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_flags = REQ_META | REQ_PRIO | 41 (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : 0); 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(REQ_OP_WRITE, write_flags, 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 const struct address_space_operations gfs2_rgrp_aops = { 101 .writepage = gfs2_aspace_writepage, 102 .releasepage = gfs2_releasepage, 103 }; 104 105 /** 106 * gfs2_getbuf - Get a buffer with a given address space 107 * @gl: the glock 108 * @blkno: the block number (filesystem scope) 109 * @create: 1 if the buffer should be created 110 * 111 * Returns: the buffer 112 */ 113 114 struct buffer_head *gfs2_getbuf(struct gfs2_glock *gl, u64 blkno, int create) 115 { 116 struct address_space *mapping = gfs2_glock2aspace(gl); 117 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; 118 struct page *page; 119 struct buffer_head *bh; 120 unsigned int shift; 121 unsigned long index; 122 unsigned int bufnum; 123 124 if (mapping == NULL) 125 mapping = &sdp->sd_aspace; 126 127 shift = PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift; 128 index = blkno >> shift; /* convert block to page */ 129 bufnum = blkno - (index << shift); /* block buf index within page */ 130 131 if (create) { 132 for (;;) { 133 page = grab_cache_page(mapping, index); 134 if (page) 135 break; 136 yield(); 137 } 138 } else { 139 page = find_get_page_flags(mapping, index, 140 FGP_LOCK|FGP_ACCESSED); 141 if (!page) 142 return NULL; 143 } 144 145 if (!page_has_buffers(page)) 146 create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0); 147 148 /* Locate header for our buffer within our page */ 149 for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page) 150 /* Do nothing */; 151 get_bh(bh); 152 153 if (!buffer_mapped(bh)) 154 map_bh(bh, sdp->sd_vfs, blkno); 155 156 unlock_page(page); 157 put_page(page); 158 159 return bh; 160 } 161 162 static void meta_prep_new(struct buffer_head *bh) 163 { 164 struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data; 165 166 lock_buffer(bh); 167 clear_buffer_dirty(bh); 168 set_buffer_uptodate(bh); 169 unlock_buffer(bh); 170 171 mh->mh_magic = cpu_to_be32(GFS2_MAGIC); 172 } 173 174 /** 175 * gfs2_meta_new - Get a block 176 * @gl: The glock associated with this block 177 * @blkno: The block number 178 * 179 * Returns: The buffer 180 */ 181 182 struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno) 183 { 184 struct buffer_head *bh; 185 bh = gfs2_getbuf(gl, blkno, CREATE); 186 meta_prep_new(bh); 187 return bh; 188 } 189 190 static void gfs2_meta_read_endio(struct bio *bio) 191 { 192 struct bio_vec *bvec; 193 int i; 194 195 bio_for_each_segment_all(bvec, bio, i) { 196 struct page *page = bvec->bv_page; 197 struct buffer_head *bh = page_buffers(page); 198 unsigned int len = bvec->bv_len; 199 200 while (bh_offset(bh) < bvec->bv_offset) 201 bh = bh->b_this_page; 202 do { 203 struct buffer_head *next = bh->b_this_page; 204 len -= bh->b_size; 205 bh->b_end_io(bh, !bio->bi_error); 206 bh = next; 207 } while (bh && len); 208 } 209 bio_put(bio); 210 } 211 212 /* 213 * Submit several consecutive buffer head I/O requests as a single bio I/O 214 * request. (See submit_bh_wbc.) 215 */ 216 static void gfs2_submit_bhs(int op, int op_flags, struct buffer_head *bhs[], 217 int num) 218 { 219 struct buffer_head *bh = bhs[0]; 220 struct bio *bio; 221 int i; 222 223 if (!num) 224 return; 225 226 bio = bio_alloc(GFP_NOIO, num); 227 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); 228 bio->bi_bdev = bh->b_bdev; 229 for (i = 0; i < num; i++) { 230 bh = bhs[i]; 231 bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); 232 } 233 bio->bi_end_io = gfs2_meta_read_endio; 234 bio_set_op_attrs(bio, op, op_flags); 235 submit_bio(bio); 236 } 237 238 /** 239 * gfs2_meta_read - Read a block from disk 240 * @gl: The glock covering the block 241 * @blkno: The block number 242 * @flags: flags 243 * @bhp: the place where the buffer is returned (NULL on failure) 244 * 245 * Returns: errno 246 */ 247 248 int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags, 249 int rahead, struct buffer_head **bhp) 250 { 251 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; 252 struct buffer_head *bh, *bhs[2]; 253 int num = 0; 254 255 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) { 256 *bhp = NULL; 257 return -EIO; 258 } 259 260 *bhp = bh = gfs2_getbuf(gl, blkno, CREATE); 261 262 lock_buffer(bh); 263 if (buffer_uptodate(bh)) { 264 unlock_buffer(bh); 265 flags &= ~DIO_WAIT; 266 } else { 267 bh->b_end_io = end_buffer_read_sync; 268 get_bh(bh); 269 bhs[num++] = bh; 270 } 271 272 if (rahead) { 273 bh = gfs2_getbuf(gl, blkno + 1, CREATE); 274 275 lock_buffer(bh); 276 if (buffer_uptodate(bh)) { 277 unlock_buffer(bh); 278 brelse(bh); 279 } else { 280 bh->b_end_io = end_buffer_read_sync; 281 bhs[num++] = bh; 282 } 283 } 284 285 gfs2_submit_bhs(REQ_OP_READ, READ_SYNC | REQ_META | REQ_PRIO, bhs, num); 286 if (!