1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
7 #include <linux/sched.h>
8 #include <linux/slab.h>
9 #include <linux/spinlock.h>
10 #include <linux/completion.h>
11 #include <linux/buffer_head.h>
13 #include <linux/pagemap.h>
14 #include <linux/writeback.h>
15 #include <linux/swap.h>
16 #include <linux/delay.h>
17 #include <linux/bio.h>
18 #include <linux/gfs2_ondisk.h>
31 #include "trace_gfs2.h"
33 static int gfs2_aspace_writepage(struct page *page, struct writeback_control *wbc)
35 struct buffer_head *bh, *head;
37 blk_opf_t write_flags = REQ_META | REQ_PRIO | wbc_to_write_flags(wbc);
39 BUG_ON(!PageLocked(page));
40 BUG_ON(!page_has_buffers(page));
42 head = page_buffers(page);
46 if (!buffer_mapped(bh))
49 * If it's a fully non-blocking write attempt and we cannot
50 * lock the buffer then redirty the page. Note that this can
51 * potentially cause a busy-wait loop from flusher thread and kswapd
52 * activity, but those code paths have their own higher-level
55 if (wbc->sync_mode != WB_SYNC_NONE) {
57 } else if (!trylock_buffer(bh)) {
58 redirty_page_for_writepage(wbc, page);
61 if (test_clear_buffer_dirty(bh)) {
62 mark_buffer_async_write(bh);
66 } while ((bh = bh->b_this_page) != head);
69 * The page and its buffers are protected by PageWriteback(), so we can
70 * drop the bh refcounts early.
72 BUG_ON(PageWriteback(page));
73 set_page_writeback(page);
76 struct buffer_head *next = bh->b_this_page;
77 if (buffer_async_write(bh)) {
78 submit_bh(REQ_OP_WRITE | write_flags, bh);
86 end_page_writeback(page);
91 const struct address_space_operations gfs2_meta_aops = {
92 .dirty_folio = block_dirty_folio,
93 .invalidate_folio = block_invalidate_folio,
94 .writepage = gfs2_aspace_writepage,
95 .release_folio = gfs2_release_folio,
98 const struct address_space_operations gfs2_rgrp_aops = {
99 .dirty_folio = block_dirty_folio,
100 .invalidate_folio = block_invalidate_folio,
101 .writepage = gfs2_aspace_writepage,
102 .release_folio = gfs2_release_folio,
106 * gfs2_getbuf - Get a buffer with a given address space
108 * @blkno: the block number (filesystem scope)
109 * @create: 1 if the buffer should be created
111 * Returns: the buffer
114 struct buffer_head *gfs2_getbuf(struct gfs2_glock *gl, u64 blkno, int create)
116 struct address_space *mapping = gfs2_glock2aspace(gl);
117 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
119 struct buffer_head *bh;
125 mapping = &sdp->sd_aspace;
127 shift = PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift;
128 index = blkno >> shift; /* convert block to page */
129 bufnum = blkno - (index << shift); /* block buf index within page */
132 folio = __filemap_get_folio(mapping, index,
133 FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
134 mapping_gfp_mask(mapping) | __GFP_NOFAIL);
135 bh = folio_buffers(folio);
137 bh = create_empty_buffers(folio,
138 sdp->sd_sb.sb_bsize, 0);
140 folio = __filemap_get_folio(mapping, index,
141 FGP_LOCK | FGP_ACCESSED, 0);
144 bh = folio_buffers(folio);
150 bh = get_nth_bh(bh, bufnum);
151 if (!buffer_mapped(bh))
152 map_bh(bh, sdp->sd_vfs, blkno);
161 static void meta_prep_new(struct buffer_head *bh)
163 struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data;
166 clear_buffer_dirty(bh);
167 set_buffer_uptodate(bh);
170 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
174 * gfs2_meta_new - Get a block
175 * @gl: The glock associated with this block
176 * @blkno: The block number
178 * Returns: The buffer
181 struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, u64 blkno)
183 struct buffer_head *bh;
184 bh = gfs2_getbuf(gl, blkno, CREATE);
189 static void gfs2_meta_read_endio(struct bio *bio)
191 struct bio_vec *bvec;
192 struct bvec_iter_all iter_all;
194 bio_for_each_segment_all(bvec, bio, iter_all) {
195 struct page *page = bvec->bv_page;
196 struct buffer_head *bh = page_buffers(page);
197 unsigned int len = bvec->bv_len;
199 while (bh_offset(bh) < bvec->bv_offset)
200 bh = bh->b_this_page;
202 struct buffer_head *next = bh->b_this_page;
204 bh->b_end_io(bh, !bio->bi_status);
212 * Submit several consecutive buffer head I/O requests as a single bio I/O
213 * request. (See submit_bh_wbc.)
