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 rgd->rd_extfail_pt = rgd->rd_free; 88 } 89 90 /** 91 * gfs2_unpin - Unpin a buffer 92 * @sdp: the filesystem the buffer belongs to 93 * @bh: The buffer to unpin 94 * @ai: 95 * @flags: The inode dirty flags 96 * 97 */ 98 99 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh, 100 struct gfs2_trans *tr) 101 { 102 struct gfs2_bufdata *bd = bh->b_private; 103 104 BUG_ON(!buffer_uptodate(bh)); 105 BUG_ON(!buffer_pinned(bh)); 106 107 lock_buffer(bh); 108 mark_buffer_dirty(bh); 109 clear_buffer_pinned(bh); 110 111 if (buffer_is_rgrp(bd)) 112 maybe_release_space(bd); 113 114 spin_lock(&sdp->sd_ail_lock); 115 if (bd->bd_tr) { 116 list_del(&bd->bd_ail_st_list); 117 brelse(bh); 118 } else { 119 struct gfs2_glock *gl = bd->bd_gl; 120 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list); 121 atomic_inc(&gl->gl_ail_count); 122 } 123 bd->bd_tr = tr; 124 list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list); 125 spin_unlock(&sdp->sd_ail_lock); 126 127 clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags); 128 trace_gfs2_pin(bd, 0); 129 unlock_buffer(bh); 130 atomic_dec(&sdp->sd_log_pinned); 131 } 132 133 void gfs2_log_incr_head(struct gfs2_sbd *sdp) 134 { 135 BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) && 136 (sdp->sd_log_flush_head != sdp->sd_log_head)); 137 138 if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) 139 sdp->sd_log_flush_head = 0; 140 } 141 142 u64 gfs2_log_bmap(struct gfs2_jdesc *jd, unsigned int lblock) 143 { 144 struct gfs2_journal_extent *je; 145 146 list_for_each_entry(je, &jd->extent_list, list) { 147 if (lblock >= je->lblock && lblock < je->lblock + je->blocks) 148 return je->dblock + lblock - je->lblock; 149 } 150 151 return -1; 152 } 153 154 /** 155 * gfs2_end_log_write_bh - end log write of pagecache data with buffers 156 * @sdp: The superblock 157 * @bvec: The bio_vec 158 * @error: The i/o status 159 * 160 * This finds the relevant buffers and unlocks them and sets the 161 * error flag according to the status of the i/o request. This is 162 * used when the log is writing data which has an in-place version 163 * that is pinned in the pagecache. 164 */ 165 166 static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp, 167 struct bio_vec *bvec, 168 blk_status_t error) 169 { 170 struct buffer_head *bh, *next; 171 struct page *page = bvec->bv_page; 172 unsigned size; 173 174 bh = page_buffers(page); 175 size = bvec->bv_len; 176 while (bh_offset(bh) < bvec->bv_offset) 177 bh = bh->b_this_page; 178 do { 179 if (error) 180 mark_buffer_write_io_error(bh); 181 unlock_buffer(bh); 182 next = bh->b_this_page; 183 size -= bh->b_size; 184 brelse(bh); 185 bh = next; 186 } while(bh && size); 187 } 188 189 /** 190 * gfs2_end_log_write - end of i/o to the log 191 * @bio: The bio 192 * 193 * Each bio_vec contains either data from the pagecache or data 194 * relating to the log itself. Here we iterate over the bio_vec 195 * array, processing both kinds of data. 196 * 197 */ 198 199 static void gfs2_end_log_write(struct bio *bio) 200 { 201 struct gfs2_sbd *sdp = bio->bi_private; 202 struct bio_vec *bvec; 203 struct page *page; 204 struct bvec_iter_all iter_all; 205 206 if (bio->bi_status) { 207 if (!cmpxchg(&sdp->sd_log_error, 0, (int)bio->bi_status)) 208 fs_err(sdp, "Error %d writing to journal, jid=%u\n", 209 bio->bi_status, sdp->sd_jdesc->jd_jid); 210 gfs2_withdraw_delayed(sdp); 211 /* prevent more writes to the