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 <linux/gfs2_ondisk.h>
19 #include <asm/semaphore.h>
22 #include "lm_interface.h"
29 #include "ops_address.h"
38 * gfs2_get_block - Fills in a buffer head with details about a block
40 * @lblock: The block number to look up
41 * @bh_result: The buffer head to return the result in
42 * @create: Non-zero if we may add block to the file
47 int gfs2_get_block(struct inode *inode, sector_t lblock,
48 struct buffer_head *bh_result, int create)
50 struct gfs2_inode *ip = inode->u.generic_ip;
55 error = gfs2_block_map(ip, lblock, &new, &dblock, NULL);
62 map_bh(bh_result, inode->i_sb, dblock);
64 set_buffer_new(bh_result);
70 * get_block_noalloc - Fills in a buffer head with details about a block
72 * @lblock: The block number to look up
73 * @bh_result: The buffer head to return the result in
74 * @create: Non-zero if we may add block to the file
79 static int get_block_noalloc(struct inode *inode, sector_t lblock,
80 struct buffer_head *bh_result, int create)
82 struct gfs2_inode *ip = inode->u.generic_ip;
87 error = gfs2_block_map(ip, lblock, &new, &dblock, NULL);
92 map_bh(bh_result, inode->i_sb, dblock);
93 else if (gfs2_assert_withdraw(ip->i_sbd, !create))
99 static int get_blocks(struct inode *inode, sector_t lblock,
100 unsigned long max_blocks, struct buffer_head *bh_result,
103 struct gfs2_inode *ip = inode->u.generic_ip;
109 error = gfs2_block_map(ip, lblock, &new, &dblock, &extlen);
116 map_bh(bh_result, inode->i_sb, dblock);
118 set_buffer_new(bh_result);
120 if (extlen > max_blocks)
122 bh_result->b_size = extlen << inode->i_blkbits;
127 static int get_blocks_noalloc(struct inode *inode, sector_t lblock,
128 unsigned long max_blocks,
129 struct buffer_head *bh_result, int create)
131 struct gfs2_inode *ip = inode->u.generic_ip;
137 error = gfs2_block_map(ip, lblock, &new, &dblock, &extlen);
142 map_bh(bh_result, inode->i_sb, dblock);
143 if (extlen > max_blocks)
145 bh_result->b_size = extlen << inode->i_blkbits;
146 } else if (gfs2_assert_withdraw(ip->i_sbd, !create))
153 * gfs2_writepage - Write complete page
154 * @page: Page to write
158 * Some of this is copied from block_write_full_page() although we still
159 * call it to do most of the work.
162 static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
164 struct inode *inode = page->mapping->host;
165 struct gfs2_inode *ip = page->mapping->host->u.generic_ip;
166 struct gfs2_sbd *sdp = ip->i_sbd;
167 loff_t i_size = i_size_read(inode);
168 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
173 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) {
177 if (current->journal_info)
180 /* Is the page fully outside i_size? (truncate in progress) */
181 offset = i_size & (PAGE_CACHE_SIZE-1);
182 if (page->index >= end_index+1 || !offset) {
183 page->mapping->a_ops->invalidatepage(page, 0);
185 return 0; /* don't care */
188 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip)) {
189 error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
192 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
195 error = block_write_full_page(page, get_block_noalloc, wbc);
198 gfs2_meta_cache_flush(ip);
202 redirty_page_for_writepage(wbc, page);
208 * stuffed_readpage - Fill in a Linux page with stuffed file data
215 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
217 struct buffer_head *dibh;
221 error = gfs2_meta_inode_buffer(ip, &dibh);
225 kaddr = kmap_atomic(page, KM_USER0);
226 memcpy((char *)kaddr,
227 dibh->b_data + sizeof(struct gfs2_dinode),
229 memset((char *)kaddr + ip->i_di.di_size,
231 PAGE_CACHE_SIZE - ip->i_di.di_size);
232 kunmap_atomic(page, KM_USER0);
236 SetPageUptodate(page);
241 static int zero_readpage(struct page *page)
245 kaddr = kmap_atomic(page, KM_USER0);
246 memset(kaddr, 0, PAGE_CACHE_SIZE);
247 kunmap_atomic(page, KM_USER0);
249 SetPageUptodate(page);
256 * gfs2_readpage - readpage with locking
257 * @file: The file to read a page for. N.B. This may be NULL if we are
258 * reading an internal file.
