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; 193 194 bio_for_each_segment_all(bvec, bio, i) { 195 struct page *page = bvec->bv_page; 196 struct buffer_head *bh = page_buffers(page); 197 unsigned int len = bvec->bv_len; 198 199 while (bh_offset(bh) < bvec->bv_offset) 200 bh = bh->b_this_page; 201 do { 202 struct buffer_head *next = bh->b_this_page; 203 len -= bh->b_size; 204 bh->b_end_io(bh, !bio->bi_status); 205 bh = next; 206 } while (bh && len); 207 } 208 bio_put(bio); 209 } 210 211 /* 212 * Submit several consecutive buffer head I/O requests as a single bio I/O 213 * request. (See submit_bh_wbc.) 214 */ 215 static void gfs2_submit_bhs(int op, int op_flags, struct buffer_head *bhs[], 216 int num) 217 { 218 while (num > 0) { 219 struct buffer_head *bh = *bhs; 220 struct bio *bio; 221 222 bio = bio_alloc(GFP_NOIO, num); 223 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); 224 bio->bi_bdev = bh->b_bdev; 225 while (num > 0) { 226 bh = *bhs; 227 if (!bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh))) { 228 BUG_ON(bio->bi_iter.bi_size == 0); 229 break; 230 } 231 bhs++; 232 num--; 233 } 234 bio->bi_end_io = gfs2_meta_read_endio; 235 bio_set_op_attrs(bio, op, op_flags); 236 submit_bio(bio); 237 } 238 } 239 240 /** 241 * gfs2_meta_read - Read a block from disk 242 * @gl: The glock covering the block 243 * @blkno: The block number 244 * @flags: flags 245 * @bhp: the place where the buffer is returned (NULL on failure) 246 * 247 * Returns: errno 248 */ 249 250 int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags, 251 int rahead, struct buffer_head **bhp) 252 { 253 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; 254 struct buffer_head *bh, *bhs[2]; 255 int num = 0; 256 257 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) { 258 *bhp = NULL; 259 return -EIO; 260 } 261 262 *bhp = bh = gfs2_getbuf(gl, blkno, CREATE); 263 264 lock_buffer(bh); 265 if (buffer_uptodate(bh)) { 266 unlock_buffer(bh); 267 flags &= ~DIO_WAIT; 268 } else { 269 bh->b_end_io = end_buffer_read_sync; 270 get_bh(bh); 271 bhs[num++] = bh; 272 } 273 274 if (rahead) { 275 bh = gfs2_getbuf(gl, blkno + 1, CREATE); 276 277 lock_buffer(bh); 278 if (buffer_uptodate(bh)) { 279 unlock_buffer(bh); 280 brelse(bh); 281 } else { 282 bh->b_end_io = end_buffer_read_sync; 283 bhs[num++] = bh; 284 } 285 } 286 287 gfs2_submit_bhs(REQ_OP_READ, REQ_META | REQ_PRIO, bhs, num); 288 if (!(flags & DIO_WAIT)) 289 return 0; 290 291 bh = *bhp; 292 wait_on_buffer(bh); 293 if (unlikely(!buffer_uptodate(bh))) { 294 struct gfs2_trans *tr = current->journal_info; 295 if (tr && test_bit(TR_TOUCHED, &tr->tr_flags)) 296 gfs2_io_error_bh(sdp, bh); 297 brelse(bh); 298 *bhp = NULL; 299 return -EIO; 300 } 301 302 return 0; 303 } 304 305 /** 306 * gfs2_meta_wait - Reread a block from disk 307 * @sdp: the filesystem 308 * @bh: The block to wait for 309 * 310 * Returns: errno 311 */ 312 313 int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh) 314 { 315 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 316 return -EIO; 317 318 wait_on_buffer(bh); 319 320 if (!buffer_uptodate(bh)) { 321 struct gfs2_trans *tr = current->journal_info; 322 if (tr && test_bit(TR_TOUCHED, &tr->tr_flags)) 323 gfs2_io_error_bh(sdp, bh); 324 return -EIO; 325 } 326 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 327 return -EIO; 328 329 return 0; 330 } 331 332 void gfs2_remove_from_journal(struct buffer_head *bh, int meta) 333 { 334 struct address_space *mapping = bh->b_page->mapping; 335 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping); 336 struct