2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 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/pagevec.h>
17 #include <linux/mpage.h>
19 #include <linux/gfs2_ondisk.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)
55 error = gfs2_block_map(inode, lblock, &new, &dblock, &boundary);
62 map_bh(bh_result, inode->i_sb, dblock);
64 set_buffer_new(bh_result);
66 set_buffer_boundary(bh_result);
72 * get_block_noalloc - Fills in a buffer head with details about a block
74 * @lblock: The block number to look up
75 * @bh_result: The buffer head to return the result in
76 * @create: Non-zero if we may add block to the file
81 static int get_block_noalloc(struct inode *inode, sector_t lblock,
82 struct buffer_head *bh_result, int create)
89 error = gfs2_block_map(inode, lblock, &new, &dblock, &boundary);
94 map_bh(bh_result, inode->i_sb, dblock);
95 else if (gfs2_assert_withdraw(GFS2_SB(inode), !create))
98 set_buffer_boundary(bh_result);
104 * gfs2_writepage - Write complete page
105 * @page: Page to write
109 * Some of this is copied from block_write_full_page() although we still
110 * call it to do most of the work.
113 static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
115 struct inode *inode = page->mapping->host;
116 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
117 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
118 loff_t i_size = i_size_read(inode);
119 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
124 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) {
128 if (current->journal_info)
131 /* Is the page fully outside i_size? (truncate in progress) */
132 offset = i_size & (PAGE_CACHE_SIZE-1);
133 if (page->index > end_index || (page->index == end_index && !offset)) {
134 page->mapping->a_ops->invalidatepage(page, 0);
136 return 0; /* don't care */
139 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip)) {
140 error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
143 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
146 error = block_write_full_page(page, get_block_noalloc, wbc);
149 gfs2_meta_cache_flush(ip);
153 redirty_page_for_writepage(wbc, page);
158 static int zero_readpage(struct page *page)
162 kaddr = kmap_atomic(page, KM_USER0);
163 memset(kaddr, 0, PAGE_CACHE_SIZE);
164 kunmap_atomic(page, KM_USER0);
166 SetPageUptodate(page);
172 * stuffed_readpage - Fill in a Linux page with stuffed file data
179 static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
181 struct buffer_head *dibh;
185 /* Only the first page of a stuffed file might contain data */
186 if (unlikely(page->index))
187 return zero_readpage(page);
189 error = gfs2_meta_inode_buffer(ip, &dibh);
193 kaddr = kmap_atomic(page, KM_USER0);
194 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
196 memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
197 kunmap_atomic(page, KM_USER0);
201 SetPageUptodate(page);
208 * gfs2_readpage - readpage with locking
209 * @file: The file to read a page for. N.B. This may be NULL if we are
210 * reading an internal file.
211 * @page: The page to read
216 static int gfs2_readpage(struct file *file, struct page *page)
218 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
219 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
220 struct gfs2_holder gh;
223 if (likely(file != &gfs2_internal_file_sentinal)) {
224 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|GL_AOP, &gh);
225 error = gfs2_glock_nq_m_atime(1, &gh);
230 if (gfs2_is_stuffed(ip)) {
231 error = stuffed_readpage(ip, page);
234 error = mpage_readpage(page, gfs2_get_block);
236 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
239 if (file != &gfs2_internal_file_sentinal) {
240 gfs2_glock_dq_m(1, &gh);
241 gfs2_holder_uninit(&gh);
247 if (file != &gfs2_internal_file_sentinal)
248 gfs2_holder_uninit(&gh);
252 #define list_to_page(head) (list_entry((head)->prev, struct page, lru))
255 * gfs2_readpages - Read a bunch of pages at once
258 * 1. This is only for readahead, so we can simply ignore any things
259 * which are slightly inconvenient (such as locking conflicts between
260 * the page lock and the glock) and return having done no I/O. Its
261 * obviously not something we'd want to do on too regular a basis.
262 * Any I/O we ignore at this time will be done via readpage later.
263 * 2. We have to handle stuffed files here too.
264 * 3. mpage_readpages() does most of the heavy lifting in the common case.
265 * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places.
266 * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as
267 * well as read-ahead.
