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 int err; 38 struct buffer_head *bh, *head; 39 int nr_underway = 0; 40 int write_op = (1 << BIO_RW_META) | ((wbc->sync_mode == WB_SYNC_ALL ? 41 WRITE_SYNC_PLUG : WRITE)); 42 43 BUG_ON(!PageLocked(page)); 44 BUG_ON(!page_has_buffers(page)); 45 46 head = page_buffers(page); 47 bh = head; 48 49 do { 50 if (!buffer_mapped(bh)) 51 continue; 52 /* 53 * If it's a fully non-blocking write attempt and we cannot 54 * lock the buffer then redirty the page. Note that this can 55 * potentially cause a busy-wait loop from pdflush and kswapd 56 * activity, but those code paths have their own higher-level 57 * throttling. 58 */ 59 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) { 60 lock_buffer(bh); 61 } else if (!trylock_buffer(bh)) { 62 redirty_page_for_writepage(wbc, page); 63 continue; 64 } 65 if (test_clear_buffer_dirty(bh)) { 66 mark_buffer_async_write(bh); 67 } else { 68 unlock_buffer(bh); 69 } 70 } while ((bh = bh->b_this_page) != head); 71 72 /* 73 * The page and its buffers are protected by PageWriteback(), so we can 74 * drop the bh refcounts early. 75 */ 76 BUG_ON(PageWriteback(page)); 77 set_page_writeback(page); 78 79 do { 80 struct buffer_head *next = bh->b_this_page; 81 if (buffer_async_write(bh)) { 82 submit_bh(write_op, bh); 83 nr_underway++; 84 } 85 bh = next; 86 } while (bh != head); 87 unlock_page(page); 88 89 err = 0; 90 if (nr_underway == 0) 91 end_page_writeback(page); 92 93 return err; 94 } 95 96 const struct address_space_operations gfs2_meta_aops = { 97 .writepage = gfs2_aspace_writepage, 98 .releasepage = gfs2_releasepage, 99 .sync_page = block_sync_page, 100 }; 101 102 /** 103 * gfs2_meta_sync - Sync all buffers associated with a glock 104 * @gl: The glock 105 * 106 */ 107 108 void gfs2_meta_sync(struct gfs2_glock *gl) 109 { 110 struct address_space *mapping = gfs2_glock2aspace(gl); 111 int error; 112 113 filemap_fdatawrite(mapping); 114 error = filemap_fdatawait(mapping); 115 116 if (error) 117 gfs2_io_error(gl->gl_sbd); 118 } 119 120 /** 121 * gfs2_getbuf - Get a buffer with a given address space 122 * @gl: the glock 123 * @blkno: the block number (filesystem scope) 124 * @create: 1 if the buffer should be created 125 * 126 * Returns: the buffer 127 */ 128 129 struct buffer_head *gfs2_getbuf(struct gfs2_glock *gl, u64 blkno, int create) 130 { 131 struct address_space *mapping = gfs2_glock2aspace(gl); 132 struct gfs2_sbd *sdp = gl->gl_sbd; 133 struct page *page; 134 struct buffer_head *bh; 135 unsigned int shift; 136 unsigned long index; 137 unsigned int bufnum; 138 139 shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift; 140 index = blkno >> shift; /* convert block to page */ 141 bufnum = blkno - (index << shift); /* block buf index within page */ 142 143 if (create) { 144 for (;;) { 145 page = grab_cache_page(mapping, index); 146 if (page) 147 break; 148 yield(); 149 } 150 } else { 151 page = find_lock_page(mapping, index); 152 if (!page) 153 return NULL; 154 } 155 156 if (!page_has_buffers(page)) 157 create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0); 158 159 /* Locate header for our buffer within our page */ 160 for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page) 161 /* Do nothing */; 162 get_bh(bh); 163 164 if (!buffer_mapped(bh)) 165 map_bh(bh, sdp->sd_vfs, blkno); 166 167 unlock_page(page); 168 mark_page_accessed(page); 169 page_cache_release(page); 170 171 return bh; 172 } 173 174 static void meta_prep_new(struct buffer_head *bh) 175 { 176 struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data; 177 178 lock_buffer(bh); 179 clear_buffer_dirty(bh); 180 set_buffer_uptodate(bh); 181 unlock_buffer(bh); 182 183 mh->mh_magic = cpu_to_be32(GFS2_MAGIC); 184 } 185 186 /** 187 * gfs2_meta_new - Get a block 188 * @gl: The glock associated with this block 189 * @blkno: The block number 190 * 191 * Returns: The buffer 192 */ 193 194 struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno) 195 { 196 struct buffer_head *bh; 197 bh = gfs2_getbuf(gl, blkno, CREATE); 198 meta_prep_new(bh); 199 return bh; 200 } 201 202 /** 203 * gfs2_meta_read - Read a block from disk 204 * @gl: The glock covering the block 205 * @blkno: The block number 206 * @flags: flags 207 * @bhp: the place where the buffer is returned (NULL on failure) 208 * 209 * Returns: errno 210 */ 211 212 int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags, 213 struct buffer_head **bhp) 214 { 215 struct gfs2_sbd *sdp = gl->gl_sbd; 216 struct buffer_head *bh; 217 218 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 219 return -EIO; 220 221 *bhp = bh = gfs2_getbuf(gl, blkno, CREATE); 222 223 lock_buffer(bh); 224 if (buffer_uptodate(bh)) { 225 unlock_buffer(bh); 226 return 0; 227 } 228 bh->b_end_io = end_buffer_read_sync; 229 get_bh(bh); 230 submit_bh(READ_SYNC | (1 << BIO_RW_META), bh); 231 if (!(flags & DIO_WAIT)) 232 return 0; 233 234 wait_on_buffer(bh); 235 if (unlikely(!buffer_uptodate(bh))) { 236 struct gfs2_trans *tr = current->journal_info; 237 if (tr && tr->tr_touched) 238 gfs2_io_error_bh(sdp, bh); 239 brelse(bh); 240 return -EIO; 241 } 242 243 return 0; 244 } 245 246 /** 247 * gfs2_meta_wait - Reread a block from disk 248 * @sdp: the filesystem 249 * @bh: The block to wait for 250 * 251 * Returns: errno 252 */ 253 254 int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh) 255 { 256 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 257 return -EIO; 258 259 wait_on_buffer(bh); 260 261 if (!buffer_uptodate(bh)) { 262 struct gfs2_trans *tr = current->journal_info; 263 if (tr && tr->tr_touched) 264 gfs2_io_error_bh(sdp, bh); 265 return -EIO; 266 } 267 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 268 return -EIO; 269 270 return 0; 271 } 272 273 /** 274 * gfs2_attach_bufdata - attach a struct gfs2_bufdata structure to a buffer 275 * @gl: the glock the buffer belongs to 276 * @bh: The buffer to be attached to 277 * @meta: Flag to indicate whether its metadata or not 278 */ 279 280 void gfs2_attach_bufdata(struct gfs2_glock *gl, struct buffer_head *bh, 281 int meta) 282 { 283 struct gfs2_bufdata *bd; 284 285 if (meta) 286 lock_page(bh->b_page); 287 288 if (bh->b_private) { 289 if (meta) 290 unlock_page(bh->b_page); 291 return; 292 } 293 294 bd = kmem_cache_zalloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL); 295 bd->bd_bh = bh; 296 bd->bd_gl = gl; 297 298 INIT_LIST_HEAD(&bd->bd_list_tr); 299 if (meta) 300 lops_init_le(&bd->bd_le, &gfs2_buf_lops); 301 else 302 lops_init_le(&bd->bd_le, &gfs2_databuf_lops); 303 bh->b_private = bd; 304 305 if (meta) 306 unlock_page(bh->b_page); 307 } 308 309 void gfs2_remove_from_journal(struct buffer_head *bh, struct gfs2_trans *tr, int meta) 310 { 311 struct address_space *mapping = bh->b_page->mapping; 312 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping); 313 struct gfs2_bufdata *bd = bh->b_private; 314 315 if (test_clear_buffer_pinned(bh)) { 316 list_del_init(&bd->bd_le.le_list); 317 if (meta) { 318 gfs2_assert_warn(sdp, sdp->sd_log_num_buf); 319 sdp->sd_log_num_buf--; 320 tr->tr_num_buf_rm++; 321 } else { 322 gfs2_assert_warn(sdp, sdp->sd_log_num_databuf); 323 sdp->sd_log_num_databuf--; 324 tr->tr_num_databuf_rm++; 325 } 326 tr->tr_touched = 1; 327 brelse(bh); 328 } 329 if (bd) { 330 if (bd->bd_ail) { 331 gfs2_remove_from_ail(bd); 332 bh->b_private = NULL; 333 bd->bd_bh = NULL; 334 bd->bd_blkno = bh->b_blocknr; 335 gfs2_trans_add_revoke(sdp, bd); 336 } 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 | (1 << BIO_RW_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 } 457 458