journal */ 212 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags); 213 wake_up(&sdp->sd_logd_waitq); 214 } 215 216 bio_for_each_segment_all(bvec, bio, iter_all) { 217 page = bvec->bv_page; 218 if (page_has_buffers(page)) 219 gfs2_end_log_write_bh(sdp, bvec, bio->bi_status); 220 else 221 mempool_free(page, gfs2_page_pool); 222 } 223 224 bio_put(bio); 225 if (atomic_dec_and_test(&sdp->sd_log_in_flight)) 226 wake_up(&sdp->sd_log_flush_wait); 227 } 228 229 /** 230 * gfs2_log_submit_bio - Submit any pending log bio 231 * @biop: Address of the bio pointer 232 * @opf: REQ_OP | op_flags 233 * 234 * Submit any pending part-built or full bio to the block device. If 235 * there is no pending bio, then this is a no-op. 236 */ 237 238 void gfs2_log_submit_bio(struct bio **biop, int opf) 239 { 240 struct bio *bio = *biop; 241 if (bio) { 242 struct gfs2_sbd *sdp = bio->bi_private; 243 atomic_inc(&sdp->sd_log_in_flight); 244 bio->bi_opf = opf; 245 submit_bio(bio); 246 *biop = NULL; 247 } 248 } 249 250 /** 251 * gfs2_log_alloc_bio - Allocate a bio 252 * @sdp: The super block 253 * @blkno: The device block number we want to write to 254 * @end_io: The bi_end_io callback 255 * 256 * Allocate a new bio, initialize it with the given parameters and return it. 257 * 258 * Returns: The newly allocated bio 259 */ 260 261 static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno, 262 bio_end_io_t *end_io) 263 { 264 struct super_block *sb = sdp->sd_vfs; 265 struct bio *bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES); 266 267 bio->bi_iter.bi_sector = blkno << sdp->sd_fsb2bb_shift; 268 bio_set_dev(bio, sb->s_bdev); 269 bio->bi_end_io = end_io; 270 bio->bi_private = sdp; 271 272 return bio; 273 } 274 275 /** 276 * gfs2_log_get_bio - Get cached log bio, or allocate a new one 277 * @sdp: The super block 278 * @blkno: The device block number we want to write to 279 * @bio: The bio to get or allocate 280 * @op: REQ_OP 281 * @end_io: The bi_end_io callback 282 * @flush: Always flush the current bio and allocate a new one? 283 * 284 * If there is a cached bio, then if the next block number is sequential 285 * with the previous one, return it, otherwise flush the bio to the 286 * device. If there is no cached bio, or we just flushed it, then 287 * allocate a new one. 288 * 289 * Returns: The bio to use for log writes 290 */ 291 292 static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno, 293 struct bio **biop, int op, 294 bio_end_io_t *end_io, bool flush) 295 { 296 struct bio *bio = *biop; 297 298 if (bio) { 299 u64 nblk; 300 301 nblk = bio_end_sector(bio); 302 nblk >>= sdp->sd_fsb2bb_shift; 303 if (blkno == nblk && !flush) 304 return bio; 305 gfs2_log_submit_bio(biop, op); 306 } 307 308 *biop = gfs2_log_alloc_bio(sdp, blkno, end_io); 309 return *biop; 310 } 311 312 /** 313 * gfs2_log_write - write to log 314 * @sdp: the filesystem 315 * @page: the page to write 316 * @size: the size of the data to write 317 * @offset: the offset within the page 318 * @blkno: block number of the log entry 319 * 320 * Try and add the page segment to the current bio. If that fails, 321 * submit the current bio to the device and create a new one, and 322 * then add the page segment to that. 323 */ 324 325 void gfs2_log_write(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd, 326 struct page *page, unsigned size, unsigned offset, 327 u64 blkno) 328 { 329 struct bio *bio; 330 int ret; 331 332 bio = gfs2_log_get_bio(sdp, blkno, &jd->jd_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, &jd->jd_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, sdp->sd_jdesc, bh->b_page, bh->b_size, 360 bh_offset(bh), dblock); 361 } 362 363 /** 364 * gfs2_log_write_page - write one block stored in a page, into the log 365 * @sdp: The superblock 366 * @page: The struct page 367 * 368 * This writes the first block-sized part of the page into the log. Note 369 * that the page must have been allocated from the gfs2_page_pool mempool 370 * and that after this has been called, ownership has been transferred and 371 * the page may be freed at any time. 372 */ 373 374 static void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page) 375 { 376 struct super_block *sb = sdp->sd_vfs; 377 u64 dblock; 378 379 dblock = gfs2_log_bmap(sdp->sd_jdesc, sdp->sd_log_flush_head); 380 gfs2_log_incr_head(sdp); 381 gfs2_log_write(sdp, sdp->sd_jdesc, page, sb->s_blocksize, 0, dblock); 382 } 383 384 /** 385 * gfs2_end_log_read - end I/O callback for reads from the log 386 * @bio: The bio 387 * 388 * Simply unlock the pages in the bio. The main thread will wait on them and 389 * process them in order as necessary. 390 */ 391 392 static void gfs2_end_log_read(struct bio *bio) 393 { 394 struct page *page; 395 struct bio_vec *bvec; 396 struct bvec_iter_all iter_all; 397 398 bio_for_each_segment_all(bvec, bio, iter_all) { 399 page = bvec->bv_page; 400 if (bio->bi_status) { 401 int err = blk_status_to_errno(bio->bi_status); 402 403 SetPageError(page); 404 mapping_set_error(page->mapping, err); 405 } 406 unlock_page(page); 407 } 408 409 bio_put(bio); 410 } 411 412 /** 413 * gfs2_jhead_pg_srch - Look for the journal head in a given page. 414 * @jd: The journal descriptor 415 * @page: The page to look in 416 * 417 * Returns: 1 if found, 0 otherwise. 418 */ 419 420 static bool gfs2_jhead_pg_srch(struct gfs2_jdesc *jd, 421 struct gfs2_log_header_host *head, 422 struct page *page) 423 { 424 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 425 struct gfs2_log_header_host lh; 426 void *kaddr = kmap_atomic(page); 427 unsigned int offset; 428 bool ret = false; 429 430 for (offset = 0; offset < PAGE_SIZE; offset += sdp->sd_sb.sb_bsize) { 431 if (!__get_log_header(sdp, kaddr + offset, 0, &lh)) { 432 if (lh.lh_sequence >= head->lh_sequence) 433 *head = lh; 434 else { 435 ret = true; 436 break; 437 } 438 } 439 } 440 kunmap_atomic(kaddr); 441 return ret; 442 } 443 444 /** 445 * gfs2_jhead_process_page - Search/cleanup a page 446 * @jd: The journal descriptor 447 * @index: Index of the page to look into 448 * @done: If set, perform only cleanup, else search and set if found. 449 * 450 * Find the page with 'index' in the journal's mapping. Search the page for 451 * the journal head if requested (cleanup == false). Release refs on the 452 * page so the page cache can reclaim it (put_page() twice). We grabbed a 453 * reference on this page two times, first when we did a find_or_create_page() 454 * to obtain the page to add it to the bio and second when we do a 455 * find_get_page() here to get the page to wait on while I/O on it is being 456 * completed. 457 * This function is also used to free up a page we might've grabbed but not 458 * used. Maybe we added it to a bio, but not submitted it for I/O. Or we 459 * submitted the I/O, but we already found the jhead so we only need to drop 460 * our references to the page. 461 */ 462 463 static void gfs2_jhead_process_page(struct gfs2_jdesc *jd, unsigned long index, 464 struct gfs2_log_header_host *head, 465 bool *done) 466 { 467 struct page *page; 468 469 page = find_get_page(jd->jd_inode->i_mapping, index); 470 wait_on_page_locked(page); 471 472 if (PageError(page)) 473 *done = true; 474 475 if (!