trans.c

  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
  4 * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
  5 */
  6
  7#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  8
  9#include <linux/sched.h>
 10#include <linux/slab.h>
 11#include <linux/spinlock.h>
 12#include <linux/completion.h>
 13#include <linux/buffer_head.h>
 14#include <linux/kallsyms.h>
 15#include <linux/gfs2_ondisk.h>
 16
 17#include "gfs2.h"
 18#include "incore.h"
 19#include "glock.h"
 20#include "inode.h"
 21#include "log.h"
 22#include "lops.h"
 23#include "meta_io.h"
 24#include "trans.h"
 25#include "util.h"
 26#include "trace_gfs2.h"
 27
 28static void gfs2_print_trans(struct gfs2_sbd *sdp, const struct gfs2_trans *tr)
 29{
 30	fs_warn(sdp, "Transaction created at: %pSR\n", (void *)tr->tr_ip);
 31	fs_warn(sdp, "blocks=%u revokes=%u reserved=%u touched=%u\n",
 32		tr->tr_blocks, tr->tr_revokes, tr->tr_reserved,
 33		test_bit(TR_TOUCHED, &tr->tr_flags));
 34	fs_warn(sdp, "Buf %u/%u Databuf %u/%u Revoke %u\n",
 35		tr->tr_num_buf_new, tr->tr_num_buf_rm,
 36		tr->tr_num_databuf_new, tr->tr_num_databuf_rm,
 37		tr->tr_num_revoke);
 38}
 39
 40int __gfs2_trans_begin(struct gfs2_trans *tr, struct gfs2_sbd *sdp,
 41		       unsigned int blocks, unsigned int revokes,
 42		       unsigned long ip)
 43{
 44	unsigned int extra_revokes;
 45
 46	if (current->journal_info) {
 47		gfs2_print_trans(sdp, current->journal_info);
 48		BUG();
 49	}
 50	BUG_ON(blocks == 0 && revokes == 0);
 51
 52	if (gfs2_withdrawn(sdp))
 53		return -EROFS;
 54
 55	tr->tr_ip = ip;
 56	tr->tr_blocks = blocks;
 57	tr->tr_revokes = revokes;
 58	tr->tr_reserved = GFS2_LOG_FLUSH_MIN_BLOCKS;
 59	if (blocks) {
 60		/*
 61		 * The reserved blocks are either used for data or metadata.
 62		 * We can have mixed data and metadata, each with its own log
 63		 * descriptor block; see calc_reserved().
 64		 */
 65		tr->tr_reserved += blocks + 1 + DIV_ROUND_UP(blocks - 1, databuf_limit(sdp));
 66	}
 67	INIT_LIST_HEAD(&tr->tr_databuf);
 68	INIT_LIST_HEAD(&tr->tr_buf);
 69	INIT_LIST_HEAD(&tr->tr_list);
 70	INIT_LIST_HEAD(&tr->tr_ail1_list);
 71	INIT_LIST_HEAD(&tr->tr_ail2_list);
 72
 73	if (gfs2_assert_warn(sdp, tr->tr_reserved <= sdp->sd_jdesc->jd_blocks))
 74		return -EINVAL;
 75
 76	sb_start_intwrite(sdp->sd_vfs);
 77
 78	/*
 79	 * Try the reservations under sd_log_flush_lock to prevent log flushes
 80	 * from creating inconsistencies between the number of allocated and
 81	 * reserved revokes.  If that fails, do a full-block allocation outside
 82	 * of the lock to avoid stalling log flushes.  Then, allot the
 83	 * appropriate number of blocks to revokes, use as many revokes locally
 84	 * as needed, and "release" the surplus into the revokes pool.
