Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1 | /* -*- mode: c; c-basic-offset: 8; -*- |
| 2 | * vim: noexpandtab sw=8 ts=8 sts=0: |
| 3 | * |
| 4 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU General Public |
| 8 | * License as published by the Free Software Foundation; either |
| 9 | * version 2 of the License, or (at your option) any later version. |
| 10 | * |
| 11 | * This program is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 14 | * General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU General Public |
| 17 | * License along with this program; if not, write to the |
| 18 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, |
| 19 | * Boston, MA 021110-1307, USA. |
| 20 | */ |
| 21 | |
| 22 | #include <linux/fs.h> |
| 23 | #include <linux/slab.h> |
| 24 | #include <linux/highmem.h> |
| 25 | #include <linux/pagemap.h> |
| 26 | #include <asm/byteorder.h> |
| 27 | |
| 28 | #define MLOG_MASK_PREFIX ML_FILE_IO |
| 29 | #include <cluster/masklog.h> |
| 30 | |
| 31 | #include "ocfs2.h" |
| 32 | |
| 33 | #include "alloc.h" |
| 34 | #include "aops.h" |
| 35 | #include "dlmglue.h" |
| 36 | #include "extent_map.h" |
| 37 | #include "file.h" |
| 38 | #include "inode.h" |
| 39 | #include "journal.h" |
| 40 | #include "super.h" |
| 41 | #include "symlink.h" |
| 42 | |
| 43 | #include "buffer_head_io.h" |
| 44 | |
| 45 | static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock, |
| 46 | struct buffer_head *bh_result, int create) |
| 47 | { |
| 48 | int err = -EIO; |
| 49 | int status; |
| 50 | struct ocfs2_dinode *fe = NULL; |
| 51 | struct buffer_head *bh = NULL; |
| 52 | struct buffer_head *buffer_cache_bh = NULL; |
| 53 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 54 | void *kaddr; |
| 55 | |
| 56 | mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode, |
| 57 | (unsigned long long)iblock, bh_result, create); |
| 58 | |
| 59 | BUG_ON(ocfs2_inode_is_fast_symlink(inode)); |
| 60 | |
| 61 | if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) { |
| 62 | mlog(ML_ERROR, "block offset > PATH_MAX: %llu", |
| 63 | (unsigned long long)iblock); |
| 64 | goto bail; |
| 65 | } |
| 66 | |
| 67 | status = ocfs2_read_block(OCFS2_SB(inode->i_sb), |
| 68 | OCFS2_I(inode)->ip_blkno, |
| 69 | &bh, OCFS2_BH_CACHED, inode); |
| 70 | if (status < 0) { |
| 71 | mlog_errno(status); |
| 72 | goto bail; |
| 73 | } |
| 74 | fe = (struct ocfs2_dinode *) bh->b_data; |
| 75 | |
| 76 | if (!OCFS2_IS_VALID_DINODE(fe)) { |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 77 | mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n", |
| 78 | (unsigned long long)fe->i_blkno, 7, fe->i_signature); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 79 | goto bail; |
| 80 | } |
| 81 | |
| 82 | if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb, |
| 83 | le32_to_cpu(fe->i_clusters))) { |
| 84 | mlog(ML_ERROR, "block offset is outside the allocated size: " |
| 85 | "%llu\n", (unsigned long long)iblock); |
| 86 | goto bail; |
| 87 | } |
| 88 | |
| 89 | /* We don't use the page cache to create symlink data, so if |
| 90 | * need be, copy it over from the buffer cache. */ |
| 91 | if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) { |
