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_getbuf - Get a buffer with a given address space 102 * @gl: the glock 103 * @blkno: the block number (filesystem scope) 104 * @create: 1 if the buffer should be created 105 * 106 * Returns: the buffer 107 */ 108 109 struct buffer_head *gfs2_getbuf(struct gfs2_glock *gl, u64 blkno, int create) 110 { 111 struct address_space *mapping = gfs2_glock2aspace(gl); 112 struct gfs2_sbd *sdp = gl->gl_sbd; 113 struct page *page; 114 struct buffer_head *bh; 115 unsigned int shift; 116 unsigned long index; 117 unsigned int bufnum; 118 119 shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift; 120 index = blkno >> shift; /* convert block to page */ 121 bufnum = blkno - (index << shift); /* block buf index within page */ 122 123 if (create) { 124 for (;;) { 125 page = grab_cache_page(mapping, index); 126 if (page) 127 break; 128 yield(); 129 } 130 } else { 131 page = find_lock_page(mapping, index); 132 if (!page) 133 return NULL; 134 } 135 136 if (!page_has_buffers(page)) 137 create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0); 138 139 /* Locate header for our buffer within our page */ 140 for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page) 141 /* Do nothing */; 142 get_bh(bh); 143 144 if (!buffer_mapped(bh)) 145 map_bh(bh, sdp->sd_vfs, blkno); 146 147 unlock_page(page); 148 mark_page_accessed(page); 149 page_cache_release(page); 150 151 return bh; 152 } 153 154 static void meta_prep_new(struct buffer_head *bh) 155 { 156 struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data; 157 158 lock_buffer(bh); 159 clear_buffer_dirty(bh); 160 set_buffer_uptodate(bh); 161 unlock_buffer(bh); 162 163 mh->mh_magic = cpu_to_be32(GFS2_MAGIC); 164 } 165 166 /** 167 * gfs2_meta_new - Get a block 168 * @gl: The glock associated with this block 169 * @blkno: The block number 170 * 171 * Returns: The buffer 172 */ 173 174 struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno) 175 { 176 struct buffer_head *bh; 177 bh = gfs2_getbuf(gl, blkno, CREATE); 178 meta_prep_new(bh); 179 return bh; 180 } 181 182 /** 183 * gfs2_meta_read - Read a block from disk 184 * @gl: The glock covering the block 185 * @blkno: The block number 186 * @flags: flags 187 * @bhp: the place where the buffer is returned (NULL on failure) 188 * 189 * Returns: errno 190 */ 191 192 int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags, 193 struct buffer_head **bhp) 194 { 195 struct gfs2_sbd *sdp = gl->gl_sbd; 196 struct buffer_head *bh; 197 198 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) { 199 *bhp = NULL; 200 return -EIO; 201 } 202 203 *bhp = bh = gfs2_getbuf(gl, blkno, CREATE); 204 205 lock_buffer(bh); 206 if (buffer_uptodate(bh)) { 207 unlock_buffer(bh); 208 return 0; 209 } 210 bh->b_end_io = end_buffer_read_sync; 211 get_bh(bh); 212 submit_bh(READ_SYNC | REQ_META | REQ_PRIO, bh); 213 if (!(flags & DIO_WAIT)) 214 return 0; 215 216 wait_on_buffer(bh); 217 if (unlikely(!buffer_uptodate(bh))) { 218 struct gfs2_trans *tr = current->journal_info; 219 if (tr && tr->tr_touched) 220 gfs2_io_error_bh(sdp, bh); 221 brelse(bh); 222 *bhp = NULL; 223 return -EIO; 224 } 225 226 return 0; 227 } 228 229 /** 230 * gfs2_meta_wait - Reread a block from disk 231 * @sdp: the filesystem 232 * @bh: The block to wait for 233 * 234 * Returns: errno 235 */ 236 237 int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh) 238 { 239 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 240 return -EIO; 241 242 wait_on_buffer(bh); 243 244 if (!buffer_uptodate(bh)) { 245 struct gfs2_trans *tr = current->journal_info; 246 if (tr && tr->tr_touched) 247 gfs2_io_error_bh(sdp, bh); 248 return -EIO; 249 } 250 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 251 return -EIO; 252 253 return 0; 254 } 255 