*done) 476 *done = gfs2_jhead_pg_srch(jd, head, page); 477 478 put_page(page); /* Once for find_get_page */ 479 put_page(page); /* Once more for find_or_create_page */ 480 } 481 482 static struct bio *gfs2_chain_bio(struct bio *prev, unsigned int nr_iovecs) 483 { 484 struct bio *new; 485 486 new = bio_alloc(GFP_NOIO, nr_iovecs); 487 bio_copy_dev(new, prev); 488 new->bi_iter.bi_sector = bio_end_sector(prev); 489 new->bi_opf = prev->bi_opf; 490 new->bi_write_hint = prev->bi_write_hint; 491 bio_chain(new, prev); 492 submit_bio(prev); 493 return new; 494 } 495 496 /** 497 * gfs2_find_jhead - find the head of a log 498 * @jd: The journal descriptor 499 * @head: The log descriptor for the head of the log is returned here 500 * 501 * Do a search of a journal by reading it in large chunks using bios and find 502 * the valid log entry with the highest sequence number. (i.e. the log head) 503 * 504 * Returns: 0 on success, errno otherwise 505 */ 506 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head, 507 bool keep_cache) 508 { 509 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 510 struct address_space *mapping = jd->jd_inode->i_mapping; 511 unsigned int block = 0, blocks_submitted = 0, blocks_read = 0; 512 unsigned int bsize = sdp->sd_sb.sb_bsize, off; 513 unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift; 514 unsigned int shift = PAGE_SHIFT - bsize_shift; 515 unsigned int max_blocks = 2 * 1024 * 1024 >> bsize_shift; 516 struct gfs2_journal_extent *je; 517 int sz, ret = 0; 518 struct bio *bio = NULL; 519 struct page *page = NULL; 520 bool done = false; 521 errseq_t since; 522 523 memset(head, 0, sizeof(*head)); 524 if (list_empty(&jd->extent_list)) 525 gfs2_map_journal_extents(sdp, jd); 526 527 since = filemap_sample_wb_err(mapping); 528 list_for_each_entry(je, &jd->extent_list, list) { 529 u64 dblock = je->dblock; 530 531 for (; block < je->lblock + je->blocks; block++, dblock++) { 532 if (!page) { 533 page = find_or_create_page(mapping, 534 block >> shift, GFP_NOFS); 535 if (!page) { 536 ret = -ENOMEM; 537 done = true; 538 goto out; 539 } 540 off = 0; 541 } 542 543 if (bio && (off || block < blocks_submitted + max_blocks)) { 544 sector_t sector = dblock << sdp->sd_fsb2bb_shift; 545 546 if (bio_end_sector(bio) == sector) { 547 sz = bio_add_page(bio, page, bsize, off); 548 if (sz == bsize) 549 goto block_added; 550 } 551 if (off) { 552 unsigned int blocks = 553 (PAGE_SIZE - off) >> bsize_shift; 554 555 bio = gfs2_chain_bio(bio, blocks); 556 goto add_block_to_new_bio; 557 } 558 } 559 560 if (bio) { 561 blocks_submitted = block; 562 submit_bio(bio); 563 } 564 565 bio = gfs2_log_alloc_bio(sdp, dblock, gfs2_end_log_read); 566 bio->bi_opf = REQ_OP_READ; 567 add_block_to_new_bio: 568 sz = bio_add_page(bio, page, bsize, off); 569 BUG_ON(sz != bsize); 570 block_added: 571 off += bsize; 572 if (off == PAGE_SIZE) 573 page = NULL; 574 if (blocks_submitted <= blocks_read + max_blocks) { 575 /* Keep at least one bio in flight */ 576 continue; 577 } 578 579 gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done); 580 blocks_read += PAGE_SIZE >> bsize_shift; 581 if (done) 582 goto out; /* found */ 583 } 584 } 585 586 out: 587 