1 /* 2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. 4 * 5 * This copyrighted material is made available to anyone wishing to use, 6 * modify, copy, or redistribute it subject to the terms and conditions 7 * of the GNU General Public License version 2. 8 */ 9 10 #include <linux/sched.h> 11 #include <linux/slab.h> 12 #include <linux/spinlock.h> 13 #include <linux/completion.h> 14 #include <linux/buffer_head.h> 15 #include <linux/mempool.h> 16 #include <linux/gfs2_ondisk.h> 17 #include <linux/bio.h> 18 #include <linux/fs.h> 19 #include <linux/list_sort.h> 20 #include <linux/blkdev.h> 21 22 #include "bmap.h" 23 #include "dir.h" 24 #include "gfs2.h" 25 #include "incore.h" 26 #include "inode.h" 27 #include "glock.h" 28 #include "log.h" 29 #include "lops.h" 30 #include "meta_io.h" 31 #include "recovery.h" 32 #include "rgrp.h" 33 #include "trans.h" 34 #include "util.h" 35 #include "trace_gfs2.h" 36 37 /** 38 * gfs2_pin - Pin a buffer in memory 39 * @sdp: The superblock 40 * @bh: The buffer to be pinned 41 * 42 * The log lock must be held when calling this function 43 */ 44 void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh) 45 { 46 struct gfs2_bufdata *bd; 47 48 BUG_ON(!current->journal_info); 49 50 clear_buffer_dirty(bh); 51 if (test_set_buffer_pinned(bh)) 52 gfs2_assert_withdraw(sdp, 0); 53 if (!buffer_uptodate(bh)) 54 gfs2_io_error_bh_wd(sdp, bh); 55 bd = bh->b_private; 56 /* If this buffer is in the AIL and it has already been written 57 * to in-place disk block, remove it from the AIL. 58 */ 59 spin_lock(&sdp->sd_ail_lock); 60 if (bd->bd_tr) 61 list_move(&bd->bd_ail_st_list, &bd->bd_tr->tr_ail2_list); 62 spin_unlock(&sdp->sd_ail_lock); 63 get_bh(bh); 64 atomic_inc(&sdp->sd_log_pinned); 65 trace_gfs2_pin(bd, 1); 66 } 67 68 static bool buffer_is_rgrp(const struct gfs2_bufdata *bd) 69 { 70 return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP; 71 } 72 73 static void maybe_release_space(struct gfs2_bufdata *bd) 74 { 75 struct gfs2_glock *gl = bd->bd_gl; 76 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; 77 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl); 78 unsigned int index = bd->bd_bh->b_blocknr - gl->gl_name.ln_number; 79 struct gfs2_bitmap *bi = rgd->rd_bits + index; 80 81 if (bi->bi_clone == NULL) 82 return; 83 if (sdp->sd_args.ar_discard) 84 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL); 85 memcpy(bi->bi_clone + bi->bi_offset, 86 bd->bd_bh->b_data + bi->bi_offset, bi->bi_bytes); 87 clear_bit(GBF_FULL, &bi->bi_flags); 88 rgd->rd_free_clone = rgd->rd_free; 89 rgd->rd_extfail_pt = rgd->rd_free; 90 } 91 92 /** 93 * gfs2_unpin - Unpin a buffer 94 * @sdp: the filesystem the buffer belongs to 95 * @bh: The buffer to unpin 96 * @ai: 97 * @flags: The inode dirty flags 98 * 99 */ 100 101 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh, 102 struct gfs2_trans *tr) 103 { 104 struct gfs2_bufdata *bd = bh->b_private; 105 106 BUG_ON(!buffer_uptodate(bh)); 107 BUG_ON(!buffer_pinned(bh)); 108 109 lock_buffer(bh); 110 mark_buffer_dirty(bh); 111 clear_buffer_pinned(bh); 112 113 if (buffer_is_rgrp(bd)) 114 maybe_release_space(bd); 115 116 spin_lock(&sdp->sd_ail_lock); 117 if (bd->bd_tr) { 118 list_del(&bd->bd_ail_st_list); 119 brelse(bh); 120 } else { 121 struct gfs2_glock *gl = bd->bd_gl; 122 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list); 123 atomic_inc(&gl->gl_ail_count); 124 } 125 bd->bd_tr = tr; 126 list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list); 127 spin_unlock(&sdp->sd_ail_lock); 128 129 clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags); 130 trace_gfs2_pin(bd, 0); 