2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved.
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 v.2.
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/pagemap.h>
16 #include <linux/mpage.h>
18 #include <asm/semaphore.h>
26 #include "ops_address.h"
34 * gfs2_get_block - Fills in a buffer head with details about a block
36 * @lblock: The block number to look up
37 * @bh_result: The buffer head to return the result in
38 * @create: Non-zero if we may add block to the file
43 int gfs2_get_block(struct inode *inode, sector_t lblock,
44 struct buffer_head *bh_result, int create)
46 struct gfs2_inode *ip = get_v2ip(inode);
51 error = gfs2_block_map(ip, lblock, &new, &dblock, NULL);
58 map_bh(bh_result, inode->i_sb, dblock);
60 set_buffer_new(bh_result);
66 * get_block_noalloc - Fills in a buffer head with details about a block
68 * @lblock: The block number to look up
69 * @bh_result: The buffer head to return the result in
70 * @create: Non-zero if we may add block to the file
75 static int get_block_noalloc(struct inode *inode, sector_t lblock,
76 struct buffer_head *bh_result, int create)
78 struct gfs2_inode *ip = get_v2ip(inode);
83 error = gfs2_block_map(ip, lblock, &new, &dblock, NULL);
88 map_bh(bh_result, inode->i_sb, dblock);
89 else if (gfs2_assert_withdraw(ip->i_sbd, !create))
95 static int get_blocks(struct inode *inode, sector_t lblock,
96 unsigned long max_blocks, struct buffer_head *bh_result,
99 struct gfs2_inode *ip = get_v2ip(inode);
105 error = gfs2_block_map(ip, lblock, &new, &dblock, &extlen);
112 map_bh(bh_result, inode->i_sb, dblock);
114 set_buffer_new(bh_result);
116 if (extlen > max_blocks)
118 bh_result->b_size = extlen << inode->i_blkbits;
123 static int get_blocks_noalloc(struct inode *inode, sector_t lblock,
124 unsigned long max_blocks,
125 struct buffer_head *bh_result, int create)
127 struct gfs2_inode *ip = get_v2ip(inode);
133 error = gfs2_block_map(ip, lblock, &new, &dblock, &extlen);
138 map_bh(bh_result, inode->i_sb, dblock);
139 if (extlen > max_blocks)
141 bh_result->b_size = extlen << inode->i_blkbits;
142 } else if (gfs2_assert_withdraw(ip->i_sbd, !create))
149 * gfs2_writepage - Write complete page
150 * @page: Page to write
154 * Some of this is copied from block_write_full_page() although we still
155 * call it to do most of the work.
158 static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
160 struct inode *inode = page->mapping->host;
161 struct gfs2_inode *ip = get_v2ip(page->mapping->host);
162 struct gfs2_sbd *sdp = ip->i_sbd;
163 loff_t i_size = i_size_read(inode);
164 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
169 atomic_inc(&sdp->sd_ops_address);
170 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) {
177 /* Is the page fully outside i_size? (truncate in progress) */
178 offset = i_size & (PAGE_CACHE_SIZE-1);
179 if (page->index >= end_index+1 || !offset) {
180 page->mapping->a_ops->invalidatepage(page, 0);
182 return 0; /* don't care */
185 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip)) {
186 error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
189 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
192 error = block_write_full_page(page, get_block_noalloc, wbc);
195 gfs2_meta_cache_flush(ip);
199 redirty_page_for_writepage(wbc, page);
205 * stuffed_readpage - Fill in a Linux page with stuffed file data
212 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
214 struct buffer_head *dibh;
218 error = gfs2_meta_inode_buffer(ip, &dibh);
222 kaddr = kmap_atomic(page, KM_USER0);
223 memcpy((char *)kaddr,
224 dibh->b_data + sizeof(struct gfs2_dinode),
226 memset((char *)kaddr + ip->i_di.di_size,
228 PAGE_CACHE_SIZE - ip->i_di.di_size);
229 kunmap_atomic(page, KM_USER0);
233 SetPageUptodate(page);
238 static int zero_readpage(struct page *page)
242 kaddr = kmap_atomic(page, KM_USER0);
243 memset(kaddr, 0, PAGE_CACHE_SIZE);
244 kunmap_atomic(page, KM_USER0);
246 SetPageUptodate(page);
253 * gfs2_readpage - readpage with locking
254 * @file: The file to read a page for. N.B. This may be NULL if we are
255 * reading an internal file.
