Pileus Git - ~andy/linux/blob - fs/gfs2/lops.c

   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
  20 #include "gfs2.h"
  21 #include "incore.h"
  22 #include "inode.h"
  23 #include "glock.h"
  24 #include "log.h"
  25 #include "lops.h"
  26 #include "meta_io.h"
  27 #include "recovery.h"
  28 #include "rgrp.h"
  29 #include "trans.h"
  30 #include "util.h"
  31 #include "trace_gfs2.h"
  32
  33 /**
  34  * gfs2_pin - Pin a buffer in memory
  35  * @sdp: The superblock
  36  * @bh: The buffer to be pinned
  37  *
  38  * The log lock must be held when calling this function
  39  */
  40 static void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
  41 {
  42         struct gfs2_bufdata *bd;
  43
  44         BUG_ON(!current->journal_info);
  45
  46         clear_buffer_dirty(bh);
  47         if (test_set_buffer_pinned(bh))
  48                 gfs2_assert_withdraw(sdp, 0);
  49         if (!buffer_uptodate(bh))
  50                 gfs2_io_error_bh(sdp, bh);
  51         bd = bh->b_private;
  52         /* If this buffer is in the AIL and it has already been written
  53          * to in-place disk block, remove it from the AIL.
  54          */
  55         spin_lock(&sdp->sd_ail_lock);
  56         if (bd->bd_ail)
  57                 list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
  58         spin_unlock(&sdp->sd_ail_lock);
  59         get_bh(bh);
  60         atomic_inc(&sdp->sd_log_pinned);
  61         trace_gfs2_pin(bd, 1);
  62 }
  63
  64 static bool buffer_is_rgrp(const struct gfs2_bufdata *bd)
  65 {
  66         return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP;
  67 }
  68
  69 static void maybe_release_space(struct gfs2_bufdata *bd)
  70 {
  71         struct gfs2_glock *gl = bd->bd_gl;
  72         struct gfs2_sbd *sdp = gl->gl_sbd;
  73         struct gfs2_rgrpd *rgd = gl->gl_object;
  74         unsigned int index = bd->bd_bh->b_blocknr - gl->gl_name.ln_number;
  75         struct gfs2_bitmap *bi = rgd->rd_bits + index;
  76
  77         if (bi->bi_clone == 0)
  78                 return;
  79         if (sdp->sd_args.ar_discard)
  80                 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL);
  81         memcpy(bi->bi_clone + bi->bi_offset,
  82                bd->bd_bh->b_data + bi->bi_offset, bi->bi_len);
  83         clear_bit(GBF_FULL, &bi->bi_flags);
  84         rgd->rd_free_clone = rgd->rd_free;
  85 }
  86
  87 /**
  88  * gfs2_unpin - Unpin a buffer
  89  * @sdp: the filesystem the buffer belongs to
  90  * @bh: The buffer to unpin
  91  * @ai:
  92  * @flags: The inode dirty flags
  93  *
  94  */
  95
  96 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
  97                        struct gfs2_ail *ai)
  98 {
  99         struct gfs2_bufdata *bd = bh->b_private;
 100
 101         BUG_ON(!buffer_uptodate(bh));
 102         BUG_ON(!buffer_pinned(bh));
 103
 104         lock_buffer(bh);
 105         mark_buffer_dirty(bh);
 106         clear_buffer_pinned(bh);
 107
 108         if (buffer_is_rgrp(bd))
 109                 maybe_release_space(bd);
 110
 111         spin_lock(&sdp->sd_ail_lock);
 112         if (bd->bd_ail) {
 113                 list_del(&bd->bd_ail_st_list);
 114                 brelse(bh);
 115         } else {
 116                 struct gfs2_glock *gl = bd->bd_gl;
 117                 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
 118                 atomic_inc(&gl->gl_ail_count);
 119         }
 120         bd->bd_ail = ai;
 121         list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
 122         spin_unlock(&sdp->sd_ail_lock);
 123
 124         clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 125         trace_gfs2_pin(bd, 0);
 126         unlock_buffer(bh);
 127         atomic_dec(&sdp->sd_log_pinned);
 128 }
 129
 130
 131 static inline struct gfs2_log_descriptor *bh_log_desc(struct buffer_head *bh)
 132 {
 133         return (struct gfs2_log_descriptor *)bh->b_data;
 134 }
 135
 136 static inline __be64 *bh_log_ptr(struct buffer_head *bh)
 137 {
 138         struct gfs2_log_descriptor *ld = bh_log_desc(bh);
 139         return (__force __be64 *)(ld + 1);
 140 }
 141
 142 static inline __be64 *bh_ptr_end(struct buffer_head *bh)
 143 {
 144         return (__force __be64 *)(bh->b_data + bh->b_size);
 145 }
 146
 147 /**
 148  * gfs2_log_write_endio - End of I/O for a log buffer
 149  * @bh: The buffer head
 150  * @uptodate: I/O Status
 151  *
 152  */
 153