| 92 | u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + |
| 93 | iblock; |
| 94 | buffer_cache_bh = sb_getblk(osb->sb, blkno); |
| 95 | if (!buffer_cache_bh) { |
| 96 | mlog(ML_ERROR, "couldn't getblock for symlink!\n"); |
| 97 | goto bail; |
| 98 | } |
| 99 | |
| 100 | /* we haven't locked out transactions, so a commit |
| 101 | * could've happened. Since we've got a reference on |
| 102 | * the bh, even if it commits while we're doing the |
| 103 | * copy, the data is still good. */ |
| 104 | if (buffer_jbd(buffer_cache_bh) |
| 105 | && ocfs2_inode_is_new(inode)) { |
| 106 | kaddr = kmap_atomic(bh_result->b_page, KM_USER0); |
| 107 | if (!kaddr) { |
| 108 | mlog(ML_ERROR, "couldn't kmap!\n"); |
| 109 | goto bail; |
| 110 | } |
| 111 | memcpy(kaddr + (bh_result->b_size * iblock), |
| 112 | buffer_cache_bh->b_data, |
| 113 | bh_result->b_size); |
| 114 | kunmap_atomic(kaddr, KM_USER0); |
| 115 | set_buffer_uptodate(bh_result); |
| 116 | } |
| 117 | brelse(buffer_cache_bh); |
| 118 | } |
| 119 | |
| 120 | map_bh(bh_result, inode->i_sb, |
| 121 | le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock); |
| 122 | |
| 123 | err = 0; |
| 124 | |
| 125 | bail: |
| 126 | if (bh) |
| 127 | brelse(bh); |
| 128 | |
| 129 | mlog_exit(err); |
| 130 | return err; |
| 131 | } |
| 132 | |
| 133 | static int ocfs2_get_block(struct inode *inode, sector_t iblock, |
| 134 | struct buffer_head *bh_result, int create) |
| 135 | { |
| 136 | int err = 0; |
| 137 | u64 p_blkno, past_eof; |
| 138 | |
| 139 | mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode, |
| 140 | (unsigned long long)iblock, bh_result, create); |
| 141 | |
| 142 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) |
| 143 | mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n", |
| 144 | inode, inode->i_ino); |
| 145 | |
| 146 | if (S_ISLNK(inode->i_mode)) { |
| 147 | /* this always does I/O for some reason. */ |
| 148 | err = ocfs2_symlink_get_block(inode, iblock, bh_result, create); |
| 149 | goto bail; |
| 150 | } |
| 151 | |
| 152 | /* this can happen if another node truncs after our extend! */ |
| 153 | spin_lock(&OCFS2_I(inode)->ip_lock); |
| 154 | if (iblock >= ocfs2_clusters_to_blocks(inode->i_sb, |
| 155 | OCFS2_I(inode)->ip_clusters)) |
| 156 | err = -EIO; |
| 157 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
| 158 | if (err) |
| 159 | goto bail; |
| 160 | |
| 161 | err = ocfs2_extent_map_get_blocks(inode, iblock, 1, &p_blkno, |
| 162 | NULL); |
| 163 | if (err) { |
| 164 | mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, " |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 165 | "%llu, NULL)\n", err, inode, (unsigned long long)iblock, |
| 166 | (unsigned long long)p_blkno); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 167 | goto bail; |
| 168 | } |
| 169 | |
| 170 | map_bh(bh_result, inode->i_sb, p_blkno); |
| 171 | |
| 172 | if (bh_result->b_blocknr == 0) { |
| 173 | err = -EIO; |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 174 | mlog(ML_ERROR, "iblock = %llu p_blkno = %llu blkno=(%llu)\n", |
| 175 | (unsigned long long)iblock, |
| 176 | (unsigned long long)p_blkno, |
| 177 | (unsigned long long)OCFS2_I(inode)->ip_blkno); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 178 | } |
