = sdp->sd_vfs; 376 u64 dblock; 377 378 dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head); 379 gfs2_log_incr_head(sdp); 380 gfs2_log_write(sdp, page, sb->s_blocksize, 0, dblock); 381 } 382 383 /** 384 * gfs2_end_log_read - end I/O callback for reads from the log 385 * @bio: The bio 386 * 387 * Simply unlock the pages in the bio. The main thread will wait on them and 388 * process them in order as necessary. 389 */ 390 391 static void gfs2_end_log_read(struct bio *bio) 392 { 393 struct page *page; 394 struct bio_vec *bvec; 395 struct bvec_iter_all iter_all; 396 397 bio_for_each_segment_all(bvec, bio, iter_all) { 398 page = bvec->bv_page; 399 if (bio->bi_status) { 400 int err = blk_status_to_errno(bio->bi_status); 401 402 SetPageError(page); 403 mapping_set_error(page->mapping, err); 404 } 405 unlock_page(page); 406 } 407 408 bio_put(bio); 409 } 410 411 /** 412 * gfs2_jhead_pg_srch - Look for the journal head in a given page. 413 * @jd: The journal descriptor 414 * @page: The page to look in 415 * 416 * Returns: 1 if found, 0 otherwise. 417 */ 418 419 static bool gfs2_jhead_pg_srch(struct gfs2_jdesc *jd, 420 struct gfs2_log_header_host *head, 421 struct page *page) 422 { 423 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 424 struct gfs2_log_header_host lh; 425 void *kaddr = kmap_atomic(page); 426 unsigned int offset; 427 bool ret = false; 428 429 for (offset = 0; offset < PAGE_SIZE; offset += sdp->sd_sb.sb_bsize) { 430 if (!__get_log_header(sdp, kaddr + offset, 0, &lh)) { 431 if (lh.lh_sequence >= head->lh_sequence) 432 *head = lh; 433 else { 434 ret = true; 435 break; 436 } 437 } 438 } 439 kunmap_atomic(kaddr); 440 return ret; 441 } 442 443 /** 444 * gfs2_jhead_process_page - Search/cleanup a page 445 * @jd: The journal descriptor 446 * @index: Index of the page to look into 447 * @done: If set, perform only cleanup, else search and set if found. 448 * 449 * Find the page with 'index' in the journal's mapping. Search the page for 450 * the journal head if requested (cleanup == false). Release refs on the 451 * page so the page cache can reclaim it (put_page() twice). We grabbed a 452 * reference on this page two times, first when we did a find_or_create_page() 453 * to obtain the page to add it to the bio and second when we do a 454 * find_get_page() here to get the page to wait on while I/O on it is being 455 * completed. 456 * This function is also used to free up a page we might've grabbed but not 457 * used. Maybe we added it to a bio, but not submitted it for I/O. Or we 458 * submitted the I/O, but we already found the jhead so we only need to drop 459 * our references to the page. 460 */ 461 462 static void gfs2_jhead_process_page(struct gfs2_jdesc *jd, unsigned long index, 463 struct gfs2_log_header_host *head, 464 bool *done) 465 { 466 struct page *page; 467 468 page = find_get_page(jd->jd_inode->i_mapping, index); 469 wait_on_page_locked(page); 470 471 if (PageError(page)) 472 *done = true; 473 474 if (!*done) 475 *done = gfs2_jhead_pg_srch(jd, head, page); 476 477 put_page(page); /* Once for find_get_page */ 478 put_page(page); /* Once more for find_or_create_page */ 479 } 480 481 static struct bio *gfs2_chain_bio(struct bio *prev, unsigned int nr_iovecs) 482 { 483 struct bio *new; 484 485 new = bio_alloc(GFP_NOIO, nr_iovecs); 486 bio_copy_dev(new, prev); 487 new->bi_iter.bi_sector = bio_end_sector(prev); 488 new->bi_opf = prev->bi_opf; 489 new->bi_write_hint = prev->bi_write_hint; 490 bio_chain(new, prev); 491 submit_bio(prev); 492 return new; 493 } 494 495 /** 496 * gfs2_find_jhead - find the head of a log 497 * @jd: The journal descriptor 498 * @head: The log descriptor for the head of the log is returned here 499 * 500 * Do a search of a journal by reading it in large chunks using bios and find 501 * the valid log entry with the highest sequence number. (i.e. the log head) 502 * 503 * Returns: 0 on success, errno otherwise 504 */ 505 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head, 506 bool