 154 static void gfs2_log_write_endio(struct buffer_head *bh, int uptodate)
 155 {
 156         struct gfs2_sbd *sdp = bh->b_private;
 157         bh->b_private = NULL;
 158
 159         end_buffer_write_sync(bh, uptodate);
 160         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 161                 wake_up(&sdp->sd_log_flush_wait);
 162 }
 163
 164 /**
 165  * gfs2_log_get_buf - Get and initialize a buffer to use for log control data
 166  * @sdp: The GFS2 superblock
 167  *
 168  * tReturns: the buffer_head
 169  */
 170
 171 static struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp)
 172 {
 173         u64 blkno = gfs2_log_bmap(sdp, sdp->sd_log_flush_head);
 174         struct buffer_head *bh;
 175
 176         bh = sb_getblk(sdp->sd_vfs, blkno);
 177         lock_buffer(bh);
 178         memset(bh->b_data, 0, bh->b_size);
 179         set_buffer_uptodate(bh);
 180         clear_buffer_dirty(bh);
 181         gfs2_log_incr_head(sdp);
 182         atomic_inc(&sdp->sd_log_in_flight);
 183         bh->b_private = sdp;
 184         bh->b_end_io = gfs2_log_write_endio;
 185
 186         return bh;
 187 }
 188
 189 /**
 190  * gfs2_fake_write_endio -
 191  * @bh: The buffer head
 192  * @uptodate: The I/O Status
 193  *
 194  */
 195
 196 static void gfs2_fake_write_endio(struct buffer_head *bh, int uptodate)
 197 {
 198         struct buffer_head *real_bh = bh->b_private;
 199         struct gfs2_bufdata *bd = real_bh->b_private;
 200         struct gfs2_sbd *sdp = bd->bd_gl->gl_sbd;
 201
 202         end_buffer_write_sync(bh, uptodate);
 203         mempool_free(bh, gfs2_bh_pool);
 204         unlock_buffer(real_bh);
 205         brelse(real_bh);
 206         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 207                 wake_up(&sdp->sd_log_flush_wait);
 208 }
 209
 210 /**
 211  * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log
 212  * @sdp: the filesystem
 213  * @data: the data the buffer_head should point to
 214  *
 215  * Returns: the log buffer descriptor
 216  */
 217
 218 static struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp,
 219                                       struct buffer_head *real)
 220 {
 221         u64 blkno = gfs2_log_bmap(sdp, sdp->sd_log_flush_head);
 222         struct buffer_head *bh;
 223
 224         bh = mempool_alloc(gfs2_bh_pool, GFP_NOFS);
 225         atomic_set(&bh->b_count, 1);
 226         bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate) | (1 << BH_Lock);
 227         set_bh_page(bh, real->b_page, bh_offset(real));
 228         bh->b_blocknr = blkno;
 229         bh->b_size = sdp->sd_sb.sb_bsize;
 230         bh->b_bdev = sdp->sd_vfs->s_bdev;
 231         bh->b_private = real;
 232         bh->b_end_io = gfs2_fake_write_endio;
 233
 234         gfs2_log_incr_head(sdp);
 235         atomic_inc(&sdp->sd_log_in_flight);
 236
 237         return bh;
 238 }
 239
 240 static struct buffer_head *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type)
 241 {
 242         struct buffer_head *bh = gfs2_log_get_buf(sdp);
 243         struct gfs2_log_descriptor *ld = bh_log_desc(bh);
 244         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 245         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 246         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 247         ld->ld_type = cpu_to_be32(ld_type);
 248         ld->ld_length = 0;
 249         ld->ld_data1 = 0;
 250         ld->ld_data2 = 0;
 251         memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
 252         return bh;
 253 }
 254
 255 static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 256 {
 257         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
 258         struct gfs2_meta_header *mh;
 259         struct gfs2_trans *tr;
 260
 261         lock_buffer(bd->bd_bh);
 262         gfs2_log_lock(sdp);
 263         if (!list_empty(&bd->bd_list_tr))
 264                 goto out;
 265         tr = current->journal_info;
 266         tr->tr_touched = 1;
 267         tr->tr_num_buf++;
 268         list_add(&bd->bd_list_tr, &tr->tr_list_buf);
 269         if (!list_empty(&le->le_list))
 270                 goto out;
 271         set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 272         set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
 273         gfs2_meta_check(sdp, bd->bd_bh);
 274         gfs2_pin(sdp, bd->bd_bh);
 275         mh = (struct gfs2_meta_header *)bd->bd_bh->b_data;
 276         mh->__pad0 = cpu_to_be64(0);