269 static int gfs2_readpages(struct file *file, struct address_space *mapping,
270 struct list_head *pages, unsigned nr_pages)
272 struct inode *inode = mapping->host;
273 struct gfs2_inode *ip = GFS2_I(inode);
274 struct gfs2_sbd *sdp = GFS2_SB(inode);
275 struct gfs2_holder gh;
279 if (likely(file != &gfs2_internal_file_sentinal)) {
280 gfs2_holder_init(ip->i_gl, LM_ST_SHARED,
281 LM_FLAG_TRY_1CB|GL_ATIME|GL_AOP, &gh);
282 ret = gfs2_glock_nq_m_atime(1, &gh);
283 if (ret == GLR_TRYFAILED)
289 if (gfs2_is_stuffed(ip)) {
290 struct pagevec lru_pvec;
291 pagevec_init(&lru_pvec, 0);
292 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
293 struct page *page = list_to_page(pages);
294 list_del(&page->lru);
295 if (!add_to_page_cache(page, mapping,
296 page->index, GFP_KERNEL)) {
297 ret = stuffed_readpage(ip, page);
299 if (!pagevec_add(&lru_pvec, page))
300 __pagevec_lru_add(&lru_pvec);
302 page_cache_release(page);
304 pagevec_lru_add(&lru_pvec);
307 /* What we really want to do .... */
308 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block);
311 if (likely(file != &gfs2_internal_file_sentinal)) {
312 gfs2_glock_dq_m(1, &gh);
313 gfs2_holder_uninit(&gh);
316 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
322 /* unlock all pages, we can't do any I/O right now */
323 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
324 struct page *page = list_to_page(pages);
325 list_del(&page->lru);
327 page_cache_release(page);
329 if (likely(file != &gfs2_internal_file_sentinal))
330 gfs2_holder_uninit(&gh);
335 * gfs2_prepare_write - Prepare to write a page to a file
336 * @file: The file to write to
337 * @page: The page which is to be prepared for writing
338 * @from: From (byte range within page)
339 * @to: To (byte range within page)
344 static int gfs2_prepare_write(struct file *file, struct page *page,
345 unsigned from, unsigned to)
347 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
348 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
349 unsigned int data_blocks, ind_blocks, rblocks;
352 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from;
353 loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
354 struct gfs2_alloc *al;
356 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|GL_AOP, &ip->i_gh);
357 error = gfs2_glock_nq_m_atime(1, &ip->i_gh);
361 gfs2_write_calc_reserv(ip, to - from, &data_blocks, &ind_blocks);
363 error = gfs2_write_alloc_required(ip, pos, from - to, &alloc_required);
368 if (alloc_required) {
369 al = gfs2_alloc_get(ip);
371 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
375 error = gfs2_quota_check(ip, ip->i_di.di_uid, ip->i_di.di_gid);
379 al->al_requested = data_blocks + ind_blocks;
380 error = gfs2_inplace_reserve(ip);
385 rblocks = RES_DINODE + ind_blocks;
386 if (gfs2_is_jdata(ip))
387 rblocks += data_blocks ? data_blocks : 1;
388 if (ind_blocks || data_blocks)
389 rblocks += RES_STATFS + RES_QUOTA;
391 error = gfs2_trans_begin(sdp, rblocks, 0);
395 if (gfs2_is_stuffed(ip)) {
396 if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
397 error = gfs2_unstuff_dinode(ip, gfs2_unstuffer_page,
401 } else if (!PageUptodate(page))
402 error = stuffed_readpage(ip, page);
407 error = block_prepare_write(page, from, to, gfs2_get_block);
412 if (alloc_required) {
413 gfs2_inplace_release(ip);
415 gfs2_quota_unlock(ip);
420 gfs2_glock_dq_m(1, &ip->i_gh);
422 gfs2_holder_uninit(&ip->i_gh);
429 * gfs2_commit_write - Commit write to a file
430 * @file: The file to write to
431 * @page: The page containing the data
432 * @from: From (byte range within page)
433 * @to: To (byte range within page)
438 static int gfs2_commit_write(struct file *file, struct page *page,
439 unsigned from, unsigned to)
441 struct inode *inode = page->mapping->host;
442 struct gfs2_inode *ip = GFS2_I(inode);
443 struct gfs2_sbd *sdp = GFS2_SB(inode);
444 int error = -EOPNOTSUPP;
445 struct buffer_head *dibh;
446 struct gfs2_alloc *al = &ip->i_alloc;;
448 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)))
451 error = gfs2_meta_inode_buffer(ip, &dibh);
455 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
457 if (gfs2_is_stuffed(ip)) {
461 file_size = ((uint64_t)page->index << PAGE_CACHE_SHIFT) + to;
463 kaddr = kmap_atomic(page, KM_USER0);
