fs/gfs2/meta_io.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/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 #include <linux/lm_interface.h>
  23
  24 #include "gfs2.h"
  25 #include "incore.h"
  26 #include "glock.h"
  27 #include "glops.h"
  28 #include "inode.h"
  29 #include "log.h"
  30 #include "lops.h"
  31 #include "meta_io.h"
  32 #include "rgrp.h"
  33 #include "trans.h"
  34 #include "util.h"
  35 #include "ops_address.h"
  36
  37 static int aspace_get_block(struct inode *inode, sector_t lblock,
  38                             struct buffer_head *bh_result, int create)
  39 {
  40         gfs2_assert_warn(inode->i_sb->s_fs_info, 0);
  41         return -EOPNOTSUPP;
  42 }
  43
  44 static int gfs2_aspace_writepage(struct page *page,
  45                                  struct writeback_control *wbc)
  46 {
  47         return block_write_full_page(page, aspace_get_block, wbc);
  48 }
  49
  50 static const struct address_space_operations aspace_aops = {
  51         .writepage = gfs2_aspace_writepage,
  52         .releasepage = gfs2_releasepage,
  53         .sync_page = block_sync_page,
  54 };
  55
  56 /**
  57  * gfs2_aspace_get - Create and initialize a struct inode structure
  58  * @sdp: the filesystem the aspace is in
  59  *
  60  * Right now a struct inode is just a struct inode.  Maybe Linux
  61  * will supply a more lightweight address space construct (that works)
  62  * in the future.
  63  *
  64  * Make sure pages/buffers in this aspace aren't in high memory.
  65  *
  66  * Returns: the aspace
  67  */
  68
  69 struct inode *gfs2_aspace_get(struct gfs2_sbd *sdp)
  70 {
  71         struct inode *aspace;
  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                 aspace->i_private = NULL;
  79                 insert_inode_hash(aspace);
  80         }
  81         return aspace;
  82 }
  83
  84 void gfs2_aspace_put(struct inode *aspace)
  85 {
  86         remove_inode_hash(aspace);
  87         iput(aspace);
  88 }
  89
  90 /**
  91  * gfs2_meta_inval - Invalidate all buffers associated with a glock
  92  * @gl: the glock
  93  *
  94  */
  95
  96 void gfs2_meta_inval(struct gfs2_glock *gl)
  97 {
  98         struct gfs2_sbd *sdp = gl->gl_sbd;
  99         struct inode *aspace = gl->gl_aspace;
 100         struct address_space *mapping = gl->gl_aspace->i_mapping;
 101
 102         gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
 103
 104         atomic_inc(&aspace->i_writecount);
 105         truncate_inode_pages(mapping, 0);
 106         atomic_dec(&aspace->i_writecount);
 107
 108         gfs2_assert_withdraw(sdp, !mapping->nrpages);
 109 }
 110
 111 /**
 112  * gfs2_meta_sync - Sync all buffers associated with a glock
 113  * @gl: The glock
 114  *
 115  */
 116
 117 void gfs2_meta_sync(struct gfs2_glock *gl)
 118 {
 119         struct address_space *mapping = gl->gl_aspace->i_mapping;
 120         int error;
 121
 122         filemap_fdatawrite(mapping);
 123         error = filemap_fdatawait(mapping);
 124
 125         if (error)
 126                 gfs2_io_error(gl->gl_sbd);
 127 }
 128
 129 /**
 130  * getbuf - Get a buffer with a given address space
 131  * @gl: the glock
 132  * @blkno: the block number (filesystem scope)
 133  * @create: 1 if the buffer should be created
 134  *
 135  * Returns: the buffer
 136  */
 137
 138 static struct buffer_head *getbuf(struct gfs2_glock *gl, u64 blkno, int create)
 139 {
 140         struct address_space *mapping = gl->gl_aspace->i_mapping;
 141         struct gfs2_sbd *sdp = gl->gl_sbd;
 142         struct page *page;
 143         struct buffer_head *bh;
 144         unsigned int shift;
 145         unsigned long index;
 146         unsigned int bufnum;
 147
 148         shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift;
 149         index = blkno >> shift;             /* convert block to page */
 150         bufnum = blkno - (index << shift);  /* block buf index within page */
 151
 152         if (create) {
 153                 for (;;) {
 154                         page = grab_cache_page(mapping, index);
 155                         if (page)
 156                                 break;
 157                         yield();
 158                 }
 159         } else {
 160                 page = find_lock_page(mapping, index);
 161                 if (!page)
 162                         return NULL;
 163         }
 164
 165         if (!page_has_buffers(page))
 166                 create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0);
 167
 168         /* Locate header for our buffer within our page */