 277         mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
 278         sdp->sd_log_num_buf++;
 279         list_add(&le->le_list, &sdp->sd_log_le_buf);
 280         tr->tr_num_buf_new++;
 281 out:
 282         gfs2_log_unlock(sdp);
 283         unlock_buffer(bd->bd_bh);
 284 }
 285
 286 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
 287 {
 288         struct buffer_head *bh;
 289         struct gfs2_log_descriptor *ld;
 290         struct gfs2_bufdata *bd1 = NULL, *bd2;
 291         unsigned int total;
 292         unsigned int limit;
 293         unsigned int num;
 294         unsigned n;
 295         __be64 *ptr;
 296
 297         limit = buf_limit(sdp);
 298         /* for 4k blocks, limit = 503 */
 299
 300         gfs2_log_lock(sdp);
 301         total = sdp->sd_log_num_buf;
 302         bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
 303         while(total) {
 304                 num = total;
 305                 if (total > limit)
 306                         num = limit;
 307                 gfs2_log_unlock(sdp);
 308                 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_METADATA);
 309                 gfs2_log_lock(sdp);
 310                 ld = bh_log_desc(bh);
 311                 ptr = bh_log_ptr(bh);
 312                 ld->ld_length = cpu_to_be32(num + 1);
 313                 ld->ld_data1 = cpu_to_be32(num);
 314
 315                 n = 0;
 316                 list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
 317                                              bd_le.le_list) {
 318                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 319                         if (++n >= num)
 320                                 break;
 321                 }
 322
 323                 gfs2_log_unlock(sdp);
 324                 submit_bh(WRITE_SYNC, bh);
 325                 gfs2_log_lock(sdp);
 326
 327                 n = 0;
 328                 list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
 329                                              bd_le.le_list) {
 330                         get_bh(bd2->bd_bh);
 331                         gfs2_log_unlock(sdp);
 332                         lock_buffer(bd2->bd_bh);
 333                         bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
 334                         submit_bh(WRITE_SYNC, bh);
 335                         gfs2_log_lock(sdp);
 336                         if (++n >= num)
 337                                 break;
 338                 }
 339
 340                 BUG_ON(total < num);
 341                 total -= num;
 342         }
 343         gfs2_log_unlock(sdp);
 344 }
 345
 346 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 347 {
 348         struct list_head *head = &sdp->sd_log_le_buf;
 349         struct gfs2_bufdata *bd;
 350
 351         while (!list_empty(head)) {
 352                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 353                 list_del_init(&bd->bd_le.le_list);
 354                 sdp->sd_log_num_buf--;
 355
 356                 gfs2_unpin(sdp, bd->bd_bh, ai);
 357         }
 358         gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
 359 }
 360
 361 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 362                                struct gfs2_log_header_host *head, int pass)
 363 {
 364         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 365
 366         if (pass != 0)
 367                 return;
 368
 369         sdp->sd_found_blocks = 0;
 370         sdp->sd_replayed_blocks = 0;
 371 }
 372
 373 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 374                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 375                                 int pass)
 376 {
 377         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 378         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 379         struct gfs2_glock *gl = ip->i_gl;
 380         unsigned int blks = be32_to_cpu(ld->ld_data1);
 381         struct buffer_head *bh_log, *bh_ip;
 382         u64 blkno;
 383         int error = 0;
 384
 385         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 386                 return 0;
 387
 388         gfs2_replay_incr_blk(sdp, &start);
 389
 390         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 391                 blkno = be64_to_cpu(*ptr++);
 392
 393                 sdp->sd_found_blocks++;
 394
 395                 if (gfs2_revoke_check(sdp, blkno, start))