gfs2_bufdata *bd = bh->b_private; 337 struct gfs2_trans *tr = current->journal_info; 338 int was_pinned = 0; 339 340 if (test_clear_buffer_pinned(bh)) { 341 trace_gfs2_pin(bd, 0); 342 atomic_dec(&sdp->sd_log_pinned); 343 list_del_init(&bd->bd_list); 344 if (meta == REMOVE_META) 345 tr->tr_num_buf_rm++; 346 else 347 tr->tr_num_databuf_rm++; 348 set_bit(TR_TOUCHED, &tr->tr_flags); 349 was_pinned = 1; 350 brelse(bh); 351 } 352 if (bd) { 353 spin_lock(&sdp->sd_ail_lock); 354 if (bd->bd_tr) { 355 gfs2_trans_add_revoke(sdp, bd); 356 } else if (was_pinned) { 357 bh->b_private = NULL; 358 kmem_cache_free(gfs2_bufdata_cachep, bd); 359 } 360 spin_unlock(&sdp->sd_ail_lock); 361 } 362 clear_buffer_dirty(bh); 363 clear_buffer_uptodate(bh); 364 } 365 366 /** 367 * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore 368 * @ip: the inode who owns the buffers 369 * @bstart: the first buffer in the run 370 * @blen: the number of buffers in the run 371 * 372 */ 373 374 void gfs2_meta_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen) 375 { 376 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 377 struct buffer_head *bh; 378 379 while (blen) { 380 bh = gfs2_getbuf(ip->i_gl, bstart, NO_CREATE); 381 if (bh) { 382 lock_buffer(bh); 383 gfs2_log_lock(sdp); 384 gfs2_remove_from_journal(bh, REMOVE_META); 385 gfs2_log_unlock(sdp); 386 unlock_buffer(bh); 387 brelse(bh); 388 } 389 390 bstart++; 391 blen--; 392 } 393 } 394 395 /** 396 * gfs2_meta_indirect_buffer - Get a metadata buffer 397 * @ip: The GFS2 inode 398 * @height: The level of this buf in the metadata (indir addr) tree (if any) 399 * @num: The block number (device relative) of the buffer 400 * @bhp: the buffer is returned here 401 * 402 * Returns: errno 403 */ 404 405 int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, u64 num, 406 struct buffer_head **bhp) 407 { 408 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 409 struct gfs2_glock *gl = ip->i_gl; 410 struct buffer_head *bh; 411 int ret = 0; 412 u32 mtype = height ? GFS2_METATYPE_IN : GFS2_METATYPE_DI; 413 int rahead = 0; 414 415 if (num == ip->i_no_addr) 416 rahead = ip->i_rahead; 417 418 ret = gfs2_meta_read(gl, num, DIO_WAIT, rahead, &bh); 419 if (ret == 0 && gfs2_metatype_check(sdp, bh, mtype)) { 420 brelse(bh); 421 ret = -EIO; 422 } 423 *bhp = bh; 424 return ret; 425 } 426 427 /** 428 * gfs2_meta_ra - start readahead on an extent of a file 429 * @gl: the glock the blocks belong to 430 * @dblock: the starting disk block 431 * @extlen: the number of blocks in the extent 432 * 433 * returns: the first buffer in the extent 434 */ 435 436 struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen) 437 { 438 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd; 439 struct buffer_head *first_bh, *bh; 440 u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >> 441 sdp->sd_sb.sb_bsize_shift; 442 443 BUG_ON(!extlen); 444 445 if (max_ra < 1) 446 max_ra = 1; 447 if (extlen > max_ra) 448 extlen = max_ra; 449 450 first_bh = gfs2_getbuf(gl, dblock, CREATE); 451 452 if (buffer_uptodate(first_bh)) 453 goto out; 454 if (!buffer_locked(first_bh)) 455 ll_rw_block(REQ_OP_READ, REQ_META, 1, &first_bh); 456 457 dblock++; 458 extlen--; 459 460 while (extlen) { 461 bh = gfs2_getbuf(gl, dblock, CREATE); 462 463 if (!buffer_uptodate(bh) && !buffer_locked(bh)) 464 ll_rw_block(REQ_OP_READ, REQ_RAHEAD | REQ_META, 1, &bh); 465 brelse(bh); 466 dblock++; 467 extlen--; 468 if (!buffer_locked(first_bh) && buffer_uptodate(first_bh)) 469 goto out; 470 } 471 472 wait_on_buffer(first_bh); 473 out: 474 return first_bh; 475 } 476 477