215 static void gfs2_submit_bhs(blk_opf_t opf, struct buffer_head *bhs[], int num)
218 struct buffer_head *bh = *bhs;
221 bio = bio_alloc(bh->b_bdev, num, opf, GFP_NOIO);
222 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
225 if (!bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh))) {
226 BUG_ON(bio->bi_iter.bi_size == 0);
232 bio->bi_end_io = gfs2_meta_read_endio;
238 * gfs2_meta_read - Read a block from disk
239 * @gl: The glock covering the block
240 * @blkno: The block number
242 * @rahead: Do read-ahead
243 * @bhp: the place where the buffer is returned (NULL on failure)
248 int gfs2_meta_read(struct gfs2_glock *gl, u64 blkno, int flags,
249 int rahead, struct buffer_head **bhp)
251 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
252 struct buffer_head *bh, *bhs[2];
255 if (unlikely(gfs2_withdrawn(sdp)) && !gfs2_withdraw_in_prog(sdp)) {
260 *bhp = bh = gfs2_getbuf(gl, blkno, CREATE);
263 if (buffer_uptodate(bh)) {
267 bh->b_end_io = end_buffer_read_sync;
273 bh = gfs2_getbuf(gl, blkno + 1, CREATE);
276 if (buffer_uptodate(bh)) {
280 bh->b_end_io = end_buffer_read_sync;
285 gfs2_submit_bhs(REQ_OP_READ | REQ_META | REQ_PRIO, bhs, num);
286 if (!(flags & DIO_WAIT))
291 if (unlikely(!buffer_uptodate(bh))) {
292 struct gfs2_trans *tr = current->journal_info;
293 if (tr && test_bit(TR_TOUCHED, &tr->tr_flags))
294 gfs2_io_error_bh_wd(sdp, bh);
304 * gfs2_meta_wait - Reread a block from disk
305 * @sdp: the filesystem
306 * @bh: The block to wait for
311 int gfs2_meta_wait(struct gfs2_sbd *sdp, struct buffer_head *bh)
313 if (unlikely(gfs2_withdrawn(sdp)) && !gfs2_withdraw_in_prog(sdp))
318 if (!buffer_uptodate(bh)) {
319 struct gfs2_trans *tr = current->journal_info;
320 if (tr && test_bit(TR_TOUCHED, &tr->tr_flags))
321 gfs2_io_error_bh_wd(sdp, bh);
324 if (unlikely(gfs2_withdrawn(sdp)) && !gfs2_withdraw_in_prog(sdp))
330 void gfs2_remove_from_journal(struct buffer_head *bh, int meta)
332 struct address_space *mapping = bh->b_folio->mapping;
333 struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping);
334 struct gfs2_bufdata *bd = bh->b_private;
335 struct gfs2_trans *tr = current->journal_info;
338 if (test_clear_buffer_pinned(bh)) {
339 trace_gfs2_pin(bd, 0);
340 atomic_dec(&sdp->sd_log_pinned);
341 list_del_init(&bd->bd_list);
342 if (meta == REMOVE_META)
345 tr->tr_num_databuf_rm++;
346 set_bit(TR_TOUCHED, &tr->tr_flags);
352 gfs2_trans_add_revoke(sdp, bd);
353 } else if (was_pinned) {
354 bh->b_private = NULL;
355 kmem_cache_free(gfs2_bufdata_cachep, bd);
356 } else if (!list_empty(&bd->bd_ail_st_list) &&
357 !list_empty(&bd->bd_ail_gl_list)) {
358 gfs2_remove_from_ail(bd);
361 clear_buffer_dirty(bh);
362 clear_buffer_uptodate(bh);
366 * gfs2_ail1_wipe - remove deleted/freed buffers from the ail1 list
368 * @bstart: starting block address of buffers to remove
369 * @blen: length of buffers to be removed
371 * This function is called from gfs2_journal wipe, whose job is to remove
372 * buffers, corresponding to deleted blocks, from the journal. If we find any
373 * bufdata elements on the system ail1 list, they haven't been written to
374 * the journal yet. So we remove them.