131 unlock_buffer(bh); 132 atomic_dec(&sdp->sd_log_pinned); 133 } 134 135 static void gfs2_log_incr_head(struct gfs2_sbd *sdp) 136 { 137 BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) && 138 (sdp->sd_log_flush_head != sdp->sd_log_head)); 139 140 if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) 141 sdp->sd_log_flush_head = 0; 142 } 143 144 u64 gfs2_log_bmap(struct gfs2_sbd *sdp) 145 { 146 unsigned int lbn = sdp->sd_log_flush_head; 147 struct gfs2_journal_extent *je; 148 u64 block; 149 150 list_for_each_entry(je, &sdp->sd_jdesc->extent_list, list) { 151 if ((lbn >= je->lblock) && (lbn < (je->lblock + je->blocks))) { 152 block = je->dblock + lbn - je->lblock; 153 gfs2_log_incr_head(sdp); 154 return block; 155 } 156 } 157 158 return -1; 159 } 160 161 /** 162 * gfs2_end_log_write_bh - end log write of pagecache data with buffers 163 * @sdp: The superblock 164 * @bvec: The bio_vec 165 * @error: The i/o status 166 * 167 * This finds the relevant buffers and unlocks them and sets the 168 * error flag according to the status of the i/o request. This is 169 * used when the log is writing data which has an in-place version 170 * that is pinned in the pagecache. 171 */ 172 173 static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp, 174 struct bio_vec *bvec, 175 blk_status_t error) 176 { 177 struct buffer_head *bh, *next; 178 struct page *page = bvec->bv_page; 179 unsigned size; 180 181 bh = page_buffers(page); 182 size = bvec->bv_len; 183 while (bh_offset(bh) < bvec->bv_offset) 184 bh = bh->b_this_page; 185 do { 186 if (error) 187 mark_buffer_write_io_error(bh); 188 unlock_buffer(bh); 189 next = bh->b_this_page; 190 size -= bh->b_size; 191 brelse(bh); 192 bh = next; 193 } while(bh && size); 194 } 195 196 /** 197 * gfs2_end_log_write - end of i/o to the log 198 * @bio: The bio 199 * 200 * Each bio_vec contains either data from the pagecache or data 201 * relating to the log itself. Here we iterate over the bio_vec 202 * array, processing both kinds of data. 203 * 204 */ 205 206 static void gfs2_end_log_write(struct bio *bio) 207 { 208 struct gfs2_sbd *sdp = bio->bi_private; 209 struct bio_vec *bvec; 210 struct page *page; 211 int i; 212 struct bvec_iter_all iter_all; 213 214 if (bio->bi_status) { 215 fs_err(sdp, "Error %d writing to journal, jid=%u\n", 216 bio->bi_status, sdp->sd_jdesc->jd_jid); 217 wake_up(&sdp->sd_logd_waitq); 218 } 219 220 bio_for_each_segment_all(bvec, bio, i, iter_all) { 221 page = bvec->bv_page; 222 if (page_has_buffers(page)) 223 gfs2_end_log_write_bh(sdp, bvec, bio->bi_status); 224 else 225 mempool_free(page, gfs2_page_pool); 226 } 227 228 bio_put(bio); 229 if (atomic_dec_and_test(&sdp->sd_log_in_flight)) 230 wake_up(&sdp->sd_log_flush_wait); 231 } 232 233 /** 234 * gfs2_log_submit_bio - Submit any pending log bio 235 * @biop: Address of the bio pointer 236 * @opf: REQ_OP | op_flags 237 * 238 * Submit any pending part-built or full bio to the block device. If 239 * there is no pending bio, then this is a no-op. 240 */ 241 242 void gfs2_log_submit_bio(struct bio **biop, int opf) 243 { 244 struct bio *bio = *biop; 245 if (bio) { 246 struct gfs2_sbd *sdp = bio->bi_private; 247 atomic_inc(&sdp->sd_log_in_flight); 248 bio->bi_opf = opf; 249 submit_bio(bio); 250 *biop = NULL; 251 } 252 } 253 254 /** 255 * gfs2_log_alloc_bio - Allocate a bio 256 * @sdp: The super block 257 * @blkno: The device block number we want to write to 258 * @end_io: The bi_end_io callback 259 * 260 * Allocate a new bio, initialize it with the given parameters and return it. 261 * 262 * Returns: The newly allocated bio 263 */ 264 265 