464 memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from,
465 (char *)kaddr + from, to - from);
466 kunmap_atomic(page, KM_USER0);
468 SetPageUptodate(page);
470 if (inode->i_size < file_size)
471 i_size_write(inode, file_size);
473 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED ||
475 gfs2_page_add_databufs(ip, page, from, to);
476 error = generic_commit_write(file, page, from, to);
481 if (ip->i_di.di_size < inode->i_size)
482 ip->i_di.di_size = inode->i_size;
484 gfs2_dinode_out(&ip->i_di, dibh->b_data);
487 if (al->al_requested) {
488 gfs2_inplace_release(ip);
489 gfs2_quota_unlock(ip);
492 gfs2_glock_dq_m(1, &ip->i_gh);
493 gfs2_holder_uninit(&ip->i_gh);
500 if (al->al_requested) {
501 gfs2_inplace_release(ip);
502 gfs2_quota_unlock(ip);
505 gfs2_glock_dq_m(1, &ip->i_gh);
506 gfs2_holder_uninit(&ip->i_gh);
508 ClearPageUptodate(page);
513 * gfs2_bmap - Block map function
514 * @mapping: Address space info
515 * @lblock: The block to map
517 * Returns: The disk address for the block or 0 on hole or error
520 static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
522 struct gfs2_inode *ip = GFS2_I(mapping->host);
523 struct gfs2_holder i_gh;
527 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
531 if (!gfs2_is_stuffed(ip))
532 dblock = generic_block_bmap(mapping, lblock, gfs2_get_block);
534 gfs2_glock_dq_uninit(&i_gh);
539 static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh)
541 struct gfs2_bufdata *bd;
547 bh->b_private = NULL;
548 gfs2_log_unlock(sdp);
551 gfs2_log_unlock(sdp);
554 clear_buffer_dirty(bh);
556 clear_buffer_mapped(bh);
557 clear_buffer_req(bh);
558 clear_buffer_new(bh);
559 clear_buffer_delay(bh);
563 static void gfs2_invalidatepage(struct page *page, unsigned long offset)
565 struct gfs2_sbd *sdp = page->mapping->host->i_sb->s_fs_info;
566 struct buffer_head *head, *bh, *next;
567 unsigned int curr_off = 0;
569 BUG_ON(!PageLocked(page));
570 if (!page_has_buffers(page))
573 bh = head = page_buffers(page);
575 unsigned int next_off = curr_off + bh->b_size;
576 next = bh->b_this_page;
578 if (offset <= curr_off)
579 discard_buffer(sdp, bh);
583 } while (bh != head);
586 try_to_release_page(page, 0);
591 static ssize_t gfs2_direct_IO_write(struct kiocb *iocb, const struct iovec *iov,
592 loff_t offset, unsigned long nr_segs)
594 struct file *file = iocb->ki_filp;
595 struct inode *inode = file->f_mapping->host;
596 struct gfs2_inode *ip = GFS2_I(inode);
597 struct gfs2_holder gh;
601 * Shared lock, even though its write, since we do no allocation
602 * on this path. All we need change is atime.
604 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
605 rv = gfs2_glock_nq_m_atime(1, &gh);
610 * Should we return an error here? I can't see that O_DIRECT for
611 * a journaled file makes any sense. For now we'll silently fall
612 * back to buffered I/O, likewise we do the same for stuffed
613 * files since they are (a) small and (b) unaligned.
615 if (gfs2_is_jdata(ip))
618 if (gfs2_is_stuffed(ip))
621 rv = __blockdev_direct_IO(WRITE, iocb, inode, inode->i_sb->s_bdev,
622 iov, offset, nr_segs, gfs2_get_block,
623 NULL, DIO_OWN_LOCKING);
625 gfs2_glock_dq_m(1, &gh);
626 gfs2_holder_uninit(&gh);
634 * This is called with a shared lock already held for the read path.
635 * Currently, no locks are held when the write path is called.
637 static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
638 const struct iovec *iov, loff_t offset,
639 unsigned long nr_segs)
641 struct file *file = iocb->ki_filp;
642 struct inode *inode = file->f_mapping->host;
643 struct gfs2_inode *ip = GFS2_I(inode);
644 struct gfs2_sbd *sdp = GFS2_SB(inode);
647 return gfs2_direct_IO_write(iocb, iov, offset, nr_segs);
649 if (gfs2_assert_warn(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)) ||
650 gfs2_assert_warn(sdp, !gfs2_is_stuffed(ip)))
653 return __blockdev_direct_IO(READ, iocb, inode, inode->i_sb->s_bdev, iov,
654 offset, nr_segs, gfs2_get_block, NULL,
658 struct address_space_operations gfs2_file_aops = {
659 .writepage = gfs2_writepage,
660 .readpage = gfs2_readpage,
661 .readpages = gfs2_readpages,
662 .sync_page = block_sync_page,
663 .prepare_write = gfs2_prepare_write,
664 .commit_write = gfs2_commit_write,
666 .invalidatepage = gfs2_invalidatepage,
667 .direct_IO = gfs2_direct_IO,