fs/gfs2/meta_io.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2008 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/mm.h>
  16 #include <linux/pagemap.h>
  17 #include <linux/writeback.h>
  18 #include <linux/swap.h>
  19 #include <linux/delay.h>
  20 #include <linux/bio.h>
  21 #include <linux/gfs2_ondisk.h>
  22
  23 #include "gfs2.h"
  24 #include "incore.h"
  25 #include "glock.h"
  26 #include "glops.h"
  27 #include "inode.h"
  28 #include "log.h"
  29 #include "lops.h"
  30 #include "meta_io.h"
  31 #include "rgrp.h"
  32 #include "trans.h"
  33 #include "util.h"
  34 #include "ops_address.h"
  35
  36 static int aspace_get_block(struct inode *inode, sector_t lblock,
  37                             struct buffer_head *bh_result, int create)
  38 {
  39         gfs2_assert_warn(inode->i_sb->s_fs_info, 0);
  40         return -EOPNOTSUPP;
  41 }
  42
  43 static int gfs2_aspace_writepage(struct page *page,
  44                                  struct writeback_control *wbc)
  45 {
  46         return block_write_full_page(page, aspace_get_block, wbc);
  47 }
  48
  49 static const struct address_space_operations aspace_aops = {
  50         .writepage = gfs2_aspace_writepage,
  51         .releasepage = gfs2_releasepage,
  52         .sync_page = block_sync_page,
  53 };
  54
  55 /**
  56  * gfs2_aspace_get - Create and initialize a struct inode structure
  57  * @sdp: the filesystem the aspace is in
  58  *
  59  * Right now a struct inode is just a struct inode.  Maybe Linux
  60  * will supply a more lightweight address space construct (that works)
  61  * in the future.
  62  *
  63  * Make sure pages/buffers in this aspace aren't in high memory.
  64  *
  65  * Returns: the aspace
  66  */
  67
  68 struct inode *gfs2_aspace_get(struct gfs2_sbd *sdp)
  69 {
  70         struct inode *aspace;
  71         struct gfs2_inode *ip;
  72
  73         aspace = new_inode(sdp->sd_vfs);
  74         if (aspace) {
  75                 mapping_set_gfp_mask(aspace->i_mapping, GFP_NOFS);
  76                 aspace->i_mapping->a_ops = &aspace_aops;
  77                 aspace->i_size = ~0ULL;
  78                 ip = GFS2_I(aspace);
  79                 clear_bit(GIF_USER, &ip->i_flags);
  80                 insert_inode_hash(aspace);
  81         }
  82         return aspace;
  83 }
  84
  85 void gfs2_aspace_put(struct inode *aspace)
  86 {
  87         remove_inode_hash(aspace);
  88         iput(aspace);
  89 }
  90
  91 /**
  92  * gfs2_meta_sync - Sync all buffers associated with a glock
  93  * @gl: The glock
  94  *
  95  */
  96
  97 void gfs2_meta_sync(struct gfs2_glock *gl)
  98 {
  99         struct address_space *mapping = gl->gl_aspace->i_mapping;
 100         int error;
 101
 102         filemap_fdatawrite(mapping);
 103         error = filemap_fdatawait(mapping);
 104
 105         if (error)
 106                 gfs2_io_error(gl->gl_sbd);
 107 }
 108
 109 /**
 110  * gfs2_getbuf - Get a buffer with a given address space
 111  * @gl: the glock
 112  * @blkno: the block number (filesystem scope)
 113  * @create: 1 if the buffer should be created
 114  *
 115  * Returns: the buffer
 116  */
 117
 118 struct buffer_head *gfs2_getbuf(struct gfs2_glock *gl, u64 blkno, int create)
 119 {
 120         struct address_space *mapping = gl->gl_aspace->i_mapping;
 121         struct gfs2_sbd *sdp = gl->gl_sbd;
 122         struct page *page;
 123         struct buffer_head *bh;
 124         unsigned int shift;
 125         unsigned long index;
 126         unsigned int bufnum;
 127
 128         shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift;
 129         index = blkno >> shift;             /* convert block to page */
 130         bufnum = blkno - (index << shift);  /* block buf index within page */
 131
 132         if (create) {
 133                 for (;;) {
 134                         page = grab_cache_page(mapping, index);
 135                         if (page)
 136                                 break;
 137                         yield();
 138                 }
 139         } else {
 140                 page = find_lock_page(mapping, index);
 141                 if (!page)
 142                         return NULL;
 143         }
 144
 145         if (!page_has_buffers(page))
 146                 create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0);
 147
 148         /* Locate header for our buffer within our page */
 149         for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page)
 150                 /* Do nothing */;
 151         get_bh(bh);
 152
 153         if (!buffer_mapped(bh))
