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 | wbc_to_write_flags(wbc); 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 flusher thread 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(REQ_OP_WRITE, write_flags, 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 const struct address_space_operations gfs2_rgrp_aops = { 100 .writepage = gfs2_aspace_writepage, 101 .releasepage = gfs2_releasepage, 102 }; 103 104 /** 105 * gfs2_getbuf - Get a buffer with a given address space 106 * @gl: the glock 107 * @blkno: the block number (filesystem scope) 108 * @create: 1 if the buffer should be created 109 * 110 * Returns: the buffer 111 */ 112 113 struct buffer_head *gfs2_getbuf(struct gfs2_glock *gl, u64 blkno, int create) 114 { 115 struct address_space *mapping = gfs2_glock2aspace(gl); 116 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; 117 struct page *page; 118 struct buffer_head *bh; 119 unsigned int shift; 120 unsigned long index; 121 unsigned int bufnum; 122 123 if (mapping == NULL) 124 mapping = &sdp->sd_aspace; 125 126 shift = PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift; 127 index = blkno >> shift; /* convert block to page */ 128 bufnum = blkno - (index << shift); /* block buf index within page */ 129 130 if (create) { 131 for (;;) { 132 page = grab_cache_page(mapping, index); 133 if (page) 134 break; 135 yield(); 136 } 137 } else { 138 page = find_get_page_flags(mapping, index, 139 FGP_LOCK|FGP_ACCESSED); 140 if (!page) 141 return NULL; 142 } 143 144 if (!page_has_buffers(page)) 145 create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0); 146 147 /* Locate header for our buffer within our page */ 148 for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page) 149 /* Do nothing */; 150 get_bh(bh); 151 152 if (!buffer_mapped(bh)) 153 map_bh(bh, sdp->sd_vfs, blkno); 154 155 unlock_page(page); 156 put_page(page); 157 158 return bh; 159 } 160 161 static void meta_prep_new(struct buffer_head *bh) 162 { 163 struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data; 164 165 lock_buffer(bh); 166 clear_buffer_dirty(bh); 167 set_buffer_uptodate(bh); 168 unlock_buffer(bh); 169 170 mh->mh_magic = cpu_to_be32(GFS2_MAGIC); 171 } 172 173 /** 174 * gfs2_meta_new - Get a block 175 * @gl: The glock associated with this block 176 * @blkno: The block number 177 * 178 * Returns: The buffer 179 */ 180 181 struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno) 182 { 183 struct buffer_head *bh; 184 bh = gfs2_getbuf(gl, blkno, CREATE); 185 meta_prep_new(bh); 186 return bh; 187 } 188 189 static void gfs2_meta_read_endio(struct bio *bio) 190 { 191 struct bio_vec *bvec; 192 int i; 193 struct bvec_iter_all iter_all; 194 195 bio_for_each_segment_all(bvec, bio, i, iter_all) { 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_status); 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 while (num > 0) { 220 struct buffer_head *bh = *bhs; 221 struct bio *bio; 222 223 bio = bio_alloc(GFP_NOIO, num); 224 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); 225 bio_set_dev(bio, bh->b_bdev); 226 while (num > 0) { 227 bh = *bhs; 228 if (!bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh))) { 229 BUG_ON(bio->bi_iter.bi_size == 0); 230 break; 231 } 232 bhs++; 233 num--; 234 } 235 bio->bi_end_io = gfs2_meta_read_endio; 236 bio_set_op_attrs(bio, op, op_flags); 237 submit_bio(bio); 238 } 239 } 240 241 /** 242 * gfs2_meta_read - Read a block from disk 243 * @gl: The glock covering the block 244 * @blkno: The block number 245 * @flags: flags 246 * @bhp: the place where the buffer is returned (NULL on failure) 247 * 248 * Returns: errno 249 */ 250 251 int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags, 252 int rahead, struct buffer_head **bhp) 253 { 254 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; 255 struct buffer_head *bh, *bhs[2]; 256 int num = 0; 257 258 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) { 259 *bhp = NULL; 260 return -EIO; 261 } 262 263 *bhp = bh = gfs2_getbuf(gl, blkno, CREATE); 264 265 lock_buffer(bh); 266 if (buffer_uptodate(bh)) { 267 unlock_buffer(bh); 268 flags &= ~DIO_WAIT; 269 } else { 270 bh->b_end_io = end_buffer_read_sync; 271 get_bh(bh); 272 bhs[num++] = bh; 273 } 274 275 if (rahead) { 276 bh = gfs2_getbuf(gl, blkno + 1, CREATE); 277 278 lock_buffer(bh); 279 if (buffer_uptodate(bh)) { 280 unlock_buffer(bh); 281 brelse(bh); 282 } else { 283 bh->b_end_io = end_buffer_read_sync; 284 bhs[num++] = bh; 285 } 286 } 287 288 gfs2_submit_bhs(REQ_OP_READ, REQ_META | REQ_PRIO, bhs, num); 289 if (!