 169         for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page)
 170                 /* Do nothing */;
 171         get_bh(bh);
 172
 173         if (!buffer_mapped(bh))
 174                 map_bh(bh, sdp->sd_vfs, blkno);
 175
 176         unlock_page(page);
 177         mark_page_accessed(page);
 178         page_cache_release(page);
 179
 180         return bh;
 181 }
 182
 183 static void meta_prep_new(struct buffer_head *bh)
 184 {
 185         struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data;
 186
 187         lock_buffer(bh);
 188         clear_buffer_dirty(bh);
 189         set_buffer_uptodate(bh);
 190         unlock_buffer(bh);
 191
 192         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 193 }
 194
 195 /**
 196  * gfs2_meta_new - Get a block
 197  * @gl: The glock associated with this block
 198  * @blkno: The block number
 199  *
 200  * Returns: The buffer
 201  */
 202
 203 struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno)
 204 {
 205         struct buffer_head *bh;
 206         bh = getbuf(gl, blkno, CREATE);
 207         meta_prep_new(bh);
 208         return bh;
 209 }
 210
 211 /**
 212  * gfs2_meta_read - Read a block from disk
 213  * @gl: The glock covering the block
 214  * @blkno: The block number
 215  * @flags: flags
 216  * @bhp: the place where the buffer is returned (NULL on failure)
 217  *
 218  * Returns: errno
 219  */
 220
 221 int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags,
 222                    struct buffer_head **bhp)
 223 {
 224         *bhp = getbuf(gl, blkno, CREATE);
 225         if (!buffer_uptodate(*bhp)) {
 226                 ll_rw_block(READ_META, 1, bhp);
 227                 if (flags & DIO_WAIT) {
 228                         int error = gfs2_meta_wait(gl->gl_sbd, *bhp);
 229                         if (error) {
 230                                 brelse(*bhp);
 231                                 return error;
 232                         }
 233                 }
 234         }
 235
 236         return 0;
 237 }
 238
 239 /**
 240  * gfs2_meta_wait - Reread a block from disk
 241  * @sdp: the filesystem
 242  * @bh: The block to wait for
 243  *
 244  * Returns: errno
 245  */
 246
 247 int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh)
 248 {
 249         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 250                 return -EIO;
 251
 252         wait_on_buffer(bh);
 253
 254         if (!buffer_uptodate(bh)) {
 255                 struct gfs2_trans *tr = current->journal_info;
 256                 if (tr && tr->tr_touched)
 257                         gfs2_io_error_bh(sdp, bh);
 258                 return -EIO;
 259         }
 260         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 261                 return -EIO;
 262
 263         return 0;
 264 }
 265
 266 /**
 267  * gfs2_attach_bufdata - attach a struct gfs2_bufdata structure to a buffer
 268  * @gl: the glock the buffer belongs to
 269  * @bh: The buffer to be attached to
 270  * @meta: Flag to indicate whether its metadata or not
 271  */
 272
 273 void gfs2_attach_bufdata(struct gfs2_glock *gl, struct buffer_head *bh,
 274                          int meta)
 275 {
 276         struct gfs2_bufdata *bd;
 277
 278         if (meta)
 279                 lock_page(bh->b_page);
 280
 281         if (bh->b_private) {
 282                 if (meta)
 283                         unlock_page(bh->b_page);
 284                 return;
 285         }
 286
 287         bd = kmem_cache_zalloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL);
 288         bd->bd_bh = bh;
 289         bd->bd_gl = gl;
 290
 291         INIT_LIST_HEAD(&bd->bd_list_tr);
 292         if (meta)
 293                 lops_init_le(&bd->bd_le, &gfs2_buf_lops);
 294         else
 295                 lops_init_le(&bd->bd_le, &gfs2_databuf_lops);
 296         bh->b_private = bd;
 297
 298         if (meta)
 299                 unlock_page(bh->b_page);
 300 }
 301
 302 void gfs2_remove_from_journal(struct buffer_head *bh, struct gfs2_trans *tr, int meta)
 303 {
 304         struct gfs2_sbd *sdp = GFS2_SB(bh->b_page->mapping->host);
 305         struct gfs2_bufdata *bd = bh->b_private;
 306         if (test_clear_buffer_pinned(bh)) {
 307                 list_del_init(&bd->bd_le.le_list);
 308                 if (meta) {
 309                         gfs2_assert_warn(sdp, sdp->sd_log_num_buf);
 310                         sdp->sd_log_num_buf--;
 311                         tr->tr_num_buf_rm++;
 312                 } else {
 313                         gfs2_assert_warn(sdp, sdp->sd_log_num_databuf);
 314                         sdp->sd_log_num_databuf--;