256 * @page: The page to read
261 static int gfs2_readpage(struct file *file, struct page *page)
263 struct gfs2_inode *ip = get_v2ip(page->mapping->host);
264 struct gfs2_sbd *sdp = ip->i_sbd;
265 struct gfs2_holder gh;
268 atomic_inc(&sdp->sd_ops_address);
270 if (file != &gfs2_internal_file_sentinal) {
271 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
272 error = gfs2_glock_nq_m_atime(1, &gh);
277 if (gfs2_is_stuffed(ip)) {
279 error = stuffed_readpage(ip, page);
282 error = zero_readpage(page);
284 error = mpage_readpage(page, gfs2_get_block);
286 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
289 if (file != &gfs2_internal_file_sentinal) {
290 gfs2_glock_dq_m(1, &gh);
291 gfs2_holder_uninit(&gh);
301 * gfs2_prepare_write - Prepare to write a page to a file
302 * @file: The file to write to
303 * @page: The page which is to be prepared for writing
304 * @from: From (byte range within page)
305 * @to: To (byte range within page)
310 static int gfs2_prepare_write(struct file *file, struct page *page,
311 unsigned from, unsigned to)
313 struct gfs2_inode *ip = get_v2ip(page->mapping->host);
314 struct gfs2_sbd *sdp = ip->i_sbd;
315 unsigned int data_blocks, ind_blocks, rblocks;
318 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from;
319 loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
320 struct gfs2_alloc *al;
322 atomic_inc(&sdp->sd_ops_address);
324 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME, &ip->i_gh);
325 error = gfs2_glock_nq_m_atime(1, &ip->i_gh);
329 gfs2_write_calc_reserv(ip, to - from, &data_blocks, &ind_blocks);
331 error = gfs2_write_alloc_required(ip, pos, from - to, &alloc_required);
336 if (alloc_required) {
337 al = gfs2_alloc_get(ip);
339 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
343 error = gfs2_quota_check(ip, ip->i_di.di_uid, ip->i_di.di_gid);
347 al->al_requested = data_blocks + ind_blocks;
348 error = gfs2_inplace_reserve(ip);
353 rblocks = RES_DINODE + ind_blocks;
354 if (gfs2_is_jdata(ip))
355 rblocks += data_blocks ? data_blocks : 1;
356 if (ind_blocks || data_blocks)
357 rblocks += RES_STATFS + RES_QUOTA;
359 error = gfs2_trans_begin(sdp, rblocks, 0);
363 if (gfs2_is_stuffed(ip)) {
364 if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
365 error = gfs2_unstuff_dinode(ip, gfs2_unstuffer_page, page);
368 } else if (!PageUptodate(page))
369 error = stuffed_readpage(ip, page);
374 error = block_prepare_write(page, from, to, gfs2_get_block);
379 if (alloc_required) {
380 gfs2_inplace_release(ip);
382 gfs2_quota_unlock(ip);
387 gfs2_glock_dq_m(1, &ip->i_gh);
389 gfs2_holder_uninit(&ip->i_gh);
396 * gfs2_commit_write - Commit write to a file
397 * @file: The file to write to
398 * @page: The page containing the data
399 * @from: From (byte range within page)
400 * @to: To (byte range within page)
405 static int gfs2_commit_write(struct file *file, struct page *page,
406 unsigned from, unsigned to)
408 struct inode *inode = page->mapping->host;
409 struct gfs2_inode *ip = get_v2ip(inode);
410 struct gfs2_sbd *sdp = ip->i_sbd;
411 int error = -EOPNOTSUPP;
412 struct buffer_head *dibh;
413 struct gfs2_alloc *al = &ip->i_alloc;;
415 atomic_inc(&sdp->sd_ops_address);
418 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)))
421 error = gfs2_meta_inode_buffer(ip, &dibh);
425 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
427 if (gfs2_is_stuffed(ip)) {
431 file_size = ((uint64_t)page->index << PAGE_CACHE_SHIFT) + to;
433 kaddr = kmap_atomic(page, KM_USER0);
434 memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from,
435 (char *)kaddr + from, to - from);
436 kunmap_atomic(page, KM_USER0);
438 SetPageUptodate(page);
440 if (inode->i_size < file_size)
441 i_size_write(inode, file_size);
443 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip))
444 gfs2_page_add_databufs(ip, page, from, to);
445 error = generic_commit_write(file, page, from, to);