259 * @page: The page to read
264 static int gfs2_readpage(struct file *file, struct page *page)
266 struct gfs2_inode *ip = page->mapping->host->u.generic_ip;
267 struct gfs2_sbd *sdp = ip->i_sbd;
268 struct gfs2_holder gh;
271 if (file != &gfs2_internal_file_sentinal) {
272 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
273 error = gfs2_glock_nq_m_atime(1, &gh);
278 if (gfs2_is_stuffed(ip)) {
280 error = stuffed_readpage(ip, page);
283 error = zero_readpage(page);
285 error = mpage_readpage(page, gfs2_get_block);
287 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
290 if (file != &gfs2_internal_file_sentinal) {
291 gfs2_glock_dq_m(1, &gh);
292 gfs2_holder_uninit(&gh);
302 * gfs2_prepare_write - Prepare to write a page to a file
303 * @file: The file to write to
304 * @page: The page which is to be prepared for writing
305 * @from: From (byte range within page)
306 * @to: To (byte range within page)
311 static int gfs2_prepare_write(struct file *file, struct page *page,
312 unsigned from, unsigned to)
314 struct gfs2_inode *ip = page->mapping->host->u.generic_ip;
315 struct gfs2_sbd *sdp = ip->i_sbd;
316 unsigned int data_blocks, ind_blocks, rblocks;
319 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from;
320 loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
321 struct gfs2_alloc *al;
323 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME, &ip->i_gh);
324 error = gfs2_glock_nq_m_atime(1, &ip->i_gh);
328 gfs2_write_calc_reserv(ip, to - from, &data_blocks, &ind_blocks);
330 error = gfs2_write_alloc_required(ip, pos, from - to, &alloc_required);
335 if (alloc_required) {
336 al = gfs2_alloc_get(ip);
338 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
342 error = gfs2_quota_check(ip, ip->i_di.di_uid, ip->i_di.di_gid);
346 al->al_requested = data_blocks + ind_blocks;
347 error = gfs2_inplace_reserve(ip);
352 rblocks = RES_DINODE + ind_blocks;
353 if (gfs2_is_jdata(ip))
354 rblocks += data_blocks ? data_blocks : 1;
355 if (ind_blocks || data_blocks)
356 rblocks += RES_STATFS + RES_QUOTA;
358 error = gfs2_trans_begin(sdp, rblocks, 0);
362 if (gfs2_is_stuffed(ip)) {
363 if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
364 error = gfs2_unstuff_dinode(ip, gfs2_unstuffer_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 = inode->u.generic_ip;
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 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)))
418 error = gfs2_meta_inode_buffer(ip, &dibh);
422 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
424 if (gfs2_is_stuffed(ip)) {
428 file_size = ((uint64_t)page->index << PAGE_CACHE_SHIFT) + to;
430 kaddr = kmap_atomic(page, KM_USER0);
431 memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from,
432 (char *)kaddr + from, to - from);
433 kunmap_atomic(page, KM_USER0);
435 SetPageUptodate(page);
437 if (inode->i_size < file_size)
438 i_size_write(inode, file_size);
440 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED ||
442 gfs2_page_add_databufs(ip, page, from, to);
443 error = generic_commit_write(file, page, from, to);
448 if (ip->i_di.di_size < inode->i_size)
449 ip->i_di.di_size = inode->i_size;
451 gfs2_dinode_out(&ip->i_di, dibh->b_data);
454 if (al->al_requested) {
455 gfs2_inplace_release(ip);
456 gfs2_quota_unlock(ip);
459 gfs2_glock_dq_m(1, &ip->i_gh);
460 gfs2_holder_uninit(&ip->i_gh);
467 if (al->al_requested) {
468 gfs2_inplace_release(ip);
469 gfs2_quota_unlock(ip);
472 gfs2_glock_dq_m(1, &ip->i_gh);
