ia64/linux-2.6.18-xen.hg

view fs/ocfs2/aops.c @ 524:7f8b544237bf

netfront: Allow netfront in domain 0.

This is useful if your physical network device is in a utility domain.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
author Keir Fraser <keir.fraser@citrix.com>
date Tue Apr 15 15:18:58 2008 +0100 (2008-04-15)
parents 831230e53067
children
line source
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 */
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>
28 #define MLOG_MASK_PREFIX ML_FILE_IO
29 #include <cluster/masklog.h>
31 #include "ocfs2.h"
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"
43 #include "buffer_head_io.h"
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;
56 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode,
57 (unsigned long long)iblock, bh_result, create);
59 BUG_ON(ocfs2_inode_is_fast_symlink(inode));
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 }
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;
76 if (!OCFS2_IS_VALID_DINODE(fe)) {
77 mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n",
78 (unsigned long long)fe->i_blkno, 7, fe->i_signature);
79 goto bail;
80 }
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 }
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 }
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 }
120 map_bh(bh_result, inode->i_sb,
121 le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock);
123 err = 0;
125 bail:
126 if (bh)
127 brelse(bh);
129 mlog_exit(err);
130 return err;
131 }
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;
139 mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode,
140 (unsigned long long)iblock, bh_result, create);
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);
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 }
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;
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, "
165 "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
166 (unsigned long long)p_blkno);
167 goto bail;
168 }
170 map_bh(bh_result, inode->i_sb, p_blkno);
172 if (bh_result->b_blocknr == 0) {
173 err = -EIO;
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);
178 }
180 past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
181 mlog(0, "Inode %lu, past_eof = %llu\n", inode->i_ino,
182 (unsigned long long)past_eof);
184 if (create && (iblock >= past_eof))
185 set_buffer_new(bh_result);
187 bail:
188 if (err < 0)
189 err = -EIO;
191 mlog_exit(err);
192 return err;
193 }
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;
201 mlog_entry("(0x%p, %lu)\n", file, (page ? page->index : 0));
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 }
211 down_read(&OCFS2_I(inode)->ip_alloc_sem);
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 }
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 }
241 ret = block_read_full_page(page, ocfs2_get_block);
242 unlock = 0;
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 }
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;
270 mlog_entry("(0x%p)\n", page);
272 ret = block_write_full_page(page, ocfs2_get_block, wbc);
274 mlog_exit(ret);
276 return ret;
277 }
279 /* This can also be called from ocfs2_write_zero_page() which has done
280 * it's own cluster locking. */
281 int ocfs2_prepare_write_nolock(struct inode *inode, struct page *page,
282 unsigned from, unsigned to)
283 {
284 int ret;
286 down_read(&OCFS2_I(inode)->ip_alloc_sem);
288 ret = block_prepare_write(page, from, to, ocfs2_get_block);
290 up_read(&OCFS2_I(inode)->ip_alloc_sem);
292 return ret;
293 }
295 /*
296 * ocfs2_prepare_write() can be an outer-most ocfs2 call when it is called
297 * from loopback. It must be able to perform its own locking around
298 * ocfs2_get_block().
299 */
300 static int ocfs2_prepare_write(struct file *file, struct page *page,
301 unsigned from, unsigned to)
302 {
303 struct inode *inode = page->mapping->host;
304 int ret;
306 mlog_entry("(0x%p, 0x%p, %u, %u)\n", file, page, from, to);
308 ret = ocfs2_meta_lock_with_page(inode, NULL, NULL, 0, page);
309 if (ret != 0) {
310 mlog_errno(ret);
311 goto out;
312 }
314 ret = ocfs2_prepare_write_nolock(inode, page, from, to);
316 ocfs2_meta_unlock(inode, 0);
317 out:
318 mlog_exit(ret);
319 return ret;
320 }
322 /* Taken from ext3. We don't necessarily need the full blown
323 * functionality yet, but IMHO it's better to cut and paste the whole
324 * thing so we can avoid introducing our own bugs (and easily pick up
325 * their fixes when they happen) --Mark */
326 static int walk_page_buffers( handle_t *handle,
327 struct buffer_head *head,
328 unsigned from,