| 179 | |
| 180 | past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); |
Mark Fasheh | b069705 | 2006-03-03 10:24:33 -0800 | [diff] [blame] | 181 | mlog(0, "Inode %lu, past_eof = %llu\n", inode->i_ino, |
| 182 | (unsigned long long)past_eof); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 183 | |
| 184 | if (create && (iblock >= past_eof)) |
| 185 | set_buffer_new(bh_result); |
| 186 | |
| 187 | bail: |
| 188 | if (err < 0) |
| 189 | err = -EIO; |
| 190 | |
| 191 | mlog_exit(err); |
| 192 | return err; |
| 193 | } |
| 194 | |
| 195 | static int ocfs2_readpage(struct file *file, struct page *page) |
| 196 | { |
| 197 | struct inode *inode = page->mapping->host; |
| 198 | loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT; |
| 199 | int ret, unlock = 1; |
| 200 | |
| 201 | mlog_entry("(0x%p, %lu)\n", file, (page ? page->index : 0)); |
| 202 | |
| 203 | ret = ocfs2_meta_lock_with_page(inode, NULL, NULL, 0, page); |
| 204 | if (ret != 0) { |
| 205 | if (ret == AOP_TRUNCATED_PAGE) |
| 206 | unlock = 0; |
| 207 | mlog_errno(ret); |
| 208 | goto out; |
| 209 | } |
| 210 | |
| 211 | down_read(&OCFS2_I(inode)->ip_alloc_sem); |
| 212 | |
| 213 | /* |
| 214 | * i_size might have just been updated as we grabed the meta lock. We |
| 215 | * might now be discovering a truncate that hit on another node. |
| 216 | * block_read_full_page->get_block freaks out if it is asked to read |
| 217 | * beyond the end of a file, so we check here. Callers |
| 218 | * (generic_file_read, fault->nopage) are clever enough to check i_size |
| 219 | * and notice that the page they just read isn't needed. |
| 220 | * |
| 221 | * XXX sys_readahead() seems to get that wrong? |
| 222 | */ |
| 223 | if (start >= i_size_read(inode)) { |
| 224 | char *addr = kmap(page); |
| 225 | memset(addr, 0, PAGE_SIZE); |
| 226 | flush_dcache_page(page); |
| 227 | kunmap(page); |
| 228 | SetPageUptodate(page); |
| 229 | ret = 0; |
| 230 | goto out_alloc; |
| 231 | } |
| 232 | |
| 233 | ret = ocfs2_data_lock_with_page(inode, 0, page); |
| 234 | if (ret != 0) { |
| 235 | if (ret == AOP_TRUNCATED_PAGE) |
| 236 | unlock = 0; |
| 237 | mlog_errno(ret); |
| 238 | goto out_alloc; |
| 239 | } |
| 240 | |
| 241 | ret = block_read_full_page(page, ocfs2_get_block); |
| 242 | unlock = 0; |
| 243 | |
| 244 | ocfs2_data_unlock(inode, 0); |
| 245 | out_alloc: |
| 246 | up_read(&OCFS2_I(inode)->ip_alloc_sem); |
| 247 | ocfs2_meta_unlock(inode, 0); |
| 248 | out: |
| 249 | if (unlock) |
| 250 | unlock_page(page); |
| 251 | mlog_exit(ret); |
| 252 | return ret; |
| 253 | } |
| 254 | |
| 255 | /* Note: Because we don't support holes, our allocation has |
| 256 | * already happened (allocation writes zeros to the file data) |
| 257 | * so we don't have to worry about ordered writes in |
| 258 | * ocfs2_writepage. |
| 259 | * |
| 260 | * ->writepage is called during the process of invalidating the page cache |
| 261 | * during blocked lock processing. It can't block on any cluster locks |
| 262 | * to during block mapping. It's relying on the fact that the block |
| 263 | * mapping can't have disappeared under the dirty pages that it is |
| 264 | * being asked to write back. |
