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 "trace_gfs2.h" 35 36 static int gfs2_aspace_writepage(struct page *page, struct writeback_control *wbc) 37 { 38 struct buffer_head *bh, *head; 39 int nr_underway = 0; 40 int write_op = REQ_META | REQ_PRIO | 41 (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : 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 flusher thread and kswapd 56 * activity, but those code paths have their own higher-level 57 * throttling. 58 */ 59 if (wbc->sync_mode != WB_SYNC_NONE) { 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 if (nr_underway == 0) 90 end_page_writeback(page); 91 92 return 0; 93 } 94 95 const struct address_space_operations gfs2_meta_aops = { 96 .writepage = gfs2_aspace_writepage, 97 .releasepage = gfs2_releasepage, 98 }; 99 100 /** 101 * gfs2_meta_sync - Sync all buffers associated with a glock 102 * @gl: The glock 103 * 104 */ 105 106 void gfs2_meta_sync(struct gfs2_glock *gl) 107 { 108 struct address_space *mapping = gfs2_glock2aspace(gl); 109 int error; 110 111 filemap_fdatawrite(mapping); 112 error = filemap_fdatawait(mapping); 113 114 if (error) 115 gfs2_io_error(gl->gl_sbd); 116 } 117 118 /** 119 * gfs2_getbuf - Get a buffer with a given address space 120 * @gl: the glock 121 * @blkno: the block number (filesystem scope) 122 * @create: 1 if the buffer should be created 123 * 124 * Returns: the buffer 125 */ 126 127 struct buffer_head *gfs2_getbuf(struct gfs2_glock *gl, u64 blkno, int create) 128 { 129 struct address_space *mapping = gfs2_glock2aspace(gl); 130 struct gfs2_sbd *sdp = gl->gl_sbd; 131 struct page *page; 132 struct buffer_head *bh; 133 unsigned int shift; 134 unsigned long index; 135 unsigned int bufnum; 136 137 shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift; 138 index = blkno >> shift; /* convert block to page */ 139 bufnum = blkno - (index << shift); /* block buf index within page */ 140 141 if (create) { 142 for (;;) { 143 page = grab_cache_page(mapping, index); 144 if (page) 145 break; 146 yield(); 147 } 148 } else { 149 page = find_lock_page(mapping, index); 150 if (!page) 151 return NULL; 152 } 153 154 if (!page_has_buffers(page)) 155 create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0); 156 157 /* Locate header for our buffer within our page */ 158 for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page) 159 /* Do nothing */; 160 get_bh(bh); 161 162 if (!buffer_mapped(bh)) 163 map_bh(bh, sdp->sd_vfs, blkno); 164 165 unlock_page(page); 166 mark_page_accessed(page); 167 page_cache_release(page); 168 169 return bh; 170 } 171 172 static void meta_prep_new(struct buffer_head *bh) 173 { 174 struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data; 175 176 lock_buffer(bh); 177 clear_buffer_dirty(bh); 178 set_buffer_uptodate(bh); 179 unlock_buffer(bh); 180 181 mh->mh_magic = cpu_to_be32(GFS2_MAGIC); 182 } 183 184 /** 185 * gfs2_meta_new - Get a block 186 * @gl: The glock associated with this block 187 * @blkno: The block number 188 * 189 * Returns: The buffer 190 */ 191 192 struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno) 193 { 194 struct buffer_head *bh; 195 bh = gfs2_getbuf(gl, blkno, CREATE); 196 meta_prep_new(bh); 197 return bh; 198 } 199 200 /** 201 * gfs2_meta_read - Read a block from disk 202 * @gl: The glock covering the block 203 * @blkno: The block number 204 * @flags: flags 205 * @bhp: the place where the buffer is returned (NULL on failure) 206 * 207 * Returns: errno 208 */ 209 210 int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags, 211 struct buffer_head **bhp) 212 { 213 struct gfs2_sbd *sdp = gl->gl_sbd; 214 struct buffer_head *bh; 215 216 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) { 217 *bhp = NULL; 218 return -EIO; 219 } 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 | REQ_META | REQ_PRIO, 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 *bhp = NULL; 241 return -EIO; 242 } 243 244 return 0; 245 } 246 247 /** 248 * gfs2_meta_wait - Reread a block from disk 249 * @sdp: the filesystem 250 * @bh: The block to wait for 251 * 252 * Returns: errno 253 */ 254 255 int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh) 256 { 257 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 258 return -EIO; 259 260 wait_on_buffer(bh); 261 262 if (!buffer_uptodate(bh)) { 263 struct gfs2_trans *tr = current->journal_info; 264 if (tr && tr->tr_touched) 265 gfs2_io_error_bh(sdp, bh); 266 return -EIO; 267 } 268 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 269 return -EIO; 270 271 return 0; 272 } 273 274 /** 275 * gfs2_attach_bufdata - attach a struct gfs2_bufdata structure to a buffer 276 * @gl: the glock the buffer belongs to 277 * @bh: The buffer to be attached to 278 * @meta: Flag to indicate whether its metadata or not 279 */ 280 281 void gfs2_attach_bufdata(struct gfs2_glock *gl, struct buffer_head *bh, 282 int meta) 283 { 284 struct gfs2_bufdata *bd; 285 286 if (meta) 287 lock_page(bh->b_page); 288 289 if (bh->b_private) { 290 if (meta) 291 unlock_page(bh->b_page); 292 return; 293 } 294 295 bd = kmem_cache_zalloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL); 296 bd->bd_bh = bh; 297 bd->bd_gl = gl; 298 299 if (meta) 300 lops_init_le(bd, &gfs2_buf_lops); 301 else 302 lops_init_le(bd, &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 trace_gfs2_pin(bd, 0); 317 atomic_dec(&sdp->sd_log_pinned); 318 list_del_init(&bd->bd_list); 319 if (meta) { 320 gfs2_assert_warn(sdp, sdp->sd_log_num_buf); 321 sdp->sd_log_num_buf--; 322 tr->tr_num_buf_rm++; 323 } else { 324 gfs2_assert_warn(sdp, sdp->sd_log_num_databuf); 325 sdp->sd_log_num_databuf--; 326 tr->tr_num_databuf_rm++; 327 } 328 tr->tr_touched = 1; 329 brelse(bh); 330 } 331 if (bd) { 332 spin_lock(&sdp->sd_ail_lock); 333 if (bd->bd_ail) { 334 gfs2_remove_from_ail(bd); 335 bh->b_private = NULL; 336 bd->bd_bh = NULL; 337 bd->bd_blkno = bh->b_blocknr; 338 gfs2_trans_add_revoke(sdp, bd); 339 } 340 spin_unlock(&sdp->sd_ail_lock); 341 } 342 clear_buffer_dirty(bh); 343 clear_buffer_uptodate(bh); 344 } 345 346 /** 347 * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore 348 * @ip: the inode who owns the buffers 349 * @bstart: the first buffer in the run 350 * @blen: the number of buffers in the run 351 * 352 */ 353 354 void gfs2_meta_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen) 355 { 356 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 357 struct buffer_head *bh; 358 359 while (blen) { 360 bh = gfs2_getbuf(ip->i_gl, bstart, NO_CREATE); 361 if (bh) { 362 lock_buffer(bh); 363 gfs2_log_lock(sdp); 364 gfs2_remove_from_journal(bh, current->journal_info, 1); 365 gfs2_log_unlock(sdp); 366 unlock_buffer(bh); 367 brelse(bh); 368 } 369 370 bstart++; 371 blen--; 372 } 373 } 374 375 /** 376 * gfs2_meta_indirect_buffer - Get a metadata buffer 377 * @ip: The GFS2 inode 378 * @height: The level of this buf in the metadata (indir addr) tree (if any) 379 * @num: The block number (device relative) of the buffer 380 * @bhp: the buffer is returned here 381 * 382 * Returns: errno 383 */ 384 385 int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, u64 num, 386 struct buffer_head **bhp) 387 { 388 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 389 struct gfs2_glock *gl = ip->i_gl; 390 struct buffer_head *bh; 391 int ret = 0; 392 u32 mtype = height ? GFS2_METATYPE_IN : GFS2_METATYPE_DI; 393 394 ret = gfs2_meta_read(gl, num, DIO_WAIT, &bh); 395 if (ret == 0 && gfs2_metatype_check(sdp, bh, mtype)) { 396 brelse(bh); 397 ret = -EIO; 398 } 399 *bhp = bh; 400 return ret; 401 } 402 403 /** 404 * gfs2_meta_ra - start readahead on an extent of a file 405 * @gl: the glock the blocks belong to 406 * @dblock: the starting disk block 407 * @extlen: the number of blocks in the extent 408 * 409 * returns: the first buffer in the extent 410 */ 411 412 struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen) 413 { 414 struct gfs2_sbd *sdp = gl->gl_sbd; 415 struct buffer_head *first_bh, *bh; 416 u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >> 417 sdp->sd_sb.sb_bsize_shift; 418 419 BUG_ON(!extlen); 420 421 if (max_ra < 1) 422 max_ra = 1; 423 if (extlen > max_ra) 424 extlen = max_ra; 425 426 first_bh = gfs2_getbuf(gl, dblock, CREATE); 427 428 if (buffer_uptodate(first_bh)) 429 goto out; 430 if (!buffer_locked(first_bh)) 431 ll_rw_block(READ_SYNC | REQ_META, 1, &first_bh); 432 433 dblock++; 434 extlen--; 435 436 while (extlen) { 437 bh = gfs2_getbuf(gl, dblock, CREATE); 438 439 if (!buffer_uptodate(bh) && !buffer_locked(bh)) 440 ll_rw_block(READA | REQ_META, 1, &bh); 441 brelse(bh); 442 dblock++; 443 extlen--; 444 if (!buffer_locked(first_bh) && buffer_uptodate(first_bh)) 445 goto out; 446 } 447 448 wait_on_buffer(first_bh); 449 out: 450 return first_bh; 451 } 452 453