 154                 map_bh(bh, sdp->sd_vfs, blkno);
 155
 156         unlock_page(page);
 157         mark_page_accessed(page);
 158         page_cache_release(page);
 159
 160         return bh;
 161 }
 162
 163 static void meta_prep_new(struct buffer_head *bh)
 164 {
 165         struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data;
 166
 167         lock_buffer(bh);
 168         clear_buffer_dirty(bh);
 169         set_buffer_uptodate(bh);
 170         unlock_buffer(bh);
 171
 172         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 173 }
 174
 175 /**
 176  * gfs2_meta_new - Get a block
 177  * @gl: The glock associated with this block
 178  * @blkno: The block number
 179  *
 180  * Returns: The buffer
 181  */
 182
 183 struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno)
 184 {
 185         struct buffer_head *bh;
 186         bh = gfs2_getbuf(gl, blkno, CREATE);
 187         meta_prep_new(bh);
 188         return bh;
 189 }
 190
 191 /**
 192  * gfs2_meta_read - Read a block from disk
 193  * @gl: The glock covering the block
 194  * @blkno: The block number
 195  * @flags: flags
 196  * @bhp: the place where the buffer is returned (NULL on failure)
 197  *
 198  * Returns: errno
 199  */
 200
 201 int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags,
 202                    struct buffer_head **bhp)
 203 {
 204         *bhp = gfs2_getbuf(gl, blkno, CREATE);
 205         if (!buffer_uptodate(*bhp)) {
 206                 ll_rw_block(READ_META, 1, bhp);
 207                 if (flags & DIO_WAIT) {
 208                         int error = gfs2_meta_wait(gl->gl_sbd, *bhp);
 209                         if (error) {
 210                                 brelse(*bhp);
 211                                 return error;
 212                         }
 213                 }
 214         }
 215
 216         return 0;
 217 }
 218
 219 /**
 220  * gfs2_meta_wait - Reread a block from disk
 221  * @sdp: the filesystem
 222  * @bh: The block to wait for
 223  *
 224  * Returns: errno
 225  */
 226
 227 int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh)
 228 {
 229         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 230                 return -EIO;
 231
 232         wait_on_buffer(bh);
 233
 234         if (!buffer_uptodate(bh)) {
 235                 struct gfs2_trans *tr = current->journal_info;
 236                 if (tr && tr->tr_touched)
 237                         gfs2_io_error_bh(sdp, bh);
 238                 return -EIO;
 239         }
 240         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 241                 return -EIO;
 242
 243         return 0;
 244 }
 245
 246 /**
 247  * gfs2_attach_bufdata - attach a struct gfs2_bufdata structure to a buffer
 248  * @gl: the glock the buffer belongs to
 249  * @bh: The buffer to be attached to
 250  * @meta: Flag to indicate whether its metadata or not
 251  */
 252
 253 void gfs2_attach_bufdata(struct gfs2_glock *gl, struct buffer_head *bh,
 254                          int meta)
 255 {
 256         struct gfs2_bufdata *bd;
 257
 258         if (meta)
 259                 lock_page(bh->b_page);
 260
 261         if (bh->b_private) {
 262                 if (meta)
 263                         unlock_page(bh->b_page);
 264                 return;
 265         }
 266
 267         bd = kmem_cache_zalloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL);
 268         bd->bd_bh = bh;
 269         bd->bd_gl = gl;
 270
 271         INIT_LIST_HEAD(&bd->bd_list_tr);
 272         if (meta)
 273                 lops_init_le(&bd->bd_le, &gfs2_buf_lops);
 274         else
 275                 lops_init_le(&bd->bd_le, &gfs2_databuf_lops);
 276         bh->b_private = bd;
 277
 278         if (meta)
 279                 unlock_page(bh->b_page);
 280 }
 281
 282 void gfs2_remove_from_journal(struct buffer_head *bh, struct gfs2_trans *tr, int meta)
 283 {
 284         struct gfs2_sbd *sdp = GFS2_SB(bh->b_page->mapping->host);
 285         struct gfs2_bufdata *bd = bh->b_private;
 286         if (test_clear_buffer_pinned(bh)) {
 287                 list_del_init(&bd->bd_le.le_list);
 288                 if (meta) {
 289                         gfs2_assert_warn(sdp, sdp->sd_log_num_buf);
 290                         sdp->sd_log_num_buf--;
 291                         tr->tr_num_buf_rm++;
 292                 } else {
 293                         gfs2_assert_warn(sdp, sdp->sd_log_num_databuf);
 294                         sdp->sd_log_num_databuf--;
 295                         tr->tr_num_databuf_rm++;