| 265 | */ |
| 266 | static int ocfs2_writepage(struct page *page, struct writeback_control *wbc) |
| 267 | { |
| 268 | int ret; |
| 269 | |
| 270 | mlog_entry("(0x%p)\n", page); |
| 271 | |
| 272 | ret = block_write_full_page(page, ocfs2_get_block, wbc); |
| 273 | |
| 274 | mlog_exit(ret); |
| 275 | |
| 276 | return ret; |
| 277 | } |
| 278 | |
| 279 | /* |
| 280 | * ocfs2_prepare_write() can be an outer-most ocfs2 call when it is called |
| 281 | * from loopback. It must be able to perform its own locking around |
| 282 | * ocfs2_get_block(). |
| 283 | */ |
| 284 | int ocfs2_prepare_write(struct file *file, struct page *page, |
| 285 | unsigned from, unsigned to) |
| 286 | { |
| 287 | struct inode *inode = page->mapping->host; |
| 288 | int ret; |
| 289 | |
| 290 | mlog_entry("(0x%p, 0x%p, %u, %u)\n", file, page, from, to); |
| 291 | |
| 292 | ret = ocfs2_meta_lock_with_page(inode, NULL, NULL, 0, page); |
| 293 | if (ret != 0) { |
| 294 | mlog_errno(ret); |
| 295 | goto out; |
| 296 | } |
| 297 | |
| 298 | down_read(&OCFS2_I(inode)->ip_alloc_sem); |
| 299 | |
| 300 | ret = block_prepare_write(page, from, to, ocfs2_get_block); |
| 301 | |
| 302 | up_read(&OCFS2_I(inode)->ip_alloc_sem); |
| 303 | |
| 304 | ocfs2_meta_unlock(inode, 0); |
| 305 | out: |
| 306 | mlog_exit(ret); |
| 307 | return ret; |
| 308 | } |
| 309 | |
| 310 | /* Taken from ext3. We don't necessarily need the full blown |
| 311 | * functionality yet, but IMHO it's better to cut and paste the whole |
| 312 | * thing so we can avoid introducing our own bugs (and easily pick up |
| 313 | * their fixes when they happen) --Mark */ |
| 314 | static int walk_page_buffers( handle_t *handle, |
| 315 | struct buffer_head *head, |
| 316 | unsigned from, |
| 317 | unsigned to, |
| 318 | int *partial, |
| 319 | int (*fn)( handle_t *handle, |
| 320 | struct buffer_head *bh)) |
| 321 | { |
| 322 | struct buffer_head *bh; |
| 323 | unsigned block_start, block_end; |
| 324 | unsigned blocksize = head->b_size; |
| 325 | int err, ret = 0; |
| 326 | struct buffer_head *next; |
| 327 | |
| 328 | for ( bh = head, block_start = 0; |
| 329 | ret == 0 && (bh != head || !block_start); |
| 330 | block_start = block_end, bh = next) |
| 331 | { |
| 332 | next = bh->b_this_page; |
| 333 | block_end = block_start + blocksize; |
| 334 | if (block_end <= from || block_start >= to) { |
| 335 | if (partial && !buffer_uptodate(bh)) |
| 336 | *partial = 1; |
| 337 | continue; |
| 338 | } |
| 339 | err = (*fn)(handle, bh); |
| 340 | if (!ret) |
| 341 | ret = err; |
| 342 | } |
| 343 | return ret; |
| 344 | } |
| 345 | |
| 346 | struct ocfs2_journal_handle *ocfs2_start_walk_page_trans(struct inode *inode, |
| 347 | struct page *page, |
| 348 | unsigned from, |
| 349 | unsigned to) |
| 350 | { |
| 351 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
| 352 | struct ocfs2_journal_handle *handle = NULL; |
| 353 | int ret = 0; |
| 354 | |
| 355 | handle = ocfs2_start_trans(osb, NULL, OCFS2_INODE_UPDATE_CREDITS); |
| 356 | if (!handle) { |
| 357 | ret = -ENOMEM; |
| 358 | mlog_errno(ret); |
| 359 | goto out; |
| 360 | } |
| 361 | |
| 362 | if (ocfs2_should_order_data(inode)) { |
| 363 | ret = walk_page_buffers(handle->k_handle, |