keep_cache) 507 { 508 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 509 struct address_space *mapping = jd->jd_inode->i_mapping; 510 unsigned int block = 0, blocks_submitted = 0, blocks_read = 0; 511 unsigned int bsize = sdp->sd_sb.sb_bsize, off; 512 unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift; 513 unsigned int shift = PAGE_SHIFT - bsize_shift; 514 unsigned int max_blocks = 2 * 1024 * 1024 >> bsize_shift; 515 struct gfs2_journal_extent *je; 516 int sz, ret = 0; 517 struct bio *bio = NULL; 518 struct page *page = NULL; 519 bool done = false; 520 errseq_t since; 521 522 memset(head, 0, sizeof(*head)); 523 if (list_empty(&jd->extent_list)) 524 gfs2_map_journal_extents(sdp, jd); 525 526 since = filemap_sample_wb_err(mapping); 527 list_for_each_entry(je, &jd->extent_list, list) { 528 u64 dblock = je->dblock; 529 530 for (; block < je->lblock + je->blocks; block++, dblock++) { 531 if (!page) { 532 page = find_or_create_page(mapping, 533 block >> shift, GFP_NOFS); 534 if (!page) { 535 ret = -ENOMEM; 536 done = true; 537 goto out; 538 } 539 off = 0; 540 } 541 542 if (bio && (off || block < blocks_submitted + max_blocks)) { 543 sector_t sector = dblock << sdp->sd_fsb2bb_shift; 544 545 if (bio_end_sector(bio) == sector) { 546 sz = bio_add_page(bio, page, bsize, off); 547 if (sz == bsize) 548 goto block_added; 549 } 550 if (off) { 551 unsigned int blocks = 552 (PAGE_SIZE - off) >> bsize_shift; 553 554 bio = gfs2_chain_bio(bio, blocks); 555 goto add_block_to_new_bio; 556 } 557 } 558 559 if (bio) { 560 blocks_submitted = block; 561 submit_bio(bio); 562 } 563 564 bio = gfs2_log_alloc_bio(sdp, dblock, gfs2_end_log_read); 565 bio->bi_opf = REQ_OP_READ; 566 add_block_to_new_bio: 567 sz = bio_add_page(bio, page, bsize, off); 568 BUG_ON(sz != bsize); 569 block_added: 570 off += bsize; 571 if (off == PAGE_SIZE) 572 page = NULL; 573 if (blocks_submitted <= blocks_read + max_blocks) { 574 /* Keep at least one bio in flight */ 575 continue; 576 } 577 578 gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done); 579 blocks_read += PAGE_SIZE >> bsize_shift; 580 if (done) 581 goto out; /* found */ 582 } 583 } 584 585 out: 586 if (bio) 587 submit_bio(bio); 588 while (blocks_read < block) { 589 gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done); 590 blocks_read += PAGE_SIZE >> bsize_shift; 591 } 592 593 if (!ret) 594 ret = filemap_check_wb_err(mapping, since); 595 596 if (!keep_cache) 597 truncate_inode_pages(mapping, 0); 598 599 return ret; 600 } 601 602 static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type, 603 u32 ld_length, u32 ld_data1) 604 { 605 struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO); 606 struct gfs2_log_descriptor *ld = page_address(page); 607 clear_page(ld); 608 ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC); 609 ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD); 610 ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD); 611 ld->ld_type = cpu_to_be32(ld_type); 612 ld->ld_length = cpu_to_be32(ld_length); 613 ld->ld_data1 = cpu_to_be32(ld_data1); 614 ld->ld_data2 = 0; 615 return page; 616 } 617 618 static void gfs2_check_magic(struct buffer_head *bh) 619 { 620 void *kaddr; 621 __be32 *ptr; 622 623 clear_buffer_escaped(bh); 624 kaddr = kmap_atomic(bh->b_page); 625 ptr = kaddr + bh_offset(bh); 626 if (*ptr == cpu_to_be32(GFS2_MAGIC)) 627 set_buffer_escaped(bh); 628 kunmap_atomic(kaddr); 629 } 630 631 static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b) 632 { 633 struct gfs2_bufdata *bda, *bdb; 634 635 bda = list_entry(a, struct gfs2_bufdata, bd_list); 636 bdb = list_entry(b, struct gfs2_bufdata, bd_list); 637 638 if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr) 639 return -1; 640 if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr) 641 return 1; 642 return 0; 643 } 644 645 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit, 646 unsigned int total, struct list_head *blist, 647 bool is_databuf) 648 { 649 struct gfs2_log_descriptor *ld; 650 struct