450 if (ip->i_di.di_size < inode->i_size)
451 ip->i_di.di_size = inode->i_size;
453 gfs2_dinode_out(&ip->i_di, dibh->b_data);
456 if (al->al_requested) {
457 gfs2_inplace_release(ip);
458 gfs2_quota_unlock(ip);
461 gfs2_glock_dq_m(1, &ip->i_gh);
462 gfs2_holder_uninit(&ip->i_gh);
469 if (al->al_requested) {
470 gfs2_inplace_release(ip);
471 gfs2_quota_unlock(ip);
474 gfs2_glock_dq_m(1, &ip->i_gh);
475 gfs2_holder_uninit(&ip->i_gh);
477 ClearPageUptodate(page);
482 * gfs2_bmap - Block map function
483 * @mapping: Address space info
484 * @lblock: The block to map
486 * Returns: The disk address for the block or 0 on hole or error
489 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
491 struct gfs2_inode *ip = get_v2ip(mapping->host);
492 struct gfs2_holder i_gh;
496 atomic_inc(&ip->i_sbd->sd_ops_address);
498 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
502 if (!gfs2_is_stuffed(ip))
503 dblock = generic_block_bmap(mapping, lblock, gfs2_get_block);
505 gfs2_glock_dq_uninit(&i_gh);
510 static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh)
512 struct gfs2_bufdata *bd;
519 gfs2_log_unlock(sdp);
522 gfs2_log_unlock(sdp);
525 clear_buffer_dirty(bh);
527 clear_buffer_mapped(bh);
528 clear_buffer_req(bh);
529 clear_buffer_new(bh);
530 clear_buffer_delay(bh);
534 static int gfs2_invalidatepage(struct page *page, unsigned long offset)
536 struct gfs2_sbd *sdp = get_v2sdp(page->mapping->host->i_sb);
537 struct buffer_head *head, *bh, *next;
538 unsigned int curr_off = 0;
541 BUG_ON(!PageLocked(page));
542 if (!page_has_buffers(page))
545 bh = head = page_buffers(page);
547 unsigned int next_off = curr_off + bh->b_size;
548 next = bh->b_this_page;
550 if (offset <= curr_off)
551 discard_buffer(sdp, bh);
555 } while (bh != head);
558 ret = try_to_release_page(page, 0);
563 static ssize_t gfs2_direct_IO_write(struct kiocb *iocb, const struct iovec *iov,
564 loff_t offset, unsigned long nr_segs)
566 struct file *file = iocb->ki_filp;
567 struct inode *inode = file->f_mapping->host;
568 struct gfs2_inode *ip = get_v2ip(inode);
569 struct gfs2_holder gh;
573 * Shared lock, even though its write, since we do no allocation
574 * on this path. All we need change is atime.
576 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
577 rv = gfs2_glock_nq_m_atime(1, &gh);
582 * Should we return an error here? I can't see that O_DIRECT for
583 * a journaled file makes any sense. For now we'll silently fall
584 * back to buffered I/O, likewise we do the same for stuffed
585 * files since they are (a) small and (b) unaligned.
587 if (gfs2_is_jdata(ip))
590 if (gfs2_is_stuffed(ip))
593 rv = __blockdev_direct_IO(WRITE, iocb, inode, inode->i_sb->s_bdev,
594 iov, offset, nr_segs, get_blocks_noalloc,
595 NULL, DIO_OWN_LOCKING);
597 gfs2_glock_dq_m(1, &gh);
598 gfs2_holder_uninit(&gh);
606 * This is called with a shared lock already held for the read path.
607 * Currently, no locks are held when the write path is called.
609 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
610 const struct iovec *iov, loff_t offset,
611 unsigned long nr_segs)
613 struct file *file = iocb->ki_filp;
614 struct inode *inode = file->f_mapping->host;
615 struct gfs2_inode *ip = get_v2ip(inode);
616 struct gfs2_sbd *sdp = ip->i_sbd;
618 atomic_inc(&sdp->sd_ops_address);
621 return gfs2_direct_IO_write(iocb, iov, offset, nr_segs);
623 if (gfs2_assert_warn(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)) ||
624 gfs2_assert_warn(sdp, !gfs2_is_stuffed(ip)))
627 return __blockdev_direct_IO(READ, iocb, inode, inode->i_sb->s_bdev, iov,
628 offset, nr_segs, get_blocks, NULL,
632 struct address_space_operations gfs2_file_aops = {
633 .writepage = gfs2_writepage,
634 .readpage = gfs2_readpage,
635 .sync_page = block_sync_page,
636 .prepare_write = gfs2_prepare_write,
637 .commit_write = gfs2_commit_write,
639 .invalidatepage = gfs2_invalidatepage,
640 .direct_IO = gfs2_direct_IO,