static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno, 266 bio_end_io_t *end_io) 267 { 268 struct super_block *sb = sdp->sd_vfs; 269 struct bio *bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES); 270 271 bio->bi_iter.bi_sector = blkno * (sb->s_blocksize >> 9); 272 bio_set_dev(bio, sb->s_bdev); 273 bio->bi_end_io = end_io; 274 bio->bi_private = sdp; 275 276 return bio; 277 } 278 279 /** 280 * gfs2_log_get_bio - Get cached log bio, or allocate a new one 281 * @sdp: The super block 282 * @blkno: The device block number we want to write to 283 * @bio: The bio to get or allocate 284 * @op: REQ_OP 285 * @end_io: The bi_end_io callback 286 * @flush: Always flush the current bio and allocate a new one? 287 * 288 * If there is a cached bio, then if the next block number is sequential 289 * with the previous one, return it, otherwise flush the bio to the 290 * device. If there is no cached bio, or we just flushed it, then 291 * allocate a new one. 292 * 293 * Returns: The bio to use for log writes 294 */ 295 296 static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno, 297 struct bio **biop, int op, 298 bio_end_io_t *end_io, bool flush) 299 { 300 struct bio *bio = *biop; 301 302 if (bio) { 303 u64 nblk; 304 305 nblk = bio_end_sector(bio); 306 nblk >>= sdp->sd_fsb2bb_shift; 307 if (blkno == nblk && !flush) 308 return bio; 309 gfs2_log_submit_bio(biop, op); 310 } 311 312 *biop = gfs2_log_alloc_bio(sdp, blkno, end_io); 313 return *biop; 314 } 315 316 /** 317 * gfs2_log_write - write to log 318 * @sdp: the filesystem 319 * @page: the page to write 320 * @size: the size of the data to write 321 * @offset: the offset within the page 322 * @blkno: block number of the log entry 323 * 324 * Try and add the page segment to the current bio. If that fails, 325 * submit the current bio to the device and create a new one, and 326 * then add the page segment to that. 327 */ 328 329 void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page, 330 unsigned size, unsigned offset, u64 blkno) 331 { 332 struct bio *bio; 333 int ret; 334 335 bio = gfs2_log_get_bio(sdp, blkno, &sdp->sd_log_bio, REQ_OP_WRITE, 336 gfs2_end_log_write, false); 337 ret = bio_add_page(bio, page, size, offset); 338 if (ret == 0) { 339 bio = gfs2_log_get_bio(sdp, blkno, &sdp->sd_log_bio, 340 REQ_OP_WRITE, gfs2_end_log_write, true); 341 ret = bio_add_page(bio, page, size, offset); 342 WARN_ON(ret == 0); 343 } 344 } 345 346 /** 347 * gfs2_log_write_bh - write a buffer's content to the log 348 * @sdp: The super block 349 * @bh: The buffer pointing to the in-place location 350 * 351 * This writes the content of the buffer to the next available location 352 * in the log. The buffer will be unlocked once the i/o to the log has 353 * completed. 354 */ 355 356 static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh) 357 { 358 gfs2_log_write(sdp, bh->b_page, bh->b_size, bh_offset(bh), 359 gfs2_log_bmap(sdp)); 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 gfs2_log_write(sdp, page, sb->s_blocksize, 0, 377 gfs2_log_bmap(sdp)); 378 } 379 380 /** 381 * gfs2_end_log_read - end I/O callback for reads from the log 382 * @bio: The bio 383 * 384 * Simply unlock the pages in the bio. The main thread will wait on them and 385 * process them in order as necessary. 386 */ 387 388 static void gfs2_end_log_read(struct bio *bio) 389 { 390 struct page *page; 391 struct bio_vec *bvec; 392 int i; 393 struct bvec_iter_all iter_all; 394 395 bio_for_each_segment_all(bvec, bio, i, iter_all) { 396 page = bvec->bv_page; 397 if (bio->bi_status) { 398 int err = blk_status_to_errno(bio->bi_status); 399 400 SetPageError(page); 401 mapping_set_error(page->mapping, err); 402 } 403 unlock_page(page); 404 } 405 406 bio_put(bio); 407 } 408 409 /** 410 * gfs2_jhead_pg_srch - Look for the journal head in a given page. 411 * @jd: The journal descriptor 412 * @page: The page to look in 413 * 414 * Returns: 1 if found, 0 otherwise. 