(flags & DIO_WAIT)) 290 return 0; 291 292 bh = *bhp; 293 wait_on_buffer(bh); 294 if (unlikely(!buffer_uptodate(bh))) { 295 struct gfs2_trans *tr = current->journal_info; 296 if (tr && test_bit(TR_TOUCHED, &tr->tr_flags)) 297 gfs2_io_error_bh_wd(sdp, bh); 298 brelse(bh); 299 *bhp = NULL; 300 return -EIO; 301 } 302 303 return 0; 304 } 305 306 /** 307 * gfs2_meta_wait - Reread a block from disk 308 * @sdp: the filesystem 309 * @bh: The block to wait for 310 * 311 * Returns: errno 312 */ 313 314 int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh) 315 { 316 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 317 return -EIO; 318 319 wait_on_buffer(bh); 320 321 if (!buffer_uptodate(bh)) { 322 struct gfs2_trans *tr = current->journal_info; 323 if (tr && test_bit(TR_TOUCHED, &tr->tr_flags)) 324 gfs2_io_error_bh_wd(sdp, bh); 325 return -EIO; 326 } 327 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 328 return -EIO; 329 330 return 0; 331 } 332 333 void gfs2_remove_from_journal(struct buffer_head *bh, int meta) 334 { 335 struct address_space *mapping = bh->b_page->mapping; 336 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping); 337 struct gfs2_bufdata *bd = bh->b_private; 338 struct gfs2_trans *tr = current->journal_info; 339 int was_pinned = 0; 340 341 if (test_clear_buffer_pinned(bh)) { 342 trace_gfs2_pin(bd, 0); 343 atomic_dec(&sdp->sd_log_pinned); 344 list_del_init(&bd->bd_list); 345 if (meta == REMOVE_META) 346 tr->tr_num_buf_rm++; 347 else 348 tr->tr_num_databuf_rm++; 349 set_bit(TR_TOUCHED, &tr->tr_flags); 350 was_pinned = 1; 351 brelse(bh); 352 } 353 if (bd) { 354 spin_lock(&sdp->sd_ail_lock); 355 if (bd->bd_tr) { 356 gfs2_trans_add_revoke(sdp, bd); 357 } else if (was_pinned) { 358 bh->b_private = NULL; 359 kmem_cache_free(gfs2_bufdata_cachep, bd); 360 } 361 spin_unlock(&sdp->sd_ail_lock); 362 } 363 clear_buffer_dirty(bh); 364 clear_buffer_uptodate(bh); 365 } 366 367 /** 368 * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore 369 * @ip: the inode who owns the buffers 370 * @bstart: the first buffer in the run 371 * @blen: the number of buffers in the run 372 * 373 */ 374 375 void gfs2_meta_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen) 376 { 377 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 378 struct buffer_head *bh; 379 380 while (blen) { 381 bh = gfs2_getbuf(ip->i_gl, bstart, NO_CREATE); 382 if (bh) { 383 lock_buffer(bh); 384 gfs2_log_lock(sdp); 385 gfs2_remove_from_journal(bh, REMOVE_META); 386 gfs2_log_unlock(sdp); 387 unlock_buffer(bh); 388 brelse(bh); 389 } 390 391 bstart++; 392 blen--; 393 } 394 } 395 396 /** 397 * gfs2_meta_indirect_buffer - Get a metadata buffer 398 * @ip: The GFS2 inode 399 * @height: The level of this buf in the metadata (indir addr) tree (if any) 400 * @num: The block number (device relative) of the buffer 401 * @bhp: the buffer is returned here 402 * 403 * Returns: errno 404 */ 405 406 int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, u64 num, 407 struct buffer_head **bhp) 408 { 409 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 410 struct gfs2_glock *gl = ip->i_gl; 411 struct buffer_head *bh; 412 int ret = 0; 413 u32 mtype = height ? GFS2_METATYPE_IN : GFS2_METATYPE_DI; 414 int rahead = 0; 415 416 if (num == ip->i_no_addr) 417 rahead = ip->i_rahead; 418 419 ret = gfs2_meta_read(gl, num, DIO_WAIT, rahead, &bh); 420 if (ret == 0 && gfs2_metatype_check(sdp, bh, mtype)) { 421 brelse(bh); 422 ret = -EIO; 423 } else { 424 *bhp = bh; 425 } 426 return ret; 427 } 428 429 /** 430 * gfs2_meta_ra - start readahead on an extent of a file 431 * @gl: the glock the blocks belong to 432 * @dblock: the starting disk block 433 * @extlen: the number of blocks in the extent 434 * 435 * returns: the first buffer in the extent 436 */ 437 438 struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen) 439 { 440 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; 441 struct buffer_head *first_bh, *bh; 442 u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >> 443 sdp->sd_sb.sb_bsize_shift; 444 445 BUG_ON(!extlen); 446 447 if (max_ra < 1) 448 max_ra = 1; 449 if (extlen > max_ra) 450 extlen = max_ra; 451 452 first_bh = gfs2_getbuf(gl, dblock, CREATE); 453 454 if (buffer_uptodate(first_bh)) 455 goto out; 456 if (!buffer_locked(first_bh)) 457 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &first_bh); 458 459 dblock++; 460 extlen--; 461 462 while (extlen) { 463 bh = gfs2_getbuf(gl, dblock, CREATE); 464 465 if (!buffer_uptodate(bh) && !buffer_locked(bh)) 466 ll_rw_block(REQ_OP_READ, 467 REQ_RAHEAD | REQ_META | REQ_PRIO, 468 1, &bh); 469 brelse(bh); 470 dblock++; 471 extlen--; 472 if (!buffer_locked(first_bh) && buffer_uptodate(first_bh)) 473 goto out; 474 } 475 476 wait_on_buffer(first_bh); 477 out: 478 return first_bh; 479 } 480 481