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
  35 static int gfs2_aspace_writepage(struct page *page, struct writeback_control *wbc)
  36 {
  37         struct buffer_head *bh, *head;
  38         int nr_underway = 0;
  39         int write_op = REQ_META |
  40                 (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE);
  41
  42         BUG_ON(!PageLocked(page));
  43         BUG_ON(!page_has_buffers(page));
  44
  45         head = page_buffers(page);
  46         bh = head;
  47
  48         do {
  49                 if (!buffer_mapped(bh))
  50                         continue;
  51                 /*
  52                  * If it's a fully non-blocking write attempt and we cannot
  53                  * lock the buffer then redirty the page.  Note that this can
  54                  * potentially cause a busy-wait loop from pdflush and kswapd
  55                  * activity, but those code paths have their own higher-level
  56                  * throttling.
  57                  */
  58                 if (wbc->sync_mode != WB_SYNC_NONE) {
  59                         lock_buffer(bh);
  60                 } else if (!trylock_buffer(bh)) {
  61                         redirty_page_for_writepage(wbc, page);
  62                         continue;
  63                 }
  64                 if (test_clear_buffer_dirty(bh)) {
  65                         mark_buffer_async_write(bh);
  66                 } else {
  67                         unlock_buffer(bh);
  68                 }
  69         } while ((bh = bh->b_this_page) != head);
  70
  71         /*
  72          * The page and its buffers are protected by PageWriteback(), so we can
  73          * drop the bh refcounts early.
  74          */
  75         BUG_ON(PageWriteback(page));
  76         set_page_writeback(page);
  77
  78         do {
  79                 struct buffer_head *next = bh->b_this_page;
  80                 if (buffer_async_write(bh)) {
  81                         submit_bh(write_op, bh);
  82                         nr_underway++;
  83                 }
  84                 bh = next;
  85         } while (bh != head);
  86         unlock_page(page);
  87
  88         if (nr_underway == 0)
  89                 end_page_writeback(page);
  90
  91         return 0;
  92 }
  93
  94 const struct address_space_operations gfs2_meta_aops = {
  95         .writepage = gfs2_aspace_writepage,
  96         .releasepage = gfs2_releasepage,
  97 };
  98
  99 /**
 100  * gfs2_meta_sync - Sync all buffers associated with a glock
 101  * @gl: The glock
 102  *
 103  */
 104
 105 void gfs2_meta_sync(struct gfs2_glock *gl)
 106 {
 107         struct address_space *mapping = gfs2_glock2aspace(gl);
 108         int error;
 109
 110         filemap_fdatawrite(mapping);
 111         error = filemap_fdatawait(mapping);
 112
 113         if (error)
 114                 gfs2_io_error(gl->gl_sbd);
 115 }
 116
 117 /**
 118  * gfs2_getbuf - Get a buffer with a given address space
 119  * @gl: the glock
 120  * @blkno: the block number (filesystem scope)
 121  * @create: 1 if the buffer should be created
 122  *
 123  * Returns: the buffer
 124  */
 125
 126 struct buffer_head *gfs2_getbuf(struct gfs2_glock *gl, u64 blkno, int create)
 127 {
 128         struct address_space *mapping = gfs2_glock2aspace(gl);
 129         struct gfs2_sbd *sdp = gl->gl_sbd;
 130         struct page *page;
 131         struct buffer_head *bh;
 132         unsigned int shift;
 133         unsigned long index;
 134         unsigned int bufnum;
 135
 136         shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift;
 137         index = blkno >> shift;             /* convert block to page */
 138         bufnum = blkno - (index << shift);  /* block buf index within page */
 139
 140         if (create) {
 141                 for (;;) {
 142                         page = grab_cache_page(mapping, index);
 143                         if (page)
 144                                 break;
 145                         yield();
 146                 }
 147         } else {
 148                 page = find_lock_page(mapping, index);
 149                 if (!page)
 150                         return NULL;
 151         }
 152
 153         if (!page_has_buffers(page))
 154                 create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0);
 155
 156         /* Locate header for our buffer within our page */
 157         for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page)
 158                 /* Do nothing */;
 159         get_bh(bh);
 160
 161         if (!buffer_mapped(bh))
 162                 map_bh(bh, sdp->sd_vfs, blkno);
 163
 164         unlock_page(page);
 165         mark_page_accessed(page);
 166         page_cache_release(page);
 167
 168         return bh;
 169 }
 170