 396                         continue;
 397
 398                 error = gfs2_replay_read_block(jd, start, &bh_log);
 399                 if (error)
 400                         return error;
 401
 402                 bh_ip = gfs2_meta_new(gl, blkno);
 403                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 404
 405                 if (gfs2_meta_check(sdp, bh_ip))
 406                         error = -EIO;
 407                 else
 408                         mark_buffer_dirty(bh_ip);
 409
 410                 brelse(bh_log);
 411                 brelse(bh_ip);
 412
 413                 if (error)
 414                         break;
 415
 416                 sdp->sd_replayed_blocks++;
 417         }
 418
 419         return error;
 420 }
 421
 422 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 423 {
 424         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 425         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 426
 427         if (error) {
 428                 gfs2_meta_sync(ip->i_gl);
 429                 return;
 430         }
 431         if (pass != 1)
 432                 return;
 433
 434         gfs2_meta_sync(ip->i_gl);
 435
 436         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 437                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 438 }
 439
 440 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 441 {
 442         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
 443         struct gfs2_glock *gl = bd->bd_gl;
 444         struct gfs2_trans *tr;
 445
 446         tr = current->journal_info;
 447         tr->tr_touched = 1;
 448         tr->tr_num_revoke++;
 449         sdp->sd_log_num_revoke++;
 450         atomic_inc(&gl->gl_revokes);
 451         set_bit(GLF_LFLUSH, &gl->gl_flags);
 452         list_add(&le->le_list, &sdp->sd_log_le_revoke);
 453 }
 454
 455 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
 456 {
 457         struct gfs2_log_descriptor *ld;
 458         struct gfs2_meta_header *mh;
 459         struct buffer_head *bh;
 460         unsigned int offset;
 461         struct list_head *head = &sdp->sd_log_le_revoke;
 462         struct gfs2_bufdata *bd;
 463
 464         if (!sdp->sd_log_num_revoke)
 465                 return;
 466
 467         bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE);
 468         ld = bh_log_desc(bh);
 469         ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
 470                                                     sizeof(u64)));
 471         ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
 472         offset = sizeof(struct gfs2_log_descriptor);
 473
 474         list_for_each_entry(bd, head, bd_le.le_list) {
 475                 sdp->sd_log_num_revoke--;
 476
 477                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 478                         submit_bh(WRITE_SYNC, bh);
 479
 480                         bh = gfs2_log_get_buf(sdp);
 481                         mh = (struct gfs2_meta_header *)bh->b_data;
 482                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 483                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 484                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 485                         offset = sizeof(struct gfs2_meta_header);
 486                 }
 487
 488                 *(__be64 *)(bh->b_data + offset) = cpu_to_be64(bd->bd_blkno);
 489                 offset += sizeof(u64);
 490         }
 491         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 492
 493         submit_bh(WRITE_SYNC, bh);
 494 }
 495
 496 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 497 {
 498         struct list_head *head = &sdp->sd_log_le_revoke;
 499         struct gfs2_bufdata *bd;
 500         struct gfs2_glock *gl;
 501
 502         while (!list_empty(head)) {
 503                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 504                 list_del_init(&bd->bd_le.le_list);
 505                 gl = bd->bd_gl;
 506                 atomic_dec(&gl->gl_revokes);
 507                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
 508                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 509         }
 510 }
 511
 512 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 513                                   struct gfs2_log_header_host *head, int pass)
 514 {
 515         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 516
 517         if (pass != 0)