if (bio) 588 submit_bio(bio); 589 while (blocks_read < block) { 590 gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done); 591 blocks_read += PAGE_SIZE >> bsize_shift; 592 } 593 594 if (!ret) 595 ret = filemap_check_wb_err(mapping, since); 596 597 if (!keep_cache) 598 truncate_inode_pages(mapping, 0); 599 600 return ret; 601 } 602 603 static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type, 604 u32 ld_length, u32 ld_data1) 605 { 606 struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO); 607 struct gfs2_log_descriptor *ld = page_address(page); 608 clear_page(ld); 609 ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC); 610 ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD); 611 ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD); 612 ld->ld_type = cpu_to_be32(ld_type); 613 ld->ld_length = cpu_to_be32(ld_length); 614 ld->ld_data1 = cpu_to_be32(ld_data1); 615 ld->ld_data2 = 0; 616 return page; 617 } 618 619 static void gfs2_check_magic(struct buffer_head *bh) 620 { 621 void *kaddr; 622 __be32 *ptr; 623 624 clear_buffer_escaped(bh); 625 kaddr = kmap_atomic(bh->b_page); 626 ptr = kaddr + bh_offset(bh); 627 if (*ptr == cpu_to_be32(GFS2_MAGIC)) 628 set_buffer_escaped(bh); 629 kunmap_atomic(kaddr); 630 } 631 632 static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b) 633 { 634 struct gfs2_bufdata *bda, *bdb; 635 636 bda = list_entry(a, struct gfs2_bufdata, bd_list); 637 bdb = list_entry(b, struct gfs2_bufdata, bd_list); 638 639 if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr) 640 return -1; 641 if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr) 642 return 1; 643 return 0; 644 } 645 646 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit, 647 unsigned int total, struct list_head *blist, 648 bool is_databuf) 649 { 650 struct gfs2_log_descriptor *ld; 651 struct gfs2_bufdata *bd1 = NULL, *bd2; 652 struct page *page; 653 unsigned int num; 654 unsigned n; 655 __be64 *ptr; 656 657 gfs2_log_lock(sdp); 658 list_sort(NULL, blist, blocknr_cmp); 659 bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list); 660 while(total) { 661 num = total; 662 if (total > limit) 663 num = limit; 664 gfs2_log_unlock(sdp); 665 page = gfs2_get_log_desc(sdp, 666 is_databuf ? GFS2_LOG_DESC_JDATA : 667 GFS2_LOG_DESC_METADATA, num + 1, num); 668 ld = page_address(page); 669 gfs2_log_lock(sdp); 670 ptr = (__be64 *)(ld + 1); 671 672 n = 0; 673 list_for_each_entry_continue(bd1, blist, bd_list) { 674 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr); 675 if (is_databuf) { 676 gfs2_check_magic(bd1->bd_bh); 677 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0); 678 } 679 if (++n >= num) 680 break; 681 } 682 683 gfs2_log_unlock(sdp); 684 gfs2_log_write_page(sdp, page); 685 gfs2_log_lock(sdp); 686 687 n = 0; 688 list_for_each_entry_continue(bd2, blist, bd_list) { 689 get_bh(bd2->bd_bh); 690 gfs2_log_unlock(sdp); 691 lock_buffer(bd2->bd_bh); 692 693 if (buffer_escaped(bd2->bd_bh)) { 694 void *kaddr; 695 page = mempool_alloc(gfs2_page_pool, GFP_NOIO); 696 ptr = page_address(page); 697 kaddr = kmap_atomic(bd2->bd_bh->b_page); 698 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh), 699 bd2->bd_bh->b_size); 700 kunmap_atomic(kaddr); 701 *(__be32 *)ptr = 0; 702 clear_buffer_escaped(bd2->bd_bh); 703 unlock_buffer(bd2->bd_bh); 704 brelse(bd2->bd_bh); 705 gfs2_log_write_page(sdp, page); 706 } else { 707 gfs2_log_write_bh(sdp, bd2->bd_bh); 708 } 709 gfs2_log_lock(sdp); 710 if (++n >= num) 711 break; 712 } 713 714 BUG_ON(total < num); 715 total -= num; 716 } 717 gfs2_log_unlock(sdp); 718 } 719 720 static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 721 { 722 unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */ 723 unsigned int nbuf; 724 if (tr == NULL) 725 return; 726 nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm; 727 gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0); 728 } 729 730 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 731 { 732 struct list_head *head; 733 struct gfs2_bufdata *bd; 734 735 if (tr == NULL) 736 return; 737 738 head = &tr->tr_buf; 739 while (!list_empty(head)) { 740 bd = list_first_entry(head, struct gfs2_bufdata, bd_list); 741 list_del_init(&bd->bd_list); 742 gfs2_unpin(sdp, bd->bd_bh, tr); 743 } 744 } 745 746 static void buf_lo_before_scan(struct gfs2_jdesc *jd, 747 struct gfs2_log_header_host *head, int pass) 748 { 749 if (pass != 0) 750 return; 751 752 jd->jd_found_blocks = 0; 753 jd->jd_replayed_blocks = 0; 754 } 755 756 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start, 757 struct gfs2_log_descriptor *ld, __be64 *ptr, 758 int pass) 759 { 760 struct gfs2_inode *ip = GFS2_I(jd->jd_inode); 761 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 762 struct gfs2_glock *gl = ip->i_gl; 763 unsigned int blks = be32_to_cpu(ld->ld_data1); 764 struct buffer_head *bh_log, *bh_ip; 765 u64 blkno; 766 int error = 0; 767 768 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA) 769 return 0; 770 771 gfs2_replay_incr_blk(jd, &start); 772 773 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) { 774 blkno = be64_to_cpu(*ptr++); 775 776 jd->jd_found_blocks++; 777 778 if (gfs2_revoke_check(jd, blkno, start)) 779 continue; 780 781 error = gfs2_replay_read_block(jd, start, &bh_log); 782 if (error) 783 return error; 784 785 bh_ip = gfs2_meta_new(gl, blkno); 786 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size); 787 788 if (gfs2_meta_check(sdp, bh_ip)) 789 error = -EIO; 790 else { 791 struct gfs2_meta_header *mh = 792 (struct gfs2_meta_header *)bh_ip->b_data; 793 794 if (mh->mh_type == cpu_to_be32(GFS2_METATYPE_RG)) { 795 struct gfs2_rgrpd *rgd; 796 797 rgd = gfs2_blk2rgrpd(sdp, blkno, false); 798 if (rgd && rgd->rd_addr == blkno && 799 rgd->rd_bits && rgd->rd_bits->bi_bh) { 800 fs_info(sdp, "Replaying 0x%llx but we " 801 "already have a bh!\n", 802 (unsigned long long)blkno); 803 fs_info(sdp, "busy:%d, pinned:%d\n", 804 buffer_busy(rgd->rd_bits->bi_bh) ? 1 : 0, 805 buffer_pinned(rgd->rd_bits->bi_bh)); 806 gfs2_dump_glock(NULL, rgd->rd_gl, true); 807 } 808 } 809 mark_buffer_dirty(bh_ip); 810 } 811 brelse(bh_log); 812 brelse(bh_ip); 813 814 if (error) 815 break; 816 817 jd->jd_replayed_blocks++; 818 } 819 820 return error; 821 } 822 823 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) 824 { 825 struct gfs2_inode *ip = GFS2_I(jd->jd_inode); 826 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 827 828 if (error) { 829 gfs2_inode_metasync(ip->i_gl); 830 return; 831 } 832 if (pass != 1) 833 return; 834 835 gfs2_inode_metasync(ip->i_gl); 836 837 fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n", 838 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks); 839 } 840 841 static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 842 { 843 struct gfs2_meta_header *mh; 844 unsigned int offset; 845 struct list_head *head = &sdp->sd_log_revokes; 846 struct gfs2_bufdata *bd; 847 struct page *page; 848 