 296                 }
 297                 tr->tr_touched = 1;
 298                 brelse(bh);
 299         }
 300         if (bd) {
 301                 if (bd->bd_ail) {
 302                         gfs2_remove_from_ail(bd);
 303                         bh->b_private = NULL;
 304                         bd->bd_bh = NULL;
 305                         bd->bd_blkno = bh->b_blocknr;
 306                         gfs2_trans_add_revoke(sdp, bd);
 307                 }
 308         }
 309         clear_buffer_dirty(bh);
 310         clear_buffer_uptodate(bh);
 311 }
 312
 313 /**
 314  * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore
 315  * @ip: the inode who owns the buffers
 316  * @bstart: the first buffer in the run
 317  * @blen: the number of buffers in the run
 318  *
 319  */
 320
 321 void gfs2_meta_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen)
 322 {
 323         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 324         struct buffer_head *bh;
 325
 326         while (blen) {
 327                 bh = gfs2_getbuf(ip->i_gl, bstart, NO_CREATE);
 328                 if (bh) {
 329                         lock_buffer(bh);
 330                         gfs2_log_lock(sdp);
 331                         gfs2_remove_from_journal(bh, current->journal_info, 1);
 332                         gfs2_log_unlock(sdp);
 333                         unlock_buffer(bh);
 334                         brelse(bh);
 335                 }
 336
 337                 bstart++;
 338                 blen--;
 339         }
 340 }
 341
 342 /**
 343  * gfs2_meta_indirect_buffer - Get a metadata buffer
 344  * @ip: The GFS2 inode
 345  * @height: The level of this buf in the metadata (indir addr) tree (if any)
 346  * @num: The block number (device relative) of the buffer
 347  * @new: Non-zero if we may create a new buffer
 348  * @bhp: the buffer is returned here
 349  *
 350  * Returns: errno
 351  */
 352
 353 int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, u64 num,
 354                               int new, struct buffer_head **bhp)
 355 {
 356         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 357         struct gfs2_glock *gl = ip->i_gl;
 358         struct buffer_head *bh;
 359         int ret = 0;
 360
 361         if (new) {
 362                 BUG_ON(height == 0);
 363                 bh = gfs2_meta_new(gl, num);
 364                 gfs2_trans_add_bh(ip->i_gl, bh, 1);
 365                 gfs2_metatype_set(bh, GFS2_METATYPE_IN, GFS2_FORMAT_IN);
 366                 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
 367         } else {
 368                 u32 mtype = height ? GFS2_METATYPE_IN : GFS2_METATYPE_DI;
 369                 ret = gfs2_meta_read(gl, num, DIO_WAIT, &bh);
 370                 if (ret == 0 && gfs2_metatype_check(sdp, bh, mtype)) {
 371                         brelse(bh);
 372                         ret = -EIO;
 373                 }
 374         }
 375         *bhp = bh;
 376         return ret;
 377 }
 378
 379 /**
 380  * gfs2_meta_ra - start readahead on an extent of a file
 381  * @gl: the glock the blocks belong to
 382  * @dblock: the starting disk block
 383  * @extlen: the number of blocks in the extent
 384  *
 385  * returns: the first buffer in the extent
 386  */
 387
 388 struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen)
 389 {
 390         struct gfs2_sbd *sdp = gl->gl_sbd;
 391         struct buffer_head *first_bh, *bh;
 392         u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >>
 393                           sdp->sd_sb.sb_bsize_shift;
 394
 395         BUG_ON(!extlen);
 396
 397         if (max_ra < 1)
 398                 max_ra = 1;
 399         if (extlen > max_ra)
 400                 extlen = max_ra;
 401
 402         first_bh = gfs2_getbuf(gl, dblock, CREATE);
 403
 404         if (buffer_uptodate(first_bh))
 405                 goto out;
 406         if (!buffer_locked(first_bh))
 407                 ll_rw_block(READ_META, 1, &first_bh);
 408
 409         dblock++;
 410         extlen--;
 411
 412         while (extlen) {
 413                 bh = gfs2_getbuf(gl, dblock, CREATE);
 414
 415                 if (!buffer_uptodate(bh) && !buffer_locked(bh))
 416                         ll_rw_block(READA, 1, &bh);
 417                 brelse(bh);
 418                 dblock++;
 419                 extlen--;
 420                 if (!buffer_locked(first_bh) && buffer_uptodate(first_bh))
 421                         goto out;
 422         }
 423
 424         wait_on_buffer(first_bh);
 425 out:
 426         return first_bh;
 427 }
 428