| 364 | page_buffers(page), |
| 365 | from, to, NULL, |
| 366 | ocfs2_journal_dirty_data); |
| 367 | if (ret < 0) |
| 368 | mlog_errno(ret); |
| 369 | } |
| 370 | out: |
| 371 | if (ret) { |
| 372 | if (handle) |
| 373 | ocfs2_commit_trans(handle); |
| 374 | handle = ERR_PTR(ret); |
| 375 | } |
| 376 | return handle; |
| 377 | } |
| 378 | |
| 379 | static int ocfs2_commit_write(struct file *file, struct page *page, |
| 380 | unsigned from, unsigned to) |
| 381 | { |
| 382 | int ret, extending = 0, locklevel = 0; |
| 383 | loff_t new_i_size; |
| 384 | struct buffer_head *di_bh = NULL; |
| 385 | struct inode *inode = page->mapping->host; |
| 386 | struct ocfs2_journal_handle *handle = NULL; |
| 387 | |
| 388 | mlog_entry("(0x%p, 0x%p, %u, %u)\n", file, page, from, to); |
| 389 | |
| 390 | /* NOTE: ocfs2_file_aio_write has ensured that it's safe for |
| 391 | * us to sample inode->i_size here without the metadata lock: |
| 392 | * |
| 393 | * 1) We're currently holding the inode alloc lock, so no |
| 394 | * nodes can change it underneath us. |
| 395 | * |
| 396 | * 2) We've had to take the metadata lock at least once |
| 397 | * already to check for extending writes, hence insuring |
| 398 | * that our current copy is also up to date. |
| 399 | */ |
| 400 | new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to; |
| 401 | if (new_i_size > i_size_read(inode)) { |
| 402 | extending = 1; |
| 403 | locklevel = 1; |
| 404 | } |
| 405 | |
| 406 | ret = ocfs2_meta_lock_with_page(inode, NULL, &di_bh, locklevel, page); |
| 407 | if (ret != 0) { |
| 408 | mlog_errno(ret); |
| 409 | goto out; |
| 410 | } |
| 411 | |
| 412 | ret = ocfs2_data_lock_with_page(inode, 1, page); |
| 413 | if (ret != 0) { |
| 414 | mlog_errno(ret); |
| 415 | goto out_unlock_meta; |
| 416 | } |
| 417 | |
| 418 | if (extending) { |
| 419 | handle = ocfs2_start_walk_page_trans(inode, page, from, to); |
| 420 | if (IS_ERR(handle)) { |
| 421 | ret = PTR_ERR(handle); |
| 422 | handle = NULL; |
| 423 | goto out_unlock_data; |
| 424 | } |
| 425 | |
| 426 | /* Mark our buffer early. We'd rather catch this error up here |
| 427 | * as opposed to after a successful commit_write which would |
| 428 | * require us to set back inode->i_size. */ |
| 429 | ret = ocfs2_journal_access(handle, inode, di_bh, |
| 430 | OCFS2_JOURNAL_ACCESS_WRITE); |
| 431 | if (ret < 0) { |
| 432 | mlog_errno(ret); |
| 433 | goto out_commit; |
| 434 | } |
| 435 | } |
| 436 | |
| 437 | /* might update i_size */ |
| 438 | ret = generic_commit_write(file, page, from, to); |
| 439 | if (ret < 0) { |
| 440 | mlog_errno(ret); |
| 441 | goto out_commit; |
| 442 | } |
| 443 | |
| 444 | if (extending) { |
| 445 | loff_t size = (u64) i_size_read(inode); |
| 446 | struct ocfs2_dinode *di = |
| 447 | (struct ocfs2_dinode *)di_bh->b_data; |
| 448 | |
| 449 | /* ocfs2_mark_inode_dirty is too heavy to use here. */ |
| 450 | inode->i_blocks = ocfs2_align_bytes_to_sectors(size); |
| 451 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
| 452 | |
| 453 | di->i_size = cpu_to_le64(size); |
| 454 | di->i_ctime = di->i_mtime = |
| 455 | cpu_to_le64(inode->i_mtime.tv_sec); |
| 456 | di->i_ctime_nsec = di->i_mtime_nsec = |