415 */ 416 417 static bool gfs2_jhead_pg_srch(struct gfs2_jdesc *jd, 418 struct gfs2_log_header_host *head, 419 struct page *page) 420 { 421 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 422 struct gfs2_log_header_host uninitialized_var(lh); 423 void *kaddr = kmap_atomic(page); 424 unsigned int offset; 425 bool ret = false; 426 427 for (offset = 0; offset < PAGE_SIZE; offset += sdp->sd_sb.sb_bsize) { 428 if (!__get_log_header(sdp, kaddr + offset, 0, &lh)) { 429 if (lh.lh_sequence > head->lh_sequence) 430 *head = lh; 431 else { 432 ret = true; 433 break; 434 } 435 } 436 } 437 kunmap_atomic(kaddr); 438 return ret; 439 } 440 441 /** 442 * gfs2_jhead_process_page - Search/cleanup a page 443 * @jd: The journal descriptor 444 * @index: Index of the page to look into 445 * @done: If set, perform only cleanup, else search and set if found. 446 * 447 * Find the page with 'index' in the journal's mapping. Search the page for 448 * the journal head if requested (cleanup == false). Release refs on the 449 * page so the page cache can reclaim it (put_page() twice). We grabbed a 450 * reference on this page two times, first when we did a find_or_create_page() 451 * to obtain the page to add it to the bio and second when we do a 452 * find_get_page() here to get the page to wait on while I/O on it is being 453 * completed. 454 * This function is also used to free up a page we might've grabbed but not 455 * used. Maybe we added it to a bio, but not submitted it for I/O. Or we 456 * submitted the I/O, but we already found the jhead so we only need to drop 457 * our references to the page. 458 */ 459 460 static void gfs2_jhead_process_page(struct gfs2_jdesc *jd, unsigned long index, 461 struct gfs2_log_header_host *head, 462 bool *done) 463 { 464 struct page *page; 465 466 page = find_get_page(jd->jd_inode->i_mapping, index); 467 wait_on_page_locked(page); 468 469 if (PageError(page)) 470 *done = true; 471 472 if (!*done) 473 *done = gfs2_jhead_pg_srch(jd, head, page); 474 475 put_page(page); /* Once for find_get_page */ 476 put_page(page); /* Once more for find_or_create_page */ 477 } 478 479 /** 480 * gfs2_find_jhead - find the head of a log 481 * @jd: The journal descriptor 482 * @head: The log descriptor for the head of the log is returned here 483 * 484 * Do a search of a journal by reading it in large chunks using bios and find 485 * the valid log entry with the highest sequence number. (i.e. the log head) 486 * 487 * Returns: 0 on success, errno otherwise 488 */ 489 490 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head) 491 { 492 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 493 struct address_space *mapping = jd->jd_inode->i_mapping; 494 struct gfs2_journal_extent *je; 495 u32 block, read_idx = 0, submit_idx = 0, index = 0; 496 int shift = PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift; 497 int blocks_per_page = 1 << shift, sz, ret = 0; 498 struct bio *bio = NULL; 499 struct page *page; 500 bool done = false; 501 errseq_t since; 502 503 memset(head, 0, sizeof(*head)); 504 if (list_empty(&jd->extent_list)) 505 gfs2_map_journal_extents(sdp, jd); 506 507 since = filemap_sample_wb_err(mapping); 508 list_for_each_entry(je, &jd->extent_list, list) { 509 for (block = 0; block < je->blocks; block += blocks_per_page) { 