unsigned int length; 849 850 gfs2_write_revokes(sdp); 851 if (!sdp->sd_log_num_revoke) 852 return; 853 854 length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke); 855 page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke); 856 offset = sizeof(struct gfs2_log_descriptor); 857 858 list_for_each_entry(bd, head, bd_list) { 859 sdp->sd_log_num_revoke--; 860 861 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) { 862 863 gfs2_log_write_page(sdp, page); 864 page = mempool_alloc(gfs2_page_pool, GFP_NOIO); 865 mh = page_address(page); 866 clear_page(mh); 867 mh->mh_magic = cpu_to_be32(GFS2_MAGIC); 868 mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB); 869 mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB); 870 offset = sizeof(struct gfs2_meta_header); 871 } 872 873 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno); 874 offset += sizeof(u64); 875 } 876 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke); 877 878 gfs2_log_write_page(sdp, page); 879 } 880 881 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 882 { 883 struct list_head *head = &sdp->sd_log_revokes; 884 struct gfs2_bufdata *bd; 885 struct gfs2_glock *gl; 886 887 while (!list_empty(head)) { 888 bd = list_first_entry(head, struct gfs2_bufdata, bd_list); 889 list_del_init(&bd->bd_list); 890 gl = bd->bd_gl; 891 gfs2_glock_remove_revoke(gl); 892 kmem_cache_free(gfs2_bufdata_cachep, bd); 893 } 894 } 895 896 static void revoke_lo_before_scan(struct gfs2_jdesc *jd, 897 struct gfs2_log_header_host *head, int pass) 898 { 899 if (pass != 0) 900 return; 901 902 jd->jd_found_revokes = 0; 903 jd->jd_replay_tail = head->lh_tail; 904 } 905 906 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, u32 start, 907 struct gfs2_log_descriptor *ld, __be64 *ptr, 908 int pass) 909 { 910 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 911 unsigned int blks = be32_to_cpu(ld->ld_length); 912 unsigned int revokes = be32_to_cpu(ld->ld_data1); 913 struct buffer_head *bh; 914 unsigned int offset; 915 u64 blkno; 916 int first = 1; 917 int error; 918 919 if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE) 920 return 0; 921 922 offset = sizeof(struct gfs2_log_descriptor); 923 924 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) { 925 error = gfs2_replay_read_block(jd, start, &bh); 926 if (error) 927 return error; 928 929 if (!first) 930 gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB); 931 932 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) { 933 blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset)); 934 935 error = gfs2_revoke_add(jd, blkno, start); 936 if (error < 0) { 937 brelse(bh); 938 return error; 939 } 940 else if (error) 941 jd->jd_found_revokes++; 942 943 if (!--revokes) 944 break; 945 offset += sizeof(u64); 946 } 947 948 brelse(bh); 949 offset = sizeof(struct gfs2_meta_header); 950 first = 0; 951 } 952 953 return 0; 954 } 955 956 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) 957 { 958 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 959 960 if (error) { 961 gfs2_revoke_clean(jd); 962 return; 963 } 964 if (pass != 1) 965 return; 966 967 fs_info(sdp, "jid=%u: Found %u revoke tags\n", 968 jd->jd_jid, jd->jd_found_revokes); 969 970 gfs2_revoke_clean(jd); 971 } 972 973 /** 974 * databuf_lo_before_commit - Scan the data buffers, writing as we go 975 * 976 */ 977 978 static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 979 { 980 unsigned int limit = databuf_limit(sdp); 