| 457 | cpu_to_le32(inode->i_mtime.tv_nsec); |
| 458 | |
| 459 | ret = ocfs2_journal_dirty(handle, di_bh); |
| 460 | if (ret < 0) { |
| 461 | mlog_errno(ret); |
| 462 | goto out_commit; |
| 463 | } |
| 464 | } |
| 465 | |
| 466 | BUG_ON(extending && (i_size_read(inode) != new_i_size)); |
| 467 | |
| 468 | out_commit: |
| 469 | if (handle) |
| 470 | ocfs2_commit_trans(handle); |
| 471 | out_unlock_data: |
| 472 | ocfs2_data_unlock(inode, 1); |
| 473 | out_unlock_meta: |
| 474 | ocfs2_meta_unlock(inode, locklevel); |
| 475 | out: |
| 476 | if (di_bh) |
| 477 | brelse(di_bh); |
| 478 | |
| 479 | mlog_exit(ret); |
| 480 | return ret; |
| 481 | } |
| 482 | |
| 483 | static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block) |
| 484 | { |
| 485 | sector_t status; |
| 486 | u64 p_blkno = 0; |
| 487 | int err = 0; |
| 488 | struct inode *inode = mapping->host; |
| 489 | |
| 490 | mlog_entry("(block = %llu)\n", (unsigned long long)block); |
| 491 | |
| 492 | /* We don't need to lock journal system files, since they aren't |
| 493 | * accessed concurrently from multiple nodes. |
| 494 | */ |
| 495 | if (!INODE_JOURNAL(inode)) { |
| 496 | err = ocfs2_meta_lock(inode, NULL, NULL, 0); |
| 497 | if (err) { |
| 498 | if (err != -ENOENT) |
| 499 | mlog_errno(err); |
| 500 | goto bail; |
| 501 | } |
| 502 | down_read(&OCFS2_I(inode)->ip_alloc_sem); |
| 503 | } |
| 504 | |
| 505 | err = ocfs2_extent_map_get_blocks(inode, block, 1, &p_blkno, |
| 506 | NULL); |
| 507 | |
| 508 | if (!INODE_JOURNAL(inode)) { |
| 509 | up_read(&OCFS2_I(inode)->ip_alloc_sem); |
| 510 | ocfs2_meta_unlock(inode, 0); |
| 511 | } |
| 512 | |
| 513 | if (err) { |
| 514 | mlog(ML_ERROR, "get_blocks() failed, block = %llu\n", |
| 515 | (unsigned long long)block); |
| 516 | mlog_errno(err); |
| 517 | goto bail; |
| 518 | } |
| 519 | |
| 520 | |
| 521 | bail: |
| 522 | status = err ? 0 : p_blkno; |
| 523 | |
| 524 | mlog_exit((int)status); |
| 525 | |
| 526 | return status; |
| 527 | } |
| 528 | |
| 529 | /* |
| 530 | * TODO: Make this into a generic get_blocks function. |
| 531 | * |
| 532 | * From do_direct_io in direct-io.c: |
| 533 | * "So what we do is to permit the ->get_blocks function to populate |
| 534 | * bh.b_size with the size of IO which is permitted at this offset and |
| 535 | * this i_blkbits." |
| 536 | * |
| 537 | * This function is called directly from get_more_blocks in direct-io.c. |
| 538 | * |
| 539 | * called like this: dio->get_blocks(dio->inode, fs_startblk, |
| 540 | * fs_count, map_bh, dio->rw == WRITE); |
| 541 | */ |
| 542 | static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock, |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 543 | struct buffer_head *bh_result, int create) |
| 544 | { |
| 545 | int ret; |
| 546 | u64 vbo_max; /* file offset, max_blocks from iblock */ |
| 547 | u64 p_blkno; |
| 548 | int contig_blocks; |
| 549 | unsigned char blocksize_bits; |
Badari Pulavarty | 1d8fa7a | 2006-03-26 01:38:02 -0800 | [diff] [blame] | 550 | unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 551 | |
| 552 | if (!inode || !bh_result) { |
| 553 | mlog(ML_ERROR, "inode or bh_result is null\n"); |
| 554 | return -EIO; |
| 555 | } |
| 556 | |
| 557 | blocksize_bits = inode->i_sb->s_blocksize_bits; |