gfs2_bufdata *bd1 = NULL, *bd2; 651 struct page *page; 652 unsigned int num; 653 unsigned n; 654 __be64 *ptr; 655 656 gfs2_log_lock(sdp); 657 list_sort(NULL, blist, blocknr_cmp); 658 bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list); 659 while(total) { 660 num = total; 661 if (total > limit) 662 num = limit; 663 gfs2_log_unlock(sdp); 664 page = gfs2_get_log_desc(sdp, 665 is_databuf ? GFS2_LOG_DESC_JDATA : 666 GFS2_LOG_DESC_METADATA, num + 1, num); 667 ld = page_address(page); 668 gfs2_log_lock(sdp); 669 ptr = (__be64 *)(ld + 1); 670 671 n = 0; 672 list_for_each_entry_continue(bd1, blist, bd_list) { 673 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr); 674 if (is_databuf) { 675 gfs2_check_magic(bd1->bd_bh); 676 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0); 677 } 678 if (++n >= num) 679 break; 680 } 681 682 gfs2_log_unlock(sdp); 683 gfs2_log_write_page(sdp, page); 684 gfs2_log_lock(sdp); 685 686 n = 0; 687 list_for_each_entry_continue(bd2, blist, bd_list) { 688 get_bh(bd2->bd_bh); 689 gfs2_log_unlock(sdp); 690 lock_buffer(bd2->bd_bh); 691 692 if (buffer_escaped(bd2->bd_bh)) { 693 void *kaddr; 694 page = mempool_alloc(gfs2_page_pool, GFP_NOIO); 695 ptr = page_address(page); 696 kaddr = kmap_atomic(bd2->bd_bh->b_page); 697 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh), 698 bd2->bd_bh->b_size); 699 kunmap_atomic(kaddr); 700 *(__be32 *)ptr = 0; 701 clear_buffer_escaped(bd2->bd_bh); 702 unlock_buffer(bd2->bd_bh); 703 brelse(bd2->bd_bh); 704 gfs2_log_write_page(sdp, page); 705 } else { 706 gfs2_log_write_bh(sdp, bd2->bd_bh); 707 } 708 gfs2_log_lock(sdp); 709 if (++n >= num) 710 break; 711 } 712 713 BUG_ON(total < num); 714 total -= num; 715 } 716 gfs2_log_unlock(sdp); 717 } 718 719 static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 720 { 721 unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */ 722 unsigned int nbuf; 723 if (tr == NULL) 724 return; 725 nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm; 726 gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0); 727 } 728 729 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 730 { 731 struct list_head *head; 732 struct gfs2_bufdata *bd; 733 734 if (tr == NULL) 735 return; 736 737 head = &tr->tr_buf; 738 while (!list_empty(head)) { 739 bd = list_first_entry(head, struct gfs2_bufdata, bd_list); 740 list_del_init(&bd->bd_list); 741 gfs2_unpin(sdp, bd->bd_bh, tr); 742 } 743 } 744 745 static void buf_lo_before_scan(struct gfs2_jdesc *jd, 746 struct gfs2_log_header_host *head, int pass) 747 { 748 if (pass != 0) 749 return; 750 751 jd->jd_found_blocks = 0; 752 jd->jd_replayed_blocks = 0; 753 } 754 755 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start, 756 struct gfs2_log_descriptor *ld, __be64 *ptr, 757 int pass) 758 { 759 struct gfs2_inode *ip = GFS2_I(jd->jd_inode); 760 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 761 struct gfs2_glock *gl = ip->i_gl; 762 unsigned int blks = be32_to_cpu(ld->ld_data1); 763 struct buffer_head *bh_log, *bh_ip; 764 u64 blkno; 765 int error = 0; 766 767 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA) 768 return 0; 769 770 gfs2_replay_incr_blk(jd, &start); 771 772 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) { 773 blkno = be64_to_cpu(*ptr++); 774 775 jd->jd_found_blocks++; 776 777 if (gfs2_revoke_check(jd, blkno, start)) 778 continue; 779 780 error = gfs2_replay_read_block(jd, start, &bh_log); 781 if (error) 782 return error; 783 784 bh_ip = gfs2_meta_new(gl, blkno); 785 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size); 786 787 if (gfs2_meta_check(sdp, bh_ip)) 788 error = -EIO; 789 else { 790 struct gfs2_meta_header *mh = 791 (struct gfs2_meta_header *)bh_ip->b_data; 792 793 if (mh->mh_type == cpu_to_be32(GFS2_METATYPE_RG)) { 794 struct gfs2_rgrpd *rgd; 795 796 rgd = gfs2_blk2rgrpd(sdp, blkno, false); 797 if (rgd && rgd->rd_addr == blkno && 798 rgd->rd_bits && rgd->rd_bits->bi_bh) { 799 fs_info(sdp, "Replaying 0x%llx but we " 800 "already have a bh!