 85	 */
 86
 87	down_read(&sdp->sd_log_flush_lock);
 88	if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)))
 89		goto out_not_live;
 90	if (gfs2_log_try_reserve(sdp, tr, &extra_revokes))
 91		goto reserved;
 92
 93	up_read(&sdp->sd_log_flush_lock);
 94	gfs2_log_reserve(sdp, tr, &extra_revokes);
 95	down_read(&sdp->sd_log_flush_lock);
 96	if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) {
 97		revokes = tr->tr_revokes + extra_revokes;
 98		gfs2_log_release_revokes(sdp, revokes);
 99		gfs2_log_release(sdp, tr->tr_reserved);
100		goto out_not_live;
101	}
102
103reserved:
104	gfs2_log_release_revokes(sdp, extra_revokes);
105	current->journal_info = tr;
106	return 0;
107
108out_not_live:
109	up_read(&sdp->sd_log_flush_lock);
110	sb_end_intwrite(sdp->sd_vfs);
111	return -EROFS;
112}
113
114int gfs2_trans_begin(struct gfs2_sbd *sdp, unsigned int blocks,
115		     unsigned int revokes)
116{
117	struct gfs2_trans *tr;
118	int error;
119
120	tr = kmem_cache_zalloc(gfs2_trans_cachep, GFP_NOFS);
121	if (!tr)
122		return -ENOMEM;
123	error = __gfs2_trans_begin(tr, sdp, blocks, revokes, _RET_IP_);
124	if (error)
125		kmem_cache_free(gfs2_trans_cachep, tr);
126	return error;
127}
128
129void gfs2_trans_end(struct gfs2_sbd *sdp)
130{
131	struct gfs2_trans *tr = current->journal_info;
132	s64 nbuf;
133
134	current->journal_info = NULL;
135
136	if (!test_bit(TR_TOUCHED, &tr->tr_flags)) {
137		gfs2_log_release_revokes(sdp, tr->tr_revokes);
138		up_read(&sdp->sd_log_flush_lock);
139		gfs2_log_release(sdp, tr->tr_reserved);
140		if (!test_bit(TR_ONSTACK, &tr->tr_flags))
141			gfs2_trans_free(sdp, tr);
142		sb_end_intwrite(sdp->sd_vfs);
143		return;
144	}
145
146	gfs2_log_release_revokes(sdp, tr->tr_revokes - tr->tr_num_revoke);
147
148	nbuf = tr->tr_num_buf_new + tr->tr_num_databuf_new;
149	nbuf -= tr->tr_num_buf_rm;
150	nbuf -= tr->tr_num_databuf_rm;
151
152	if (gfs2_assert_withdraw(sdp, nbuf <= tr->tr_blocks) ||
153	    gfs2_assert_withdraw(sdp, tr->tr_num_revoke <= tr->tr_revokes))
154		gfs2_print_trans(sdp, tr);
155
156	gfs2_log_commit(sdp, tr);
157	if (!test_bit(TR_ONSTACK, &tr->tr_flags) &&
158	    !test_bit(TR_ATTACHED, &tr->tr_flags))
159		gfs2_trans_free(sdp, tr);
160	up_read(&sdp->sd_log_flush_lock);
161
162	if (sdp->sd_vfs->s_flags & SB_SYNCHRONOUS)
163		gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
164			       GFS2_LFC_TRANS_END);
165	sb_end_intwrite(sdp->sd_vfs);
166}
167
168static struct gfs2_bufdata *gfs2_alloc_bufdata(struct gfs2_glock *gl,
169					       struct buffer_head *bh)
170{
171	struct gfs2_bufdata *bd;
172
173	bd = kmem_cache_zalloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL);
174	bd->bd_bh = bh;
175	bd->bd_gl = gl;
176	INIT_LIST_HEAD(&bd->bd_list);
177	INIT_LIST_HEAD(&bd->bd_ail_st_list);
178	INIT_LIST_HEAD(&bd->bd_ail_gl_list);
179	bh->b_private = bd;
180	return bd;
181}
182
183/**
184 * gfs2_trans_add_data - Add a databuf to the transaction.
185 * @gl: The inode glock associated with the buffer
186 * @bh: The buffer to add
187 *
188 * This is used in journaled data mode.
189 * We need to journal the data block in the same way as metadata in
190 * the functions above. The difference is that here we have a tag
191 * which is two __be64's being the block number (as per meta data)
192 * and a flag which says whether the data block needs escaping or
193 * not. This means we need a new log entry for each 251 or so data
194 * blocks, which isn't an enormous overhead but twice as much as
195 * for normal metadata blocks.