 171 static void meta_prep_new(struct buffer_head *bh)
 172 {
 173         struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data;
 174
 175         lock_buffer(bh);
 176         clear_buffer_dirty(bh);
 177         set_buffer_uptodate(bh);
 178         unlock_buffer(bh);
 179
 180         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 181 }
 182
 183 /**
 184  * gfs2_meta_new - Get a block
 185  * @gl: The glock associated with this block
 186  * @blkno: The block number
 187  *
 188  * Returns: The buffer
 189  */
 190
 191 struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno)
 192 {
 193         struct buffer_head *bh;
 194         bh = gfs2_getbuf(gl, blkno, CREATE);
 195         meta_prep_new(bh);
 196         return bh;
 197 }
 198
 199 /**
 200  * gfs2_meta_read - Read a block from disk
 201  * @gl: The glock covering the block
 202  * @blkno: The block number
 203  * @flags: flags
 204  * @bhp: the place where the buffer is returned (NULL on failure)
 205  *
 206  * Returns: errno
 207  */
 208
 209 int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags,
 210                    struct buffer_head **bhp)
 211 {
 212         struct gfs2_sbd *sdp = gl->gl_sbd;
 213         struct buffer_head *bh;
 214
 215         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 216                 return -EIO;
 217
 218         *bhp = bh = gfs2_getbuf(gl, blkno, CREATE);
 219
 220         lock_buffer(bh);
 221         if (buffer_uptodate(bh)) {
 222                 unlock_buffer(bh);
 223                 return 0;
 224         }
 225         bh->b_end_io = end_buffer_read_sync;
 226         get_bh(bh);
 227         submit_bh(READ_SYNC | REQ_META, bh);
 228         if (!(flags & DIO_WAIT))
 229                 return 0;
 230
 231         wait_on_buffer(bh);
 232         if (unlikely(!buffer_uptodate(bh))) {
 233                 struct gfs2_trans *tr = current->journal_info;
 234                 if (tr && tr->tr_touched)
 235                         gfs2_io_error_bh(sdp, bh);
 236                 brelse(bh);
 237                 return -EIO;
 238         }
 239
 240         return 0;
 241 }
 242
 243 /**
 244  * gfs2_meta_wait - Reread a block from disk
 245  * @sdp: the filesystem
 246  * @bh: The block to wait for
 247  *
 248  * Returns: errno
 249  */
 250
 251 int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh)
 252 {
 253         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 254                 return -EIO;
 255
 256         wait_on_buffer(bh);
 257
 258         if (!buffer_uptodate(bh)) {
 259                 struct gfs2_trans *tr = current->journal_info;
 260                 if (tr && tr->tr_touched)
 261                         gfs2_io_error_bh(sdp, bh);
 262                 return -EIO;
 263         }
 264         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 265                 return -EIO;
 266
 267         return 0;
 268 }
 269
 270 /**
 271  * gfs2_attach_bufdata - attach a struct gfs2_bufdata structure to a buffer
 272  * @gl: the glock the buffer belongs to
 273  * @bh: The buffer to be attached to
 274  * @meta: Flag to indicate whether its metadata or not
 275  */
 276
 277 void gfs2_attach_bufdata(struct gfs2_glock *gl, struct buffer_head *bh,
 278                          int meta)
 279 {
 280         struct gfs2_bufdata *bd;
 281
 282         if (meta)
 283                 lock_page(bh->b_page);
 284
 285         if (bh->b_private) {
 286                 if (meta)
 287                         unlock_page(bh->b_page);
 288                 return;
 289         }
 290
 291         bd = kmem_cache_zalloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL);
 292         bd->bd_bh = bh;
 293         bd->bd_gl = gl;
 294
 295         INIT_LIST_HEAD(&bd->bd_list_tr);
 296         if (meta)
 297                 lops_init_le(&bd->bd_le, &gfs2_buf_lops);
 298         else
 299                 lops_init_le(&bd->bd_le, &gfs2_databuf_lops);
 300         bh->b_private = bd;
 301
 302         if (meta)
 303                 unlock_page(bh->b_page);
 304 }
 305
 306 void gfs2_remove_from_journal(struct buffer_head *bh, struct gfs2_trans *tr, int meta)
 307 {
 308         struct address_space *mapping = bh->b_page->mapping;
 309         struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
 310         struct gfs2_bufdata *bd = bh->b_private;
 311
 312         if (test_clear_buffer_pinned(bh)) {
 313                 atomic_dec(&sdp->sd_log_pinned);
 314                 list_del_init(&bd->bd_le.le_list);
 315                 if (meta) {
 316                         gfs2_assert_warn(sdp, sdp->sd_log_num_buf);
 317                         sdp->sd_log_num_buf--;
 318                         tr->tr_num_buf_rm++;
 319                 } else {
 320                         gfs2_assert_warn(sdp, sdp->sd_log_num_databuf);