(flags & DIO_WAIT)) 287 return 0; 288 289 bh = *bhp; 290 wait_on_buffer(bh); 291 if (unlikely(!buffer_uptodate(bh))) { 292 struct gfs2_trans *tr = current->journal_info; 293 if (tr && tr->tr_touched) 294 gfs2_io_error_bh(sdp, bh); 295 brelse(bh); 296 *bhp = NULL; 297 return -EIO; 298 } 299 300 return 0; 301 } 302 303 /** 304 * gfs2_meta_wait - Reread a block from disk 305 * @sdp: the filesystem 306 * @bh: The block to wait for 307 * 308 * Returns: errno 309 */ 310 311 int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh) 312 { 313 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 314 return -EIO; 315 316 wait_on_buffer(bh); 317 318 if (!buffer_uptodate(bh)) { 319 struct gfs2_trans *tr = current->journal_info; 320 if (tr && tr->tr_touched) 321 gfs2_io_error_bh(sdp, bh); 322 return -EIO; 323 } 324 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 325 return -EIO; 326 327 return 0; 328 } 329 330 void gfs2_remove_from_journal(struct buffer_head *bh, int meta) 331 { 332 struct address_space *mapping = bh->b_page->mapping; 333 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping); 334 struct gfs2_bufdata *bd = bh->b_private; 335 struct gfs2_trans *tr = current->journal_info; 336 int was_pinned = 0; 337 338 if (test_clear_buffer_pinned(bh)) { 339 trace_gfs2_pin(bd, 0); 340 atomic_dec(&sdp->sd_log_pinned); 341 list_del_init(&bd->bd_list); 342 if (meta == REMOVE_META) 343 tr->tr_num_buf_rm++; 344 else 345 tr->tr_num_databuf_rm++; 346 tr->tr_touched = 1; 347 was_pinned = 1; 348 brelse(bh); 349 } 350 if (bd) { 351 spin_lock(&sdp->sd_ail_lock); 352 if (bd->bd_tr) { 353 gfs2_trans_add_revoke(sdp, bd); 354 } else if (was_pinned) { 355 bh->b_private = NULL; 356 kmem_cache_free(gfs2_bufdata_cachep, bd); 357 } 358 spin_unlock(&sdp->sd_ail_lock); 359 } 360 clear_buffer_dirty(bh); 361 clear_buffer_uptodate(bh); 362 } 363 364 /** 365 * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore 366 * @ip: the inode who owns the buffers 367 * @bstart: the first buffer in the run 368 * @blen: the number of buffers in the run 369 * 370 */ 371 372 void gfs2_meta_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen) 373 { 374 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 375 struct buffer_head *bh; 376 377 while (blen) { 378 bh = gfs2_getbuf(ip->i_gl, bstart, NO_CREATE); 379 if (bh) { 380 lock_buffer(bh); 381 gfs2_log_lock(sdp); 382 gfs2_remove_from_journal(bh, REMOVE_META); 383 gfs2_log_unlock(sdp); 384 unlock_buffer(bh); 385 brelse(bh); 386 } 387 388 bstart++; 389 blen--; 390 } 391 } 392 393 /** 394 * gfs2_meta_indirect_buffer - Get a metadata buffer 395 * @ip: The GFS2 inode 396 * @height: The level of this buf in the metadata (indir addr) tree (if any) 397 * @num: The block number (device relative) of the buffer 398 * @bhp: the buffer is returned here 399 * 400 * Returns: errno 401 */ 402 403 int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, u64 num, 404 struct buffer_head **bhp) 405 { 406 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 407 struct gfs2_glock *gl = ip->i_gl; 408 struct buffer_head *bh; 409 int ret = 0; 410 u32 mtype = height ? GFS2_METATYPE_IN : GFS2_METATYPE_DI; 411 int rahead = 0; 412 413 if (num == ip->i_no_addr) 414 rahead = ip->i_rahead; 415 416 ret = gfs2_meta_read(gl, num, DIO_WAIT, rahead, &bh); 417 if (ret == 0 && gfs2_metatype_check(sdp, bh, mtype)) { 418 brelse(bh); 419 ret = -EIO; 420 } 421 *bhp = bh; 422 return ret; 423 } 424 425 /** 426 * gfs2_meta_ra - start readahead on an extent of a file 427 * @gl: the glock the blocks belong to 428 * @dblock: the starting disk block 429 * @extlen: the number of blocks in the extent 430 * 431 * returns: the first buffer in the extent 432 */ 433 434 struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen) 435 { 436 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; 437 struct buffer_head *first_bh, *bh; 438 u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >> 439 sdp->sd_sb.sb_bsize_shift; 440 441 BUG_ON(!extlen); 442 443 if (max_ra < 1) 444 max_ra = 1; 445 if (extlen > max_ra) 446 extlen = max_ra; 447 448 first_bh = gfs2_getbuf(gl, dblock, CREATE); 449 450 if (buffer_uptodate(first_bh)) 451 goto out; 452 if (!buffer_locked(first_bh)) 453 ll_rw_block(REQ_OP_READ, READ_SYNC | REQ_META, 1, &first_bh); 454 455 dblock++; 456 extlen--; 457 458 while (extlen) { 459 bh = gfs2_getbuf(gl, dblock, CREATE); 460 461 if (!buffer_uptodate(bh) && !buffer_locked(bh)) 462 ll_rw_block(REQ_OP_READ, REQ_RAHEAD | REQ_META, 1, &bh); 463 brelse(bh); 464 dblock++; 465 extlen--; 466 if (!buffer_locked(first_bh) && buffer_uptodate(first_bh)) 467 goto out; 468 } 469 470 wait_on_buffer(first_bh); 471 out: 472 return first_bh; 473 } 474 475