196 */
197void gfs2_trans_add_data(struct gfs2_glock *gl, struct buffer_head *bh)
198{
199	struct gfs2_trans *tr = current->journal_info;
200	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
201	struct gfs2_bufdata *bd;
202
203	lock_buffer(bh);
204	if (buffer_pinned(bh)) {
205		set_bit(TR_TOUCHED, &tr->tr_flags);
206		goto out;
207	}
208	gfs2_log_lock(sdp);
209	bd = bh->b_private;
210	if (bd == NULL) {
211		gfs2_log_unlock(sdp);
212		unlock_buffer(bh);
213		if (bh->b_private == NULL)
214			bd = gfs2_alloc_bufdata(gl, bh);
215		else
216			bd = bh->b_private;
217		lock_buffer(bh);
218		gfs2_log_lock(sdp);
219	}
220	gfs2_assert(sdp, bd->bd_gl == gl);
221	set_bit(TR_TOUCHED, &tr->tr_flags);
222	if (list_empty(&bd->bd_list)) {
223		set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
224		set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
225		gfs2_pin(sdp, bd->bd_bh);
226		tr->tr_num_databuf_new++;
227		list_add_tail(&bd->bd_list, &tr->tr_databuf);
228	}
229	gfs2_log_unlock(sdp);
230out:
231	unlock_buffer(bh);
232}
233
234void gfs2_trans_add_databufs(struct gfs2_glock *gl, struct folio *folio,
235			     size_t from, size_t len)
236{
237	struct buffer_head *head = folio_buffers(folio);
238	unsigned int bsize = head->b_size;
239	struct buffer_head *bh;
240	size_t to = from + len;
241	size_t start, end;
242
243	for (bh = head, start = 0; bh != head || !start;
244	     bh = bh->b_this_page, start = end) {
245		end = start + bsize;
246		if (end <= from)
247			continue;
248		if (start >= to)
249			break;
250		set_buffer_uptodate(bh);
251		gfs2_trans_add_data(gl, bh);
252	}
253}
254
255void gfs2_trans_add_meta(struct gfs2_glock *gl, struct buffer_head *bh)
256{
257
258	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
259	struct super_block *sb = sdp->sd_vfs;
260	struct gfs2_bufdata *bd;
261	struct gfs2_meta_header *mh;
262	struct gfs2_trans *tr = current->journal_info;
263
264	lock_buffer(bh);
265	if (buffer_pinned(bh)) {
266		set_bit(TR_TOUCHED, &tr->tr_flags);
267		goto out;
268	}
269	gfs2_log_lock(sdp);
270	bd = bh->b_private;
271	if (bd == NULL) {
272		gfs2_log_unlock(sdp);
273		unlock_buffer(bh);
274		folio_lock(bh->b_folio);
275		if (bh->b_private == NULL)
276			bd = gfs2_alloc_bufdata(gl, bh);
277		else
278			bd = bh->b_private;
279		folio_unlock(bh->b_folio);
280		lock_buffer(bh);
281		gfs2_log_lock(sdp);
282	}
283	gfs2_assert(sdp, bd->bd_gl == gl);
284	set_bit(TR_TOUCHED, &tr->tr_flags);
285	if (!list_empty(&bd->bd_list))
286		goto out_unlock;
287	set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
288	set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
289	mh = (struct gfs2_meta_header *)bd->bd_bh->b_data;
290	if (unlikely(mh->mh_magic != cpu_to_be32(GFS2_MAGIC))) {
291		fs_err(sdp, "Attempting to add uninitialised block to "
292		       "journal (inplace block=%lld)\n",
293		       (unsigned long long)bd->bd_bh->b_blocknr);
294		BUG();
295	}
296	if (gfs2_withdrawn(sdp)) {
297		fs_info(sdp, "GFS2:adding buf while withdrawn! 0x%llx\n",
298			(unsigned long long)bd->bd_bh->b_blocknr);
299		goto out_unlock;
300	}
301	if (unlikely(sb->s_writers.frozen == SB_FREEZE_COMPLETE)) {
302		fs_info(sdp, "GFS2:adding buf while frozen\n");
303		gfs2_withdraw(sdp);
304		goto out_unlock;
305	}
306	gfs2_pin(sdp, bd->bd_bh);
307	mh->__pad0 = cpu_to_be64(0);
308	mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
309	list_add(&bd->bd_list, &tr->tr_buf);
310	tr->tr_num_buf_new++;
311out_unlock:
312	gfs2_log_unlock(sdp);
313out:
314	unlock_buffer(bh);
315}
316
317void gfs2_trans_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
318{
319	struct gfs2_trans *tr = current->journal_info;
320
321	BUG_ON(!list_empty(&bd->bd_list));
322	gfs2_add_revoke(sdp, bd);
323	set_bit(TR_TOUCHED, &tr->tr_flags);
324	tr->tr_num_revoke++;
325}
326
327void gfs2_trans_remove_revoke(struct gfs2_sbd *sdp, u64 blkno, unsigned int len)
328{
329	struct gfs2_bufdata *bd, *tmp;
330	unsigned int n = len;
331
332	gfs2_log_lock(sdp);
333	list_for_each_entry_safe(bd, tmp, &sdp->sd_log_revokes, bd_list) {
334		if ((bd->bd_blkno >= blkno) && (bd->bd_blkno < (blkno + len))) {
335			list_del_init(&bd->bd_list);
336			gfs2_assert_withdraw(sdp, sdp->sd_log_num_revoke);
337			sdp->sd_log_num_revoke--;
338			if (bd->bd_gl)
339				gfs2_glock_remove_revoke(bd->bd_gl);
340			kmem_cache_free(gfs2_bufdata_cachep, bd);
341			gfs2_log_release_revokes(sdp, 1);
342			if (--n == 0)
343				break;
344		}
345	}
346	gfs2_log_unlock(sdp);
347}
348
349void gfs2_trans_free(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
350{
351	if (tr == NULL)
352		return;
353
354	gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list));
355	gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list));
356	gfs2_assert_warn(sdp, list_empty(&tr->tr_databuf));
357	gfs2_assert_warn(sdp, list_empty(&tr->tr_buf));
358	kmem_cache_free(gfs2_trans_cachep, tr);
359}