 518                 return;
 519
 520         sdp->sd_found_revokes = 0;
 521         sdp->sd_replay_tail = head->lh_tail;
 522 }
 523
 524 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 525                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 526                                    int pass)
 527 {
 528         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 529         unsigned int blks = be32_to_cpu(ld->ld_length);
 530         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 531         struct buffer_head *bh;
 532         unsigned int offset;
 533         u64 blkno;
 534         int first = 1;
 535         int error;
 536
 537         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 538                 return 0;
 539
 540         offset = sizeof(struct gfs2_log_descriptor);
 541
 542         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 543                 error = gfs2_replay_read_block(jd, start, &bh);
 544                 if (error)
 545                         return error;
 546
 547                 if (!first)
 548                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 549
 550                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 551                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 552
 553                         error = gfs2_revoke_add(sdp, blkno, start);
 554                         if (error < 0) {
 555                                 brelse(bh);
 556                                 return error;
 557                         }
 558                         else if (error)
 559                                 sdp->sd_found_revokes++;
 560
 561                         if (!--revokes)
 562                                 break;
 563                         offset += sizeof(u64);
 564                 }
 565
 566                 brelse(bh);
 567                 offset = sizeof(struct gfs2_meta_header);
 568                 first = 0;
 569         }
 570
 571         return 0;
 572 }
 573
 574 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 575 {
 576         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 577
 578         if (error) {
 579                 gfs2_revoke_clean(sdp);
 580                 return;
 581         }
 582         if (pass != 1)
 583                 return;
 584
 585         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 586                 jd->jd_jid, sdp->sd_found_revokes);
 587
 588         gfs2_revoke_clean(sdp);
 589 }
 590
 591 /**
 592  * databuf_lo_add - Add a databuf to the transaction.
 593  *
 594  * This is used in two distinct cases:
 595  * i) In ordered write mode
 596  *    We put the data buffer on a list so that we can ensure that its
 597  *    synced to disk at the right time
 598  * ii) In journaled data mode
 599  *    We need to journal the data block in the same way as metadata in
 600  *    the functions above. The difference is that here we have a tag
 601  *    which is two __be64's being the block number (as per meta data)
 602  *    and a flag which says whether the data block needs escaping or
 603  *    not. This means we need a new log entry for each 251 or so data
 604  *    blocks, which isn't an enormous overhead but twice as much as
 605  *    for normal metadata blocks.
 606  */
 607 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 608 {
 609         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
 610         struct gfs2_trans *tr = current->journal_info;
 611         struct address_space *mapping = bd->bd_bh->b_page->mapping;
 612         struct gfs2_inode *ip = GFS2_I(mapping->host);
 613
 614         lock_buffer(bd->bd_bh);
 615         gfs2_log_lock(sdp);
 616         if (tr) {
 617                 if (!list_empty(&bd->bd_list_tr))
 618                         goto out;
 619                 tr->tr_touched = 1;
 620                 if (gfs2_is_jdata(ip)) {
 621                         tr->tr_num_buf++;
 622                         list_add(&bd->bd_list_tr, &tr->tr_list_buf);
 623                 }
 624         }
 625         if (!list_empty(&le->le_list))
 626                 goto out;
 627
 628         set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 629         set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
 630         if (gfs2_is_jdata(ip)) {
 631                 gfs2_pin(sdp, bd->bd_bh);
 632                 tr->tr_num_databuf_new++;
 633                 sdp->sd_log_num_databuf++;