473 gfs2_holder_uninit(&ip->i_gh);
475 ClearPageUptodate(page);
480 * gfs2_bmap - Block map function
481 * @mapping: Address space info
482 * @lblock: The block to map
484 * Returns: The disk address for the block or 0 on hole or error
487 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
489 struct gfs2_inode *ip = mapping->host->u.generic_ip;
490 struct gfs2_holder i_gh;
494 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
498 if (!gfs2_is_stuffed(ip))
499 dblock = generic_block_bmap(mapping, lblock, gfs2_get_block);
501 gfs2_glock_dq_uninit(&i_gh);
506 static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh)
508 struct gfs2_bufdata *bd;
514 bh->b_private = NULL;
515 gfs2_log_unlock(sdp);
518 gfs2_log_unlock(sdp);
521 clear_buffer_dirty(bh);
523 clear_buffer_mapped(bh);
524 clear_buffer_req(bh);
525 clear_buffer_new(bh);
526 clear_buffer_delay(bh);
530 static int gfs2_invalidatepage(struct page *page, unsigned long offset)
532 struct gfs2_sbd *sdp = page->mapping->host->i_sb->s_fs_info;
533 struct buffer_head *head, *bh, *next;
534 unsigned int curr_off = 0;
537 BUG_ON(!PageLocked(page));
538 if (!page_has_buffers(page))
541 bh = head = page_buffers(page);
543 unsigned int next_off = curr_off + bh->b_size;
544 next = bh->b_this_page;
546 if (offset <= curr_off)
547 discard_buffer(sdp, bh);
551 } while (bh != head);
554 ret = try_to_release_page(page, 0);
559 static ssize_t gfs2_direct_IO_write(struct kiocb *iocb, const struct iovec *iov,
560 loff_t offset, unsigned long nr_segs)
562 struct file *file = iocb->ki_filp;
563 struct inode *inode = file->f_mapping->host;
564 struct gfs2_inode *ip = inode->u.generic_ip;
565 struct gfs2_holder gh;
569 * Shared lock, even though its write, since we do no allocation
570 * on this path. All we need change is atime.
572 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
573 rv = gfs2_glock_nq_m_atime(1, &gh);
578 * Should we return an error here? I can't see that O_DIRECT for
579 * a journaled file makes any sense. For now we'll silently fall
580 * back to buffered I/O, likewise we do the same for stuffed
581 * files since they are (a) small and (b) unaligned.
583 if (gfs2_is_jdata(ip))
586 if (gfs2_is_stuffed(ip))
589 rv = __blockdev_direct_IO(WRITE, iocb, inode, inode->i_sb->s_bdev,
590 iov, offset, nr_segs, get_blocks_noalloc,
591 NULL, DIO_OWN_LOCKING);
593 gfs2_glock_dq_m(1, &gh);
594 gfs2_holder_uninit(&gh);
602 * This is called with a shared lock already held for the read path.
603 * Currently, no locks are held when the write path is called.
605 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
606 const struct iovec *iov, loff_t offset,
607 unsigned long nr_segs)
609 struct file *file = iocb->ki_filp;
610 struct inode *inode = file->f_mapping->host;
611 struct gfs2_inode *ip = inode->u.generic_ip;
612 struct gfs2_sbd *sdp = ip->i_sbd;
615 return gfs2_direct_IO_write(iocb, iov, offset, nr_segs);
617 if (gfs2_assert_warn(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)) ||
618 gfs2_assert_warn(sdp, !gfs2_is_stuffed(ip)))
621 return __blockdev_direct_IO(READ, iocb, inode, inode->i_sb->s_bdev, iov,
622 offset, nr_segs, get_blocks, NULL,
626 struct address_space_operations gfs2_file_aops = {
627 .writepage = gfs2_writepage,
628 .readpage = gfs2_readpage,
629 .sync_page = block_sync_page,
630 .prepare_write = gfs2_prepare_write,
631 .commit_write = gfs2_commit_write,
633 .invalidatepage = gfs2_invalidatepage,
634 .direct_IO = gfs2_direct_IO,