329 unsigned to,
330 int *partial,
331 int (*fn)( handle_t *handle,
332 struct buffer_head *bh))
333 {
334 struct buffer_head *bh;
335 unsigned block_start, block_end;
336 unsigned blocksize = head->b_size;
337 int err, ret = 0;
338 struct buffer_head *next;
340 for ( bh = head, block_start = 0;
341 ret == 0 && (bh != head || !block_start);
342 block_start = block_end, bh = next)
343 {
344 next = bh->b_this_page;
345 block_end = block_start + blocksize;
346 if (block_end <= from || block_start >= to) {
347 if (partial && !buffer_uptodate(bh))
348 *partial = 1;
349 continue;
350 }
351 err = (*fn)(handle, bh);
352 if (!ret)
353 ret = err;
354 }
355 return ret;
356 }
358 struct ocfs2_journal_handle *ocfs2_start_walk_page_trans(struct inode *inode,
359 struct page *page,
360 unsigned from,
361 unsigned to)
362 {
363 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
364 struct ocfs2_journal_handle *handle = NULL;
365 int ret = 0;
367 handle = ocfs2_start_trans(osb, NULL, OCFS2_INODE_UPDATE_CREDITS);
368 if (!handle) {
369 ret = -ENOMEM;
370 mlog_errno(ret);
371 goto out;
372 }
374 if (ocfs2_should_order_data(inode)) {
375 ret = walk_page_buffers(handle->k_handle,
376 page_buffers(page),
377 from, to, NULL,
378 ocfs2_journal_dirty_data);
379 if (ret < 0)
380 mlog_errno(ret);
381 }
382 out:
383 if (ret) {
384 if (handle)
385 ocfs2_commit_trans(handle);
386 handle = ERR_PTR(ret);
387 }
388 return handle;
389 }
391 static int ocfs2_commit_write(struct file *file, struct page *page,
392 unsigned from, unsigned to)
393 {
394 int ret, extending = 0, locklevel = 0;
395 loff_t new_i_size;
396 struct buffer_head *di_bh = NULL;
397 struct inode *inode = page->mapping->host;
398 struct ocfs2_journal_handle *handle = NULL;
400 mlog_entry("(0x%p, 0x%p, %u, %u)\n", file, page, from, to);
402 /* NOTE: ocfs2_file_aio_write has ensured that it's safe for
403 * us to sample inode->i_size here without the metadata lock:
404 *
405 * 1) We're currently holding the inode alloc lock, so no
406 * nodes can change it underneath us.
407 *
408 * 2) We've had to take the metadata lock at least once
409 * already to check for extending writes, hence insuring
410 * that our current copy is also up to date.
411 */
412 new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
413 if (new_i_size > i_size_read(inode)) {
414 extending = 1;
415 locklevel = 1;
416 }
418 ret = ocfs2_meta_lock_with_page(inode, NULL, &di_bh, locklevel, page);
419 if (ret != 0) {
420 mlog_errno(ret);
421 goto out;
422 }
424 ret = ocfs2_data_lock_with_page(inode, 1, page);
425 if (ret != 0) {
426 mlog_errno(ret);
427 goto out_unlock_meta;
428 }
430 if (extending) {
431 handle = ocfs2_start_walk_page_trans(inode, page, from, to);
432 if (IS_ERR(handle)) {
433 ret = PTR_ERR(handle);
434 handle = NULL;
435 goto out_unlock_data;
436 }
438 /* Mark our buffer early. We'd rather catch this error up here
439 * as opposed to after a successful commit_write which would
440 * require us to set back inode->i_size. */
441 ret = ocfs2_journal_access(handle, inode, di_bh,
442 OCFS2_JOURNAL_ACCESS_WRITE);
443 if (ret < 0) {
444 mlog_errno(ret);
445 goto out_commit;
446 }
447 }
449 /* might update i_size */
450 ret = generic_commit_write(file, page, from, to);
451 if (ret < 0) {
452 mlog_errno(ret);
453 goto out_commit;
454 }
456 if (extending) {
457 loff_t size = (u64) i_size_read(inode);
458 struct ocfs2_dinode *di =
459 (struct ocfs2_dinode *)di_bh->b_data;
461 /* ocfs2_mark_inode_dirty is too heavy to use here. */
462 inode->i_blocks = ocfs2_align_bytes_to_sectors(size);
463 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
465 di->i_size = cpu_to_le64(size);
466 di->i_ctime = di->i_mtime =
467 cpu_to_le64(inode->i_mtime.tv_sec);
468 di->i_ctime_nsec = di->i_mtime_nsec =
469 cpu_to_le32(inode->i_mtime.tv_nsec);
471 ret = ocfs2_journal_dirty(handle, di_bh);
472 if (ret < 0) {
473 mlog_errno(ret);
474 goto out_commit;
475 }
476 }
478 BUG_ON(extending && (i_size_read(inode) != new_i_size));
480 out_commit:
481 if (handle)
482 ocfs2_commit_trans(handle);
483 out_unlock_data:
484 ocfs2_data_unlock(inode, 1);
485 out_unlock_meta:
486 ocfs2_meta_unlock(inode, locklevel);
487 out:
488 if (di_bh)
489 brelse(di_bh);
491 mlog_exit(ret);
492 return ret;
493 }
495 static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
496 {
497 sector_t status;
498 u64 p_blkno = 0;
499 int err = 0;
500 struct inode *inode = mapping->host;
502 mlog_entry("(block = %llu)\n", (unsigned long long)block);
504 /* We don't need to lock journal system files, since they aren't
505 * accessed concurrently from multiple nodes.