510 index = (je->lblock + block) >> shift; 511 512 page = find_or_create_page(mapping, index, GFP_NOFS); 513 if (!page) { 514 ret = -ENOMEM; 515 done = true; 516 goto out; 517 } 518 519 if (bio) { 520 sz = bio_add_page(bio, page, PAGE_SIZE, 0); 521 if (sz == PAGE_SIZE) 522 goto page_added; 523 submit_idx = index; 524 submit_bio(bio); 525 bio = NULL; 526 } 527 528 bio = gfs2_log_alloc_bio(sdp, 529 je->dblock + (index << shift), 530 gfs2_end_log_read); 531 bio->bi_opf = REQ_OP_READ; 532 sz = bio_add_page(bio, page, PAGE_SIZE, 0); 533 gfs2_assert_warn(sdp, sz == PAGE_SIZE); 534 535 page_added: 536 if (submit_idx <= read_idx + BIO_MAX_PAGES) { 537 /* Keep at least one bio in flight */ 538 continue; 539 } 540 541 gfs2_jhead_process_page(jd, read_idx++, head, &done); 542 if (done) 543 goto out; /* found */ 544 } 545 } 546 547 out: 548 if (bio) 549 submit_bio(bio); 550 while (read_idx <= index) 551 gfs2_jhead_process_page(jd, read_idx++, head, &done); 552 553 if (!ret) 554 ret = filemap_check_wb_err(mapping, since); 555 556 return ret; 557 } 558 559 static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type, 560 u32 ld_length, u32 ld_data1) 561 { 562 struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO); 563 struct gfs2_log_descriptor *ld = page_address(page); 564 clear_page(ld); 565 ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC); 566 ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD); 567 ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD); 568 ld->ld_type = cpu_to_be32(ld_type); 569 ld->ld_length = cpu_to_be32(ld_length); 570 ld->ld_data1 = cpu_to_be32(ld_data1); 571 ld->ld_data2 = 0; 572 return page; 573 } 574 575 static void gfs2_check_magic(struct buffer_head *bh) 576 { 577 void *kaddr; 578 __be32 *ptr; 579 580 clear_buffer_escaped(bh); 581 kaddr = kmap_atomic(bh->b_page); 582 ptr = kaddr + bh_offset(bh); 583 if (*ptr == cpu_to_be32(GFS2_MAGIC)) 584 set_buffer_escaped(bh); 585 kunmap_atomic(kaddr); 586 } 587 588 static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b) 589 { 590 struct gfs2_bufdata *bda, *bdb; 591 592 bda = list_entry(a, struct gfs2_bufdata, bd_list); 593 bdb = list_entry(b, struct gfs2_bufdata, bd_list); 594 595 if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr) 596 return -1; 597 if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr) 598 return 1; 599 return 0; 600 } 601 602 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit, 603 unsigned int total, struct list_head *blist, 604 bool is_databuf) 605 { 606 struct gfs2_log_descriptor *ld; 607 struct gfs2_bufdata *bd1 = NULL, *bd2; 608 struct page *page; 609 unsigned int num; 610 unsigned n; 611 __be64 *ptr; 612 613 gfs2_log_lock(sdp); 614 list_sort(NULL, blist, blocknr_cmp); 615 bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list); 616 while(total) { 617 num = total; 618 if (total > limit) 619 num = limit; 620 gfs2_log_unlock(sdp); 621 page = gfs2_get_log_desc(sdp, 622 is_databuf ? GFS2_LOG_DESC_JDATA : 623 GFS2_LOG_DESC_METADATA, num + 1, num); 624 ld = page_address(page); 625 gfs2_log_lock(sdp); 626 ptr = (__be64 *)(ld + 1); 627 628 n = 0; 629 list_for_each_entry_continue(bd1, blist, bd_list) { 630 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr); 631 if (is_databuf) { 632 gfs2_check_magic(bd1->bd_bh); 633 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0); 634 } 635 if (++n >= num) 636 break; 637 } 638 639 gfs2_log_unlock(sdp); 640 gfs2_log_write_page(sdp, page); 641 gfs2_log_lock(sdp); 642 643 n = 0; 644 list_for_each_entry_continue(bd2, blist, bd_list) { 645 get_bh(bd2->bd_bh); 646 gfs2_log_unlock(sdp); 647 