| 558 | |
| 559 | /* This function won't even be called if the request isn't all |
| 560 | * nicely aligned and of the right size, so there's no need |
| 561 | * for us to check any of that. */ |
| 562 | |
| 563 | vbo_max = ((u64)iblock + max_blocks) << blocksize_bits; |
| 564 | |
| 565 | spin_lock(&OCFS2_I(inode)->ip_lock); |
| 566 | if ((iblock + max_blocks) > |
| 567 | ocfs2_clusters_to_blocks(inode->i_sb, |
| 568 | OCFS2_I(inode)->ip_clusters)) { |
| 569 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
| 570 | ret = -EIO; |
| 571 | goto bail; |
| 572 | } |
| 573 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
| 574 | |
| 575 | /* This figures out the size of the next contiguous block, and |
| 576 | * our logical offset */ |
| 577 | ret = ocfs2_extent_map_get_blocks(inode, iblock, 1, &p_blkno, |
| 578 | &contig_blocks); |
| 579 | if (ret) { |
| 580 | mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n", |
| 581 | (unsigned long long)iblock); |
| 582 | ret = -EIO; |
| 583 | goto bail; |
| 584 | } |
| 585 | |
| 586 | map_bh(bh_result, inode->i_sb, p_blkno); |
| 587 | |
| 588 | /* make sure we don't map more than max_blocks blocks here as |
| 589 | that's all the kernel will handle at this point. */ |
| 590 | if (max_blocks < contig_blocks) |
| 591 | contig_blocks = max_blocks; |
| 592 | bh_result->b_size = contig_blocks << blocksize_bits; |
| 593 | bail: |
| 594 | return ret; |
| 595 | } |
| 596 | |
| 597 | /* |
| 598 | * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're |
| 599 | * particularly interested in the aio/dio case. Like the core uses |
| 600 | * i_alloc_sem, we use the rw_lock DLM lock to protect io on one node from |
| 601 | * truncation on another. |
| 602 | */ |
| 603 | static void ocfs2_dio_end_io(struct kiocb *iocb, |
| 604 | loff_t offset, |
| 605 | ssize_t bytes, |
| 606 | void *private) |
| 607 | { |
| 608 | struct inode *inode = iocb->ki_filp->f_dentry->d_inode; |
| 609 | |
| 610 | /* this io's submitter should not have unlocked this before we could */ |
| 611 | BUG_ON(!ocfs2_iocb_is_rw_locked(iocb)); |
| 612 | ocfs2_iocb_clear_rw_locked(iocb); |
| 613 | up_read(&inode->i_alloc_sem); |
| 614 | ocfs2_rw_unlock(inode, 0); |
| 615 | } |
| 616 | |
| 617 | static ssize_t ocfs2_direct_IO(int rw, |
| 618 | struct kiocb *iocb, |
| 619 | const struct iovec *iov, |
| 620 | loff_t offset, |
| 621 | unsigned long nr_segs) |
| 622 | { |
| 623 | struct file *file = iocb->ki_filp; |
| 624 | struct inode *inode = file->f_dentry->d_inode->i_mapping->host; |
| 625 | int ret; |
| 626 | |
| 627 | mlog_entry_void(); |
| 628 | ret = blockdev_direct_IO_no_locking(rw, iocb, inode, |
| 629 | inode->i_sb->s_bdev, iov, offset, |
| 630 | nr_segs, |
| 631 | ocfs2_direct_IO_get_blocks, |
| 632 | ocfs2_dio_end_io); |
| 633 | mlog_exit(ret); |
| 634 | return ret; |
| 635 | } |
| 636 | |
| 637 | struct address_space_operations ocfs2_aops = { |
| 638 | .readpage = ocfs2_readpage, |
| 639 | .writepage = ocfs2_writepage, |
| 640 | .prepare_write = ocfs2_prepare_write, |
| 641 | .commit_write = ocfs2_commit_write, |
| 642 | .bmap = ocfs2_bmap, |
| 643 | .sync_page = block_sync_page, |
| 644 | .direct_IO = ocfs2_direct_IO |
| 645 | }; |