 315                         tr->tr_num_databuf_rm++;
 316                 }
 317                 tr->tr_touched = 1;
 318                 brelse(bh);
 319         }
 320         if (bd) {
 321                 if (bd->bd_ail) {
 322                         gfs2_remove_from_ail(bd);
 323                         bh->b_private = NULL;
 324                         bd->bd_bh = NULL;
 325                         bd->bd_blkno = bh->b_blocknr;
 326                         gfs2_trans_add_revoke(sdp, bd);
 327                 }
 328         }
 329         clear_buffer_dirty(bh);
 330         clear_buffer_uptodate(bh);
 331 }
 332
 333 /**
 334  * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore
 335  * @ip: the inode who owns the buffers
 336  * @bstart: the first buffer in the run
 337  * @blen: the number of buffers in the run
 338  *
 339  */
 340
 341 void gfs2_meta_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen)
 342 {
 343         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 344         struct buffer_head *bh;
 345
 346         while (blen) {
 347                 bh = getbuf(ip->i_gl, bstart, NO_CREATE);
 348                 if (bh) {
 349                         lock_buffer(bh);
 350                         gfs2_log_lock(sdp);
 351                         gfs2_remove_from_journal(bh, current->journal_info, 1);
 352                         gfs2_log_unlock(sdp);
 353                         unlock_buffer(bh);
 354                         brelse(bh);
 355                 }
 356
 357                 bstart++;
 358                 blen--;
 359         }
 360 }
 361
 362 /**
 363  * gfs2_meta_indirect_buffer - Get a metadata buffer
 364  * @ip: The GFS2 inode
 365  * @height: The level of this buf in the metadata (indir addr) tree (if any)
 366  * @num: The block number (device relative) of the buffer
 367  * @new: Non-zero if we may create a new buffer
 368  * @bhp: the buffer is returned here
 369  *
 370  * Returns: errno
 371  */
 372
 373 int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, u64 num,
 374                               int new, struct buffer_head **bhp)
 375 {
 376         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 377         struct gfs2_glock *gl = ip->i_gl;
 378         struct buffer_head *bh;
 379         int ret = 0;
 380
 381         if (new) {
 382                 BUG_ON(height == 0);
 383                 bh = gfs2_meta_new(gl, num);
 384                 gfs2_trans_add_bh(ip->i_gl, bh, 1);
 385                 gfs2_metatype_set(bh, GFS2_METATYPE_IN, GFS2_FORMAT_IN);
 386                 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
 387         } else {
 388                 u32 mtype = height ? GFS2_METATYPE_IN : GFS2_METATYPE_DI;
 389                 ret = gfs2_meta_read(gl, num, DIO_WAIT, &bh);
 390                 if (ret == 0 && gfs2_metatype_check(sdp, bh, mtype)) {
 391                         brelse(bh);
 392                         ret = -EIO;
 393                 }
 394         }
 395         *bhp = bh;
 396         return ret;
 397 }
 398
 399 /**
 400  * gfs2_meta_ra - start readahead on an extent of a file
 401  * @gl: the glock the blocks belong to
 402  * @dblock: the starting disk block
 403  * @extlen: the number of blocks in the extent
 404  *
 405  * returns: the first buffer in the extent
 406  */
 407
 408 struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen)
 409 {
 410         struct gfs2_sbd *sdp = gl->gl_sbd;
 411         struct buffer_head *first_bh, *bh;
 412         u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >>
 413                           sdp->sd_sb.sb_bsize_shift;
 414
 415         BUG_ON(!extlen);
 416
 417         if (max_ra < 1)
 418                 max_ra = 1;
 419         if (extlen > max_ra)
 420                 extlen = max_ra;
 421
 422         first_bh = getbuf(gl, dblock, CREATE);
 423
 424         if (buffer_uptodate(first_bh))
 425                 goto out;
 426         if (!buffer_locked(first_bh))
 427                 ll_rw_block(READ_META, 1, &first_bh);
 428
 429         dblock++;
 430         extlen--;
 431
 432         while (extlen) {
 433                 bh = getbuf(gl, dblock, CREATE);
 434
 435                 if (!buffer_uptodate(bh) && !buffer_locked(bh))
 436                         ll_rw_block(READA, 1, &bh);
 437                 brelse(bh);
 438                 dblock++;
 439                 extlen--;
 440                 if (!buffer_locked(first_bh) && buffer_uptodate(first_bh))
 441                         goto out;
 442         }
 443
 444         wait_on_buffer(first_bh);
 445 out:
 446         return first_bh;
 447 }
 448