lock_buffer(bd2->bd_bh); 648 649 if (buffer_escaped(bd2->bd_bh)) { 650 void *kaddr; 651 page = mempool_alloc(gfs2_page_pool, GFP_NOIO); 652 ptr = page_address(page); 653 kaddr = kmap_atomic(bd2->bd_bh->b_page); 654 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh), 655 bd2->bd_bh->b_size); 656 kunmap_atomic(kaddr); 657 *(__be32 *)ptr = 0; 658 clear_buffer_escaped(bd2->bd_bh); 659 unlock_buffer(bd2->bd_bh); 660 brelse(bd2->bd_bh); 661 gfs2_log_write_page(sdp, page); 662 } else { 663 gfs2_log_write_bh(sdp, bd2->bd_bh); 664 } 665 gfs2_log_lock(sdp); 666 if (++n >= num) 667 break; 668 } 669 670 BUG_ON(total < num); 671 total -= num; 672 } 673 gfs2_log_unlock(sdp); 674 } 675 676 static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 677 { 678 unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */ 679 unsigned int nbuf; 680 if (tr == NULL) 681 return; 682 nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm; 683 gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0); 684 } 685 686 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 687 { 688 struct list_head *head; 689 struct gfs2_bufdata *bd; 690 691 if (tr == NULL) 692 return; 693 694 head = &tr->tr_buf; 695 while (!list_empty(head)) { 696 bd = list_entry(head->next, struct gfs2_bufdata, bd_list); 697 list_del_init(&bd->bd_list); 698 gfs2_unpin(sdp, bd->bd_bh, tr); 699 } 700 } 701 702 static void buf_lo_before_scan(struct gfs2_jdesc *jd, 703 struct gfs2_log_header_host *head, int pass) 704 { 705 if (pass != 0) 706 return; 707 708 jd->jd_found_blocks = 0; 709 jd->jd_replayed_blocks = 0; 710 } 711 712 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start, 713 struct gfs2_log_descriptor *ld, __be64 *ptr, 714 int pass) 715 { 716 struct gfs2_inode *ip = GFS2_I(jd->jd_inode); 717 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 718 struct gfs2_glock *gl = ip->i_gl; 719 unsigned int blks = be32_to_cpu(ld->ld_data1); 720 struct buffer_head *bh_log, *bh_ip; 721 u64 blkno; 722 int error = 0; 723 724 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA) 725 return 0; 726 727 gfs2_replay_incr_blk(jd, &start); 728 729 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) { 730 blkno = be64_to_cpu(*ptr++); 731 732 jd->jd_found_blocks++; 733 734 if (gfs2_revoke_check(jd, blkno, start)) 735 continue; 736 737 error = gfs2_replay_read_block(jd, start, &bh_log); 738 if (error) 739 return error; 740 741 bh_ip = gfs2_meta_new(gl, blkno); 742 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size); 743 744 if (gfs2_meta_check(sdp, bh_ip)) 745 error = -EIO; 746 else 747 mark_buffer_dirty(bh_ip); 748 749 brelse(bh_log); 750 brelse(bh_ip); 751 752 if (error) 753 break; 754 755 jd->jd_replayed_blocks++; 756 } 757 758 return error; 759 } 760 761 /** 762 * gfs2_meta_sync - Sync all buffers associated with a glock 763 * @gl: The glock 764 * 765 */ 766 767 static void gfs2_meta_sync(struct gfs2_glock *gl) 768 { 769 struct address_space *mapping = gfs2_glock2aspace(gl); 770 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; 771 int error; 772 773 if (mapping == NULL) 774 mapping = &sdp->sd_aspace; 775 776 filemap_fdatawrite(mapping); 777 error = filemap_fdatawait(mapping); 778 779 if (error) 780 gfs2_io_error(gl->gl_name.ln_sbd); 781 } 782 783 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) 784 { 785 struct gfs2_inode *ip = GFS2_I(jd->jd_inode); 786 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 787 788 if (error) { 789 gfs2_meta_sync(ip->i_gl); 790 return; 791 } 792 if (pass != 1) 793 return; 794 795 gfs2_meta_sync(ip->i_gl); 796 797 fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n", 