256 void gfs2_remove_from_journal(struct buffer_head *bh, struct gfs2_trans *tr, int meta) 257 { 258 struct address_space *mapping = bh->b_page->mapping; 259 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping); 260 struct gfs2_bufdata *bd = bh->b_private; 261 int was_pinned = 0; 262 263 if (test_clear_buffer_pinned(bh)) { 264 trace_gfs2_pin(bd, 0); 265 atomic_dec(&sdp->sd_log_pinned); 266 list_del_init(&bd->bd_list); 267 if (meta) { 268 gfs2_assert_warn(sdp, sdp->sd_log_num_buf); 269 sdp->sd_log_num_buf--; 270 tr->tr_num_buf_rm++; 271 } else { 272 gfs2_assert_warn(sdp, sdp->sd_log_num_databuf); 273 sdp->sd_log_num_databuf--; 274 tr->tr_num_databuf_rm++; 275 } 276 tr->tr_touched = 1; 277 was_pinned = 1; 278 brelse(bh); 279 } 280 if (bd) { 281 spin_lock(&sdp->sd_ail_lock); 282 if (bd->bd_tr) { 283 gfs2_trans_add_revoke(sdp, bd); 284 } else if (was_pinned) { 285 bh->b_private = NULL; 286 kmem_cache_free(gfs2_bufdata_cachep, bd); 287 } 288 spin_unlock(&sdp->sd_ail_lock); 289 } 290 clear_buffer_dirty(bh); 291 clear_buffer_uptodate(bh); 292 } 293 294 /** 295 * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore 296 * @ip: the inode who owns the buffers 297 * @bstart: the first buffer in the run 298 * @blen: the number of buffers in the run 299 * 300 */ 301 302 void gfs2_meta_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen) 303 { 304 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 305 struct buffer_head *bh; 306 307 while (blen) { 308 bh = gfs2_getbuf(ip->i_gl, bstart, NO_CREATE); 309 if (bh) { 310 lock_buffer(bh); 311 gfs2_log_lock(sdp); 312 gfs2_remove_from_journal(bh, current->journal_info, 1); 313 gfs2_log_unlock(sdp); 314 unlock_buffer(bh); 315 brelse(bh); 316 } 317 318 bstart++; 319 blen--; 320 } 321 } 322 323 /** 324 * gfs2_meta_indirect_buffer - Get a metadata buffer 325 * @ip: The GFS2 inode 326 * @height: The level of this buf in the metadata (indir addr) tree (if any) 327 * @num: The block number (device relative) of the buffer 328 * @bhp: the buffer is returned here 329 * 330 * Returns: errno 331 */ 332 333 int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, u64 num, 334 struct buffer_head **bhp) 335 { 336 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 337 struct gfs2_glock *gl = ip->i_gl; 338 struct buffer_head *bh; 339 int ret = 0; 340 u32 mtype = height ? GFS2_METATYPE_IN : GFS2_METATYPE_DI; 341 342 ret = gfs2_meta_read(gl, num, DIO_WAIT, &bh); 343 if (ret == 0 && gfs2_metatype_check(sdp, bh, mtype)) { 344 brelse(bh); 345 ret = -EIO; 346 } 347 *bhp = bh; 348 return ret; 349 } 350 351 /** 352 * gfs2_meta_ra - start readahead on an extent of a file 353 * @gl: the glock the blocks belong to 354 * @dblock: the starting disk block 355 * @extlen: the number of blocks in the extent 356 * 357 * returns: the first buffer in the extent 358 */ 359 360 struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen) 361 { 362 struct gfs2_sbd *sdp = gl->gl_sbd; 363 struct buffer_head *first_bh, *bh; 364 u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >> 365 sdp->sd_sb.sb_bsize_shift; 366 367 BUG_ON(!extlen); 368 369 if (max_ra < 1) 370 max_ra = 1; 371 if (extlen > max_ra) 372 extlen = max_ra; 373 374 first_bh = gfs2_getbuf(gl, dblock, CREATE); 375 376 if (buffer_uptodate(first_bh)) 377 goto out; 378 if (!buffer_locked(first_bh)) 379 ll_rw_block(READ_SYNC | REQ_META, 1, &first_bh); 380 381 dblock++; 382 extlen--; 383 384 while (extlen) { 385 bh = gfs2_getbuf(gl, dblock, CREATE); 386 387 if (!buffer_uptodate(bh) && !buffer_locked(bh)) 388 ll_rw_block(READA | REQ_META, 1, &bh); 389 brelse(bh); 390 dblock++; 391 extlen--; 392 if (!buffer_locked(first_bh) && buffer_uptodate(first_bh)) 393 goto out; 394 } 395 396 wait_on_buffer(first_bh); 397 out: 398 return first_bh; 399 } 400 401