 634                 list_add_tail(&le->le_list, &sdp->sd_log_le_databuf);
 635         } else {
 636                 list_add_tail(&le->le_list, &sdp->sd_log_le_ordered);
 637         }
 638 out:
 639         gfs2_log_unlock(sdp);
 640         unlock_buffer(bd->bd_bh);
 641 }
 642
 643 static void gfs2_check_magic(struct buffer_head *bh)
 644 {
 645         void *kaddr;
 646         __be32 *ptr;
 647
 648         clear_buffer_escaped(bh);
 649         kaddr = kmap_atomic(bh->b_page);
 650         ptr = kaddr + bh_offset(bh);
 651         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 652                 set_buffer_escaped(bh);
 653         kunmap_atomic(kaddr);
 654 }
 655
 656 static void gfs2_write_blocks(struct gfs2_sbd *sdp, struct buffer_head *bh,
 657                               struct list_head *list, struct list_head *done,
 658                               unsigned int n)
 659 {
 660         struct buffer_head *bh1;
 661         struct gfs2_log_descriptor *ld;
 662         struct gfs2_bufdata *bd;
 663         __be64 *ptr;
 664
 665         if (!bh)
 666                 return;
 667
 668         ld = bh_log_desc(bh);
 669         ld->ld_length = cpu_to_be32(n + 1);
 670         ld->ld_data1 = cpu_to_be32(n);
 671
 672         ptr = bh_log_ptr(bh);
 673
 674         get_bh(bh);
 675         submit_bh(WRITE_SYNC, bh);
 676         gfs2_log_lock(sdp);
 677         while(!list_empty(list)) {
 678                 bd = list_entry(list->next, struct gfs2_bufdata, bd_le.le_list);
 679                 list_move_tail(&bd->bd_le.le_list, done);
 680                 get_bh(bd->bd_bh);
 681                 while (be64_to_cpu(*ptr) != bd->bd_bh->b_blocknr) {
 682                         gfs2_log_incr_head(sdp);
 683                         ptr += 2;
 684                 }
 685                 gfs2_log_unlock(sdp);
 686                 lock_buffer(bd->bd_bh);
 687                 if (buffer_escaped(bd->bd_bh)) {
 688                         void *kaddr;
 689                         bh1 = gfs2_log_get_buf(sdp);
 690                         kaddr = kmap_atomic(bd->bd_bh->b_page);
 691                         memcpy(bh1->b_data, kaddr + bh_offset(bd->bd_bh),
 692                                bh1->b_size);
 693                         kunmap_atomic(kaddr);
 694                         *(__be32 *)bh1->b_data = 0;
 695                         clear_buffer_escaped(bd->bd_bh);
 696                         unlock_buffer(bd->bd_bh);
 697                         brelse(bd->bd_bh);
 698                 } else {
 699                         bh1 = gfs2_log_fake_buf(sdp, bd->bd_bh);
 700                 }
 701                 submit_bh(WRITE_SYNC, bh1);
 702                 gfs2_log_lock(sdp);
 703                 ptr += 2;
 704         }
 705         gfs2_log_unlock(sdp);
 706         brelse(bh);
 707 }
 708
 709 /**
 710  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 711  *
 712  */
 713
 714 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
 715 {
 716         struct gfs2_bufdata *bd = NULL;
 717         struct buffer_head *bh = NULL;
 718         unsigned int n = 0;
 719         __be64 *ptr = NULL, *end = NULL;
 720         LIST_HEAD(processed);
 721         LIST_HEAD(in_progress);
 722
 723         gfs2_log_lock(sdp);
 724         while (!list_empty(&sdp->sd_log_le_databuf)) {
 725                 if (ptr == end) {
 726                         gfs2_log_unlock(sdp);
 727                         gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
 728                         n = 0;
 729                         bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_JDATA);
 730                         ptr = bh_log_ptr(bh);
 731                         end = bh_ptr_end(bh) - 1;
 732                         gfs2_log_lock(sdp);
 733                         continue;
 734                 }
 735                 bd = list_entry(sdp->sd_log_le_databuf.next, struct gfs2_bufdata, bd_le.le_list);
 736                 list_move_tail(&bd->bd_le.le_list, &in_progress);
 737                 gfs2_check_magic(bd->bd_bh);
 738                 *ptr++ = cpu_to_be64(bd->bd_bh->b_blocknr);
 739                 *ptr++ = cpu_to_be64(buffer_escaped(bh) ? 1 : 0);
 740                 n++;
 741         }
 742         gfs2_log_unlock(sdp);
 743         gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
 744         gfs2_log_lock(sdp);
 745         list_splice(&processed, &sdp->sd_log_le_databuf);
 746         gfs2_log_unlock(sdp);
 747 }
 748