798 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks); 799 } 800 801 static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 802 { 803 struct gfs2_meta_header *mh; 804 unsigned int offset; 805 struct list_head *head = &sdp->sd_log_le_revoke; 806 struct gfs2_bufdata *bd; 807 struct page *page; 808 unsigned int length; 809 810 gfs2_write_revokes(sdp); 811 if (!sdp->sd_log_num_revoke) 812 return; 813 814 length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64)); 815 page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke); 816 offset = sizeof(struct gfs2_log_descriptor); 817 818 list_for_each_entry(bd, head, bd_list) { 819 sdp->sd_log_num_revoke--; 820 821 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) { 822 823 gfs2_log_write_page(sdp, page); 824 page = mempool_alloc(gfs2_page_pool, GFP_NOIO); 825 mh = page_address(page); 826 clear_page(mh); 827 mh->mh_magic = cpu_to_be32(GFS2_MAGIC); 828 mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB); 829 mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB); 830 offset = sizeof(struct gfs2_meta_header); 831 } 832 833 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno); 834 offset += sizeof(u64); 835 } 836 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke); 837 838 gfs2_log_write_page(sdp, page); 839 } 840 841 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 842 { 843 struct list_head *head = &sdp->sd_log_le_revoke; 844 struct gfs2_bufdata *bd; 845 struct gfs2_glock *gl; 846 847 while (!list_empty(head)) { 848 bd = list_entry(head->next, struct gfs2_bufdata, bd_list); 849 list_del_init(&bd->bd_list); 850 gl = bd->bd_gl; 851 atomic_dec(&gl->gl_revokes); 852 clear_bit(GLF_LFLUSH, &gl->gl_flags); 853 kmem_cache_free(gfs2_bufdata_cachep, bd); 854 } 855 } 856 857 static void revoke_lo_before_scan(struct gfs2_jdesc *jd, 858 struct gfs2_log_header_host *head, int pass) 859 { 860 if (pass != 0) 861 return; 862 863 jd->jd_found_revokes = 0; 864 jd->jd_replay_tail = head->lh_tail; 865 } 866 867 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start, 868 struct gfs2_log_descriptor *ld, __be64 *ptr, 869 int pass) 870 { 871 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 872 unsigned int blks = be32_to_cpu(ld->ld_length); 873 unsigned int revokes = be32_to_cpu(ld->ld_data1); 874 struct buffer_head *bh; 875 unsigned int offset; 876 u64 blkno; 877 int first = 1; 878 int error; 879 880 if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE) 881 return 0; 882 883 offset = sizeof(struct gfs2_log_descriptor); 884 885 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) { 886 error = gfs2_replay_read_block(jd, start, &bh); 887 if (error) 888 return error; 889 890 if (!first) 891 gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB); 892 893 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) { 894 blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset)); 895 896 error = gfs2_revoke_add(jd, blkno, start); 897 if (error < 0) { 898 brelse(bh); 899 return error; 900 } 901 else if (error) 902 jd->jd_found_revokes++; 903 904 if (!