376 static void gfs2_ail1_wipe(struct gfs2_sbd *sdp, u64 bstart, u32 blen)
378 struct gfs2_trans *tr, *s;
379 struct gfs2_bufdata *bd, *bs;
380 struct buffer_head *bh;
381 u64 end = bstart + blen;
384 spin_lock(&sdp->sd_ail_lock);
385 list_for_each_entry_safe(tr, s, &sdp->sd_ail1_list, tr_list) {
386 list_for_each_entry_safe(bd, bs, &tr->tr_ail1_list,
389 if (bh->b_blocknr < bstart || bh->b_blocknr >= end)
392 gfs2_remove_from_journal(bh, REMOVE_JDATA);
395 spin_unlock(&sdp->sd_ail_lock);
396 gfs2_log_unlock(sdp);
399 static struct buffer_head *gfs2_getjdatabuf(struct gfs2_inode *ip, u64 blkno)
401 struct address_space *mapping = ip->i_inode.i_mapping;
402 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
404 struct buffer_head *bh;
405 unsigned int shift = PAGE_SHIFT - sdp->sd_sb.sb_bsize_shift;
406 unsigned long index = blkno >> shift; /* convert block to page */
407 unsigned int bufnum = blkno - (index << shift);
409 folio = __filemap_get_folio(mapping, index, FGP_LOCK | FGP_ACCESSED, 0);
412 bh = folio_buffers(folio);
414 bh = get_nth_bh(bh, bufnum);
421 * gfs2_journal_wipe - make inode's buffers so they aren't dirty/pinned anymore
422 * @ip: the inode who owns the buffers
423 * @bstart: the first buffer in the run
424 * @blen: the number of buffers in the run
428 void gfs2_journal_wipe(struct gfs2_inode *ip, u64 bstart, u32 blen)
430 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
431 struct buffer_head *bh;
435 /* This can only happen during incomplete inode creation. */
436 BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
440 gfs2_ail1_wipe(sdp, bstart, blen);
443 bh = gfs2_getbuf(ip->i_gl, bstart, NO_CREATE);
444 if (!bh && gfs2_is_jdata(ip)) {
445 bh = gfs2_getjdatabuf(ip, bstart);
451 spin_lock(&sdp->sd_ail_lock);
452 gfs2_remove_from_journal(bh, ty);
453 spin_unlock(&sdp->sd_ail_lock);
454 gfs2_log_unlock(sdp);
465 * gfs2_meta_buffer - Get a metadata buffer
466 * @ip: The GFS2 inode
467 * @mtype: The block type (GFS2_METATYPE_*)
468 * @num: The block number (device relative) of the buffer
469 * @bhp: the buffer is returned here
474 int gfs2_meta_buffer(struct gfs2_inode *ip, u32 mtype, u64 num,
475 struct buffer_head **bhp)
477 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
478 struct gfs2_glock *gl = ip->i_gl;
479 struct buffer_head *bh;
483 if (num == ip->i_no_addr)
484 rahead = ip->i_rahead;
486 ret = gfs2_meta_read(gl, num, DIO_WAIT, rahead, &bh);
487 if (ret == 0 && gfs2_metatype_check(sdp, bh, mtype)) {
497 * gfs2_meta_ra - start readahead on an extent of a file
498 * @gl: the glock the blocks belong to
499 * @dblock: the starting disk block
500 * @extlen: the number of blocks in the extent
502 * returns: the first buffer in the extent
505 struct buffer_head *gfs2_meta_ra(struct gfs2_glock *gl, u64 dblock, u32 extlen)
507 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
508 struct buffer_head *first_bh, *bh;
509 u32 max_ra = gfs2_tune_get(sdp, gt_max_readahead) >>
510 sdp->sd_sb.sb_bsize_shift;
519 first_bh = gfs2_getbuf(gl, dblock, CREATE);
521 if (buffer_uptodate(first_bh))
523 bh_read_nowait(first_bh, REQ_META | REQ_PRIO);
529 bh = gfs2_getbuf(gl, dblock, CREATE);
531 bh_readahead(bh, REQ_RAHEAD | REQ_META | REQ_PRIO);
535 if (!buffer_locked(first_bh) && buffer_uptodate(first_bh))
539 wait_on_buffer(first_bh);