981 unsigned int nbuf; 982 if (tr == NULL) 983 return; 984 nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm; 985 gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1); 986 } 987 988 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start, 989 struct gfs2_log_descriptor *ld, 990 __be64 *ptr, int pass) 991 { 992 struct gfs2_inode *ip = GFS2_I(jd->jd_inode); 993 struct gfs2_glock *gl = ip->i_gl; 994 unsigned int blks = be32_to_cpu(ld->ld_data1); 995 struct buffer_head *bh_log, *bh_ip; 996 u64 blkno; 997 u64 esc; 998 int error = 0; 999 1000 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA) 1001 return 0; 1002 1003 gfs2_replay_incr_blk(jd, &start); 1004 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) { 1005 blkno = be64_to_cpu(*ptr++); 1006 esc = be64_to_cpu(*ptr++); 1007 1008 jd->jd_found_blocks++; 1009 1010 if (gfs2_revoke_check(jd, blkno, start)) 1011 continue; 1012 1013 error = gfs2_replay_read_block(jd, start, &bh_log); 1014 if (error) 1015 return error; 1016 1017 bh_ip = gfs2_meta_new(gl, blkno); 1018 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size); 1019 1020 /* Unescape */ 1021 if (esc) { 1022 __be32 *eptr = (__be32 *)bh_ip->b_data; 1023 *eptr = cpu_to_be32(GFS2_MAGIC); 1024 } 1025 mark_buffer_dirty(bh_ip); 1026 1027 brelse(bh_log); 1028 brelse(bh_ip); 1029 1030 jd->jd_replayed_blocks++; 1031 } 1032 1033 return error; 1034 } 1035 1036 /* FIXME: sort out accounting for log blocks etc. */ 1037 1038 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) 1039 { 1040 struct gfs2_inode *ip = GFS2_I(jd->jd_inode); 1041 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 1042 1043 if (error) { 1044 gfs2_inode_metasync(ip->i_gl); 1045 return; 1046 } 1047 if (pass != 1) 1048 return; 1049 1050 /* data sync? */ 1051 gfs2_inode_metasync(ip->i_gl); 1052 1053 fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n", 1054 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks); 1055 } 1056 1057 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 1058 { 1059 struct list_head *head; 1060 struct gfs2_bufdata *bd; 1061 1062 if (tr == NULL) 1063 return; 1064 1065 head = &tr->tr_databuf; 1066 while (!list_empty(head)) { 1067 bd = list_first_entry(head, struct gfs2_bufdata, bd_list); 1068 list_del_init(&bd->bd_list); 1069 gfs2_unpin(sdp, bd->bd_bh, tr); 1070 } 1071 } 1072 1073 1074 static const struct gfs2_log_operations gfs2_buf_lops = { 1075 .lo_before_commit = buf_lo_before_commit, 1076 .lo_after_commit = buf_lo_after_commit, 1077 .lo_before_scan = buf_lo_before_scan, 1078 .lo_scan_elements = buf_lo_scan_elements, 1079 .lo_after_scan = buf_lo_after_scan, 1080 .lo_name = "buf", 1081 }; 1082 1083 static const struct gfs2_log_operations gfs2_revoke_lops = { 1084 .lo_before_commit = revoke_lo_before_commit, 1085 .lo_after_commit = revoke_lo_after_commit, 1086 .lo_before_scan = revoke_lo_before_scan, 1087 .lo_scan_elements = revoke_lo_scan_elements, 1088 .lo_after_scan = revoke_lo_after_scan, 1089 .lo_name = "revoke", 1090 }; 1091 1092 static const struct gfs2_log_operations gfs2_databuf_lops = { 1093 .lo_before_commit = databuf_lo_before_commit, 1094 .lo_after_commit = databuf_lo_after_commit, 1095 .lo_scan_elements = databuf_lo_scan_elements, 1096 .lo_after_scan = databuf_lo_after_scan, 1097 .lo_name = "databuf", 1098 }; 1099 1100 const struct gfs2_log_operations *gfs2_log_ops[] = { 1101 &gfs2_databuf_lops, 1102 &gfs2_buf_lops, 1103 &gfs2_revoke_lops, 1104 NULL, 1105 }; 1106 1107