\n", 801 (unsigned long long)blkno); 802 fs_info(sdp, "busy:%d, pinned:%d\n", 803 buffer_busy(rgd->rd_bits->bi_bh) ? 1 : 0, 804 buffer_pinned(rgd->rd_bits->bi_bh)); 805 gfs2_dump_glock(NULL, rgd->rd_gl, true); 806 } 807 } 808 mark_buffer_dirty(bh_ip); 809 } 810 brelse(bh_log); 811 brelse(bh_ip); 812 813 if (error) 814 break; 815 816 jd->jd_replayed_blocks++; 817 } 818 819 return error; 820 } 821 822 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) 823 { 824 struct gfs2_inode *ip = GFS2_I(jd->jd_inode); 825 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 826 827 if (error) { 828 gfs2_inode_metasync(ip->i_gl); 829 return; 830 } 831 if (pass != 1) 832 return; 833 834 gfs2_inode_metasync(ip->i_gl); 835 836 fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n", 837 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks); 838 } 839 840 static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 841 { 842 struct gfs2_meta_header *mh; 843 unsigned int offset; 844 struct list_head *head = &sdp->sd_log_revokes; 845 struct gfs2_bufdata *bd; 846 struct page *page; 847 unsigned int length; 848 849 gfs2_write_revokes(sdp); 850 if (!sdp->sd_log_num_revoke) 851 return; 852 853 length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke); 854 page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke); 855 offset = sizeof(struct gfs2_log_descriptor); 856 857 list_for_each_entry(bd, head, bd_list) { 858 sdp->sd_log_num_revoke--; 859 860 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) { 861 862 gfs2_log_write_page(sdp, page); 863 page = mempool_alloc(gfs2_page_pool, GFP_NOIO); 864 mh = page_address(page); 865 clear_page(mh); 866 mh->mh_magic = cpu_to_be32(GFS2_MAGIC); 867 mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB); 868 mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB); 869 offset = sizeof(struct gfs2_meta_header); 870 } 871 872 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno); 873 offset += sizeof(u64); 874 } 875 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke); 876 877 gfs2_log_write_page(sdp, page); 878 } 879 880 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 881 { 882 struct list_head *head = &sdp->sd_log_revokes; 883 struct gfs2_bufdata *bd; 884 struct gfs2_glock *gl; 885 886 while (!list_empty(head)) { 887 bd = list_first_entry(head, struct gfs2_bufdata, bd_list); 888 list_del_init(&bd->bd_list); 889 gl = bd->bd_gl; 890 gfs2_glock_remove_revoke(gl); 891 kmem_cache_free(gfs2_bufdata_cachep, bd); 892 } 893 } 894 895 static void revoke_lo_before_scan(struct gfs2_jdesc *jd, 896 struct gfs2_log_header_host *head, int pass) 897 { 898 if (pass != 0) 899 return; 900 901 jd->jd_found_revokes = 0; 902 jd->jd_replay_tail = head->lh_tail; 903 } 904 905 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, u32 start, 906 struct gfs2_log_descriptor *ld, __be64 *ptr, 907 int pass) 908 { 909 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 910 unsigned int blks = be32_to_cpu(ld->ld_length); 911 unsigned int revokes = be32_to_cpu(ld->ld_data1); 912 struct buffer_head *bh; 913 unsigned int offset; 914 u64 blkno; 915 int first = 1; 916 int error; 917 918 if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE) 919 return 0; 920 921 offset = sizeof(struct gfs2_log_descriptor); 922 923 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) { 924 error = gfs2_replay_read_block(jd, start, &bh); 925 if (error) 926 return error; 927 928 if (!first) 929 gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB); 930 931 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) { 932 blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset)); 933 934 error = gfs2_revoke_add(jd, blkno, start); 935 if (error < 0) { 936 brelse(bh); 937 return error; 938 } 939 else if (error) 940 jd->jd_found_revokes++; 941 942 if (!