 321                         sdp->sd_log_num_databuf--;
 322                         tr->tr_num_databuf_rm++;
 323                 }
 324                 tr->tr_touched = 1;
 325                 brelse(bh);
 326         }
 327         if (bd) {
 328                 spin_lock(&sdp->sd_ail_lock);
 329                 if (bd->bd_ail) {
 330                         gfs2_remove_from_ail(bd);
 331                         bh->b_private = NULL;
 332                         bd->bd_bh = NULL;
 333                         bd->bd_blkno = bh->b_blocknr;
 334                         gfs2_trans_add_revoke(sdp, bd);
 335                 }
 336                 spin_unlock(&sdp->sd_ail_lock);
 337         }
 338         clear_buffer_dirty(bh);
 339         clear_buffer_uptodate(bh);
 340 }
 341
 342 /**
 343  * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore
 344  * @ip: the inode who owns the buffers
 345  * @bstart: the first buffer in the run
 346  * @blen: the number of buffers in the run
 347  *
 348  */
 349
 350 void gfs2_meta_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen)
 351 {
 352         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 353         struct buffer_head *bh;
 354
 355         while (blen) {
 356                 bh = gfs2_getbuf(ip->i_gl, bstart, NO_CREATE);
 357                 if (bh) {
 358                         lock_buffer(bh);
 359                         gfs2_log_lock(sdp);
 360                         gfs2_remove_from_journal(bh, current->journal_info, 1);
 361                         gfs2_log_unlock(sdp);
 362                         unlock_buffer(bh);
 363                         brelse(bh);
 364                 }
 365
 366                 bstart++;
 367                 blen--;
 368         }
 369 }
 370
 371 /**
 372  * gfs2_meta_indirect_buffer - Get a metadata buffer
 373  * @ip: The GFS2 inode
 374  * @height: The level of this buf in the metadata (indir addr) tree (if any)
 375  * @num: The block number (device relative) of the buffer
 376  * @new: Non-zero if we may create a new buffer
 377  * @bhp: the buffer is returned here
 378  *
 379  * Returns: errno
 380  */
 381
 382 int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, u64 num,
 383                               int new, struct buffer_head **bhp)
 384 {
 385         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 386         struct gfs2_glock *gl = ip->i_gl;
 387         struct buffer_head *bh;
 388         int ret = 0;
 389
 390         if (new) {
 391                 BUG_ON(height == 0);
 392                 bh = gfs2_meta_new(gl, num);
 393                 gfs2_trans_add_bh(ip->i_gl, bh, 1);
 394                 gfs2_metatype_set(bh, GFS2_METATYPE_IN, GFS2_FORMAT_IN);
 395                 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
 396         } else {
 397                 u32 mtype = height ? GFS2_METATYPE_IN : GFS2_METATYPE_DI;
 398                 ret = gfs2_meta_read(gl, num, DIO_WAIT, &bh);
 399                 if (ret == 0 && gfs2_metatype_check(sdp, bh, mtype)) {
 400                         brelse(bh);
 401                         ret = -EIO;
 402                 }
 403         }
 404         *bhp = bh;
 405         return ret;
 406 }
 407
 408 /**
 409  * gfs2_meta_ra - start readahead on an extent of a file
 410  * @gl: the glock the blocks belong to
 411  * @dblock: the starting disk block
 412  * @extlen: the number of blocks in the extent
 413  *
 414  * returns: the first buffer in the extent
 415  */
 416
 417 struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen)
 418 {
 419         struct gfs2_sbd *sdp = gl->gl_sbd;
 420         struct buffer_head *first_bh, *bh;
 421         u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >>
 422                           sdp->sd_sb.sb_bsize_shift;
 423
 424         BUG_ON(!extlen);
 425
 426         if (max_ra < 1)
 427                 max_ra = 1;
 428         if (extlen > max_ra)
 429                 extlen = max_ra;
 430
 431         first_bh = gfs2_getbuf(gl, dblock, CREATE);
 432
 433         if (buffer_uptodate(first_bh))
 434                 goto out;
 435         if (!buffer_locked(first_bh))
 436                 ll_rw_block(READ_SYNC | REQ_META, 1, &first_bh);
 437
 438         dblock++;
 439         extlen--;
 440
 441         while (extlen) {
 442                 bh = gfs2_getbuf(gl, dblock, CREATE);
 443
 444                 if (!buffer_uptodate(bh) && !buffer_locked(bh))
 445                         ll_rw_block(READA, 1, &bh);
 446                 brelse(bh);
 447                 dblock++;
 448                 extlen--;
 449                 if (!buffer_locked(first_bh) && buffer_uptodate(first_bh))
 450                         goto out;
 451         }
 452
 453         wait_on_buffer(first_bh);
 454 out:
 455         return first_bh;
 456 }
 457