--revokes) 905 break; 906 offset += sizeof(u64); 907 } 908 909 brelse(bh); 910 offset = sizeof(struct gfs2_meta_header); 911 first = 0; 912 } 913 914 return 0; 915 } 916 917 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) 918 { 919 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 920 921 if (error) { 922 gfs2_revoke_clean(jd); 923 return; 924 } 925 if (pass != 1) 926 return; 927 928 fs_info(sdp, "jid=%u: Found %u revoke tags\n", 929 jd->jd_jid, jd->jd_found_revokes); 930 931 gfs2_revoke_clean(jd); 932 } 933 934 /** 935 * databuf_lo_before_commit - Scan the data buffers, writing as we go 936 * 937 */ 938 939 static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 940 { 941 unsigned int limit = databuf_limit(sdp); 942 unsigned int nbuf; 943 if (tr == NULL) 944 return; 945 nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm; 946 gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1); 947 } 948 949 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start, 950 struct gfs2_log_descriptor *ld, 951 __be64 *ptr, int pass) 952 { 953 struct gfs2_inode *ip = GFS2_I(jd->jd_inode); 954 struct gfs2_glock *gl = ip->i_gl; 955 unsigned int blks = be32_to_cpu(ld->ld_data1); 956 struct buffer_head *bh_log, *bh_ip; 957 u64 blkno; 958 u64 esc; 959 int error = 0; 960 961 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA) 962 return 0; 963 964 gfs2_replay_incr_blk(jd, &start); 965 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) { 966 blkno = be64_to_cpu(*ptr++); 967 esc = be64_to_cpu(*ptr++); 968 969 jd->jd_found_blocks++; 970 971 if (gfs2_revoke_check(jd, blkno, start)) 972 continue; 973 974 error = gfs2_replay_read_block(jd, start, &bh_log); 975 if (error) 976 return error; 977 978 bh_ip = gfs2_meta_new(gl, blkno); 979 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size); 980 981 /* Unescape */ 982 if (esc) { 983 __be32 *eptr = (__be32 *)bh_ip->b_data; 984 *eptr = cpu_to_be32(GFS2_MAGIC); 985 } 986 mark_buffer_dirty(bh_ip); 987 988 brelse(bh_log); 989 brelse(bh_ip); 990 991 jd->jd_replayed_blocks++; 992 } 993 994 return error; 995 } 996 997 /* FIXME: sort out accounting for log blocks etc. */ 998 999 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) 1000 { 1001 struct gfs2_inode *ip = GFS2_I(jd->jd_inode); 1002 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 1003 1004 if (error) { 1005 gfs2_meta_sync(ip->i_gl); 1006 return; 1007 } 1008 if (pass != 1) 1009 return; 1010 1011 /* data sync? */ 1012 gfs2_meta_sync(ip->i_gl); 1013 1014 fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n", 1015 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks); 1016 } 1017 1018 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) 1019 { 1020 struct list_head *head; 1021 struct gfs2_bufdata *bd; 1022 1023 if (tr == NULL) 1024 return; 1025 1026 head = &tr->tr_databuf; 1027 while (!list_empty(head)) { 1028 bd = list_entry(head->next, struct gfs2_bufdata, bd_list); 1029 list_del_init(&bd->bd_list); 1030 gfs2_unpin(sdp, bd->bd_bh, tr); 1031 } 1032 } 1033 1034 1035 const struct gfs2_log_operations gfs2_buf_lops = { 1036 .lo_before_commit = buf_lo_before_commit, 1037 .lo_after_commit = buf_lo_after_commit, 1038 .lo_before_scan = buf_lo_before_scan, 1039 .lo_scan_elements = buf_lo_scan_elements, 1040 .lo_after_scan = buf_lo_after_scan, 1041 .lo_name = "buf", 1042 }; 1043 1044 const struct gfs2_log_operations gfs2_revoke_lops = { 1045 .lo_before_commit = revoke_lo_before_commit, 1046 .lo_after_commit = revoke_lo_after_commit, 1047 .lo_before_scan = revoke_lo_before_scan, 1048 .lo_scan_elements = revoke_lo_scan_elements, 1049 .lo_after_scan = revoke_lo_after_scan, 1050 .lo_name = "revoke", 1051 }; 1052 1053 const struct gfs2_log_operations gfs2_databuf_lops = { 1054 .lo_before_commit = databuf_lo_before_commit, 1055 .lo_after_commit = databuf_lo_after_commit, 1056 .lo_scan_elements = databuf_lo_scan_elements, 1057 .lo_after_scan = databuf_lo_after_scan, 1058 .lo_name = "databuf", 1059 }; 1060 1061 const struct gfs2_log_operations *gfs2_log_ops[] = { 1062 &gfs2_databuf_lops, 1063 &gfs2_buf_lops, 1064 &gfs2_revoke_lops, 1065 NULL, 1066 }; 1067 1068