--revokes) 943 break; 944 offset += sizeof(u64); 945 } 946 947 brelse(bh); 948 offset = sizeof(struct gfs2_meta_header); 949 first = 0; 950 } 951 952 return 0; 953 } 954 955 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) 956 { 957 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 958 959 if (error) { 960 gfs2_revoke_clean(jd); 961 return; 962 } 963 if (pass != 1) 964 return; 965 966 fs_info(sdp, "jid=%u: Found %u revoke tags\n", 967 jd->jd_jid, jd->jd_found_revokes); 968 969 gfs2_revoke_clean(jd); 970 } 971 972 /** 973 * databuf_lo_before_commit - Scan the data buffers, writing as we go 974 * 975 */ 976 977 static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 978 { 979 unsigned int limit = databuf_limit(sdp); 980 unsigned int nbuf; 981 if (tr == NULL) 982 return; 983 nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm; 984 gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1); 985 } 986 987 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start, 988 struct gfs2_log_descriptor *ld, 989 __be64 *ptr, int pass) 990 { 991 struct gfs2_inode *ip = GFS2_I(jd->jd_inode); 992 struct gfs2_glock *gl = ip->i_gl; 993 unsigned int blks = be32_to_cpu(ld->ld_data1); 994 struct buffer_head *bh_log, *bh_ip; 995 u64 blkno; 996 u64 esc; 997 int error = 0; 998 999 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA) 1000 return 0; 1001 1002 gfs2_replay_incr_blk(jd, &start); 1003 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) { 1004 blkno = be64_to_cpu(*ptr++); 1005 esc = be64_to_cpu(*ptr++); 1006 1007 jd->jd_found_blocks++; 1008 1009 if (gfs2_revoke_check(jd, blkno, start)) 1010 continue; 1011 1012 error = gfs2_replay_read_block(jd, start, &bh_log); 1013 if (error) 1014 return error; 1015 1016 bh_ip = gfs2_meta_new(gl, blkno); 1017 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size); 1018 1019 /* Unescape */ 1020 if (esc) { 1021 __be32 *eptr = (__be32 *)bh_ip->b_data; 1022 *eptr = cpu_to_be32(GFS2_MAGIC); 1023 } 1024 mark_buffer_dirty(bh_ip); 1025 1026 brelse(bh_log); 1027 brelse(bh_ip); 1028 1029 jd->jd_replayed_blocks++; 1030 } 1031 1032 return error; 1033 } 1034 1035 /* FIXME: sort out accounting for log blocks etc. */ 1036 1037 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) 1038 { 1039 struct gfs2_inode *ip = GFS2_I(jd->jd_inode); 1040 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 1041 1042 if (error) { 1043 gfs2_inode_metasync(ip->i_gl); 1044 return; 1045 } 1046 if (pass != 1) 1047 return; 1048 1049 /* data sync? */ 1050 gfs2_inode_metasync(ip->i_gl); 1051 1052 fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n", 1053 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks); 1054 } 1055 1056 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 1057 { 1058 struct list_head *head; 1059 struct gfs2_bufdata *bd; 1060 1061 if (tr == NULL) 1062 return; 1063 1064 head = &tr->tr_databuf; 1065 while (!list_empty(head)) { 1066 bd = list_first_entry(head, struct gfs2_bufdata, bd_list); 1067 list_del_init(&bd->bd_list); 1068 gfs2_unpin(sdp, bd->bd_bh, tr); 1069 } 1070 } 1071 1072 1073 static const struct gfs2_log_operations gfs2_buf_lops = { 1074 .lo_before_commit = buf_lo_before_commit, 1075 .lo_after_commit = buf_lo_after_commit, 1076 .lo_before_scan = buf_lo_before_scan, 1077 .lo_scan_elements = buf_lo_scan_elements, 1078 .lo_after_scan = buf_lo_after_scan, 1079 .lo_name = "buf", 1080 }; 1081 1082 static const struct gfs2_log_operations gfs2_revoke_lops = { 1083 .lo_before_commit = revoke_lo_before_commit, 1084 .lo_after_commit = revoke_lo_after_commit, 1085 .lo_before_scan = revoke_lo_before_scan, 1086 .lo_scan_elements = revoke_lo_scan_elements, 1087 .lo_after_scan = revoke_lo_after_scan, 1088 .lo_name = "revoke", 1089 }; 1090 1091 static const struct gfs2_log_operations gfs2_databuf_lops = { 1092 .lo_before_commit = databuf_lo_before_commit, 1093 .lo_after_commit = databuf_lo_after_commit, 1094 .lo_scan_elements = databuf_lo_scan_elements, 1095 .lo_after_scan = databuf_lo_after_scan, 1096 .lo_name = "databuf", 1097 }; 1098 1099 const struct gfs2_log_operations *gfs2_log_ops[] = { 1100 &gfs2_databuf_lops, 1101 &gfs2_buf_lops, 1102 &gfs2_revoke_lops, 1103 NULL, 1104 }; 1105 1106