506 */
507 if (!INODE_JOURNAL(inode)) {
508 err = ocfs2_meta_lock(inode, NULL, NULL, 0);
509 if (err) {
510 if (err != -ENOENT)
511 mlog_errno(err);
512 goto bail;
513 }
514 down_read(&OCFS2_I(inode)->ip_alloc_sem);
515 }
517 err = ocfs2_extent_map_get_blocks(inode, block, 1, &p_blkno,
518 NULL);
520 if (!INODE_JOURNAL(inode)) {
521 up_read(&OCFS2_I(inode)->ip_alloc_sem);
522 ocfs2_meta_unlock(inode, 0);
523 }
525 if (err) {
526 mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
527 (unsigned long long)block);
528 mlog_errno(err);
529 goto bail;
530 }
533 bail:
534 status = err ? 0 : p_blkno;
536 mlog_exit((int)status);
538 return status;
539 }
541 /*
542 * TODO: Make this into a generic get_blocks function.
543 *
544 * From do_direct_io in direct-io.c:
545 * "So what we do is to permit the ->get_blocks function to populate
546 * bh.b_size with the size of IO which is permitted at this offset and
547 * this i_blkbits."
548 *
549 * This function is called directly from get_more_blocks in direct-io.c.
550 *
551 * called like this: dio->get_blocks(dio->inode, fs_startblk,
552 * fs_count, map_bh, dio->rw == WRITE);
553 */
554 static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock,
555 struct buffer_head *bh_result, int create)
556 {
557 int ret;
558 u64 vbo_max; /* file offset, max_blocks from iblock */
559 u64 p_blkno;
560 int contig_blocks;
561 unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits;
562 unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
564 /* This function won't even be called if the request isn't all
565 * nicely aligned and of the right size, so there's no need
566 * for us to check any of that. */
568 vbo_max = ((u64)iblock + max_blocks) << blocksize_bits;
570 spin_lock(&OCFS2_I(inode)->ip_lock);
571 if ((iblock + max_blocks) >
572 ocfs2_clusters_to_blocks(inode->i_sb,
573 OCFS2_I(inode)->ip_clusters)) {
574 spin_unlock(&OCFS2_I(inode)->ip_lock);
575 ret = -EIO;
576 goto bail;
577 }
578 spin_unlock(&OCFS2_I(inode)->ip_lock);
580 /* This figures out the size of the next contiguous block, and
581 * our logical offset */
582 ret = ocfs2_extent_map_get_blocks(inode, iblock, 1, &p_blkno,
583 &contig_blocks);
584 if (ret) {
585 mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n",
586 (unsigned long long)iblock);
587 ret = -EIO;
588 goto bail;
589 }
591 map_bh(bh_result, inode->i_sb, p_blkno);
593 /* make sure we don't map more than max_blocks blocks here as
594 that's all the kernel will handle at this point. */
595 if (max_blocks < contig_blocks)
596 contig_blocks = max_blocks;
597 bh_result->b_size = contig_blocks << blocksize_bits;
598 bail:
599 return ret;
600 }
602 /*
603 * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
604 * particularly interested in the aio/dio case. Like the core uses
605 * i_alloc_sem, we use the rw_lock DLM lock to protect io on one node from
606 * truncation on another.
607 */
608 static void ocfs2_dio_end_io(struct kiocb *iocb,
609 loff_t offset,
610 ssize_t bytes,
611 void *private)
612 {
613 struct inode *inode = iocb->ki_filp->f_dentry->d_inode;
615 /* this io's submitter should not have unlocked this before we could */
616 BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
617 ocfs2_iocb_clear_rw_locked(iocb);
618 up_read(&inode->i_alloc_sem);
619 ocfs2_rw_unlock(inode, 0);
620 }
622 static ssize_t ocfs2_direct_IO(int rw,
623 struct kiocb *iocb,
624 const struct iovec *iov,
625 loff_t offset,
626 unsigned long nr_segs)
627 {
628 struct file *file = iocb->ki_filp;
629 struct inode *inode = file->f_dentry->d_inode->i_mapping->host;
630 int ret;
632 mlog_entry_void();
634 /*
635 * We get PR data locks even for O_DIRECT. This allows
636 * concurrent O_DIRECT I/O but doesn't let O_DIRECT with
637 * extending and buffered zeroing writes race. If they did
638 * race then the buffered zeroing could be written back after
639 * the O_DIRECT I/O. It's one thing to tell people not to mix
640 * buffered and O_DIRECT writes, but expecting them to
641 * understand that file extension is also an implicit buffered
642 * write is too much. By getting the PR we force writeback of
643 * the buffered zeroing before proceeding.
644 */
645 ret = ocfs2_data_lock(inode, 0);
646 if (ret < 0) {
647 mlog_errno(ret);
648 goto out;
649 }
650 ocfs2_data_unlock(inode, 0);
652 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
653 inode->i_sb->s_bdev, iov, offset,
654 nr_segs,
655 ocfs2_direct_IO_get_blocks,
656 ocfs2_dio_end_io);
657 out:
658 mlog_exit(ret);
659 return ret;
660 }
662 const struct address_space_operations ocfs2_aops = {
663 .readpage = ocfs2_readpage,
664 .writepage = ocfs2_writepage,
665 .prepare_write = ocfs2_prepare_write,
666 .commit_write = ocfs2_commit_write,
667 .bmap = ocfs2_bmap,
668 .sync_page = block_sync_page,
669 .direct_IO = ocfs2_direct_IO
670 };