 749 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 750                                     struct gfs2_log_descriptor *ld,
 751                                     __be64 *ptr, int pass)
 752 {
 753         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 754         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 755         struct gfs2_glock *gl = ip->i_gl;
 756         unsigned int blks = be32_to_cpu(ld->ld_data1);
 757         struct buffer_head *bh_log, *bh_ip;
 758         u64 blkno;
 759         u64 esc;
 760         int error = 0;
 761
 762         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
 763                 return 0;
 764
 765         gfs2_replay_incr_blk(sdp, &start);
 766         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 767                 blkno = be64_to_cpu(*ptr++);
 768                 esc = be64_to_cpu(*ptr++);
 769
 770                 sdp->sd_found_blocks++;
 771
 772                 if (gfs2_revoke_check(sdp, blkno, start))
 773                         continue;
 774
 775                 error = gfs2_replay_read_block(jd, start, &bh_log);
 776                 if (error)
 777                         return error;
 778
 779                 bh_ip = gfs2_meta_new(gl, blkno);
 780                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 781
 782                 /* Unescape */
 783                 if (esc) {
 784                         __be32 *eptr = (__be32 *)bh_ip->b_data;
 785                         *eptr = cpu_to_be32(GFS2_MAGIC);
 786                 }
 787                 mark_buffer_dirty(bh_ip);
 788
 789                 brelse(bh_log);
 790                 brelse(bh_ip);
 791
 792                 sdp->sd_replayed_blocks++;
 793         }
 794
 795         return error;
 796 }
 797
 798 /* FIXME: sort out accounting for log blocks etc. */
 799
 800 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 801 {
 802         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 803         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 804
 805         if (error) {
 806                 gfs2_meta_sync(ip->i_gl);
 807                 return;
 808         }
 809         if (pass != 1)
 810                 return;
 811
 812         /* data sync? */
 813         gfs2_meta_sync(ip->i_gl);
 814
 815         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
 816                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 817 }
 818
 819 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 820 {
 821         struct list_head *head = &sdp->sd_log_le_databuf;
 822         struct gfs2_bufdata *bd;
 823
 824         while (!list_empty(head)) {
 825                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 826                 list_del_init(&bd->bd_le.le_list);
 827                 sdp->sd_log_num_databuf--;
 828                 gfs2_unpin(sdp, bd->bd_bh, ai);
 829         }
 830         gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
 831 }
 832
 833
 834 const struct gfs2_log_operations gfs2_buf_lops = {
 835         .lo_add = buf_lo_add,
 836         .lo_before_commit = buf_lo_before_commit,
 837         .lo_after_commit = buf_lo_after_commit,
 838         .lo_before_scan = buf_lo_before_scan,
 839         .lo_scan_elements = buf_lo_scan_elements,
 840         .lo_after_scan = buf_lo_after_scan,
 841         .lo_name = "buf",
 842 };
 843
 844 const struct gfs2_log_operations gfs2_revoke_lops = {
 845         .lo_add = revoke_lo_add,
 846         .lo_before_commit = revoke_lo_before_commit,
 847         .lo_after_commit = revoke_lo_after_commit,
 848         .lo_before_scan = revoke_lo_before_scan,
 849         .lo_scan_elements = revoke_lo_scan_elements,
 850         .lo_after_scan = revoke_lo_after_scan,
 851         .lo_name = "revoke",
 852 };
 853
 854 const struct gfs2_log_operations gfs2_rg_lops = {
 855         .lo_name = "rg",
 856 };
 857
 858 const struct gfs2_log_operations gfs2_databuf_lops = {
 859         .lo_add = databuf_lo_add,
 860         .lo_before_commit = databuf_lo_before_commit,
 861         .lo_after_commit = databuf_lo_after_commit,
 862         .lo_scan_elements = databuf_lo_scan_elements,
 863         .lo_after_scan = databuf_lo_after_scan,
 864         .lo_name = "databuf",
 865 };
 866
 867 const struct gfs2_log_operations *gfs2_log_ops[] = {
 868         &gfs2_databuf_lops,
 869         &gfs2_buf_lops,
 870         &gfs2_rg_lops,
 871         &gfs2_revoke_lops,
 872         NULL,
 873 };
 874