1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2013 Jie Liu.
4 * All Rights Reserved.
5 */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_da_format.h"
14 #include "xfs_trans_space.h"
15 #include "xfs_da_btree.h"
16 #include "xfs_bmap_btree.h"
17 #include "xfs_trace.h"
18
19 /*
20 * Shortly after enabling the large extents count feature in 2023, longstanding
21 * bugs were found in the code that computes the minimum log size. Luckily,
22 * the bugs resulted in over-estimates of that size, so there's no impact to
23 * existing users. However, we don't want to reduce the minimum log size
24 * because that can create the situation where a newer mkfs writes a new
25 * filesystem that an older kernel won't mount.
26 *
27 * Several years prior, we also discovered that the transaction reservations
28 * for rmap and reflink operations were unnecessarily large. That was fixed,
29 * but the minimum log size computation was left alone to avoid the
30 * compatibility problems noted above. Fix that too.
31 *
32 * Therefore, we only may correct the computation starting with filesystem
33 * features that didn't exist in 2023. In other words, only turn this on if
34 * the filesystem has parent pointers.
35 *
36 * This function can be called before the XFS_HAS_* flags have been set up,
37 * (e.g. mkfs) so we must check the ondisk superblock.
38 */
39 static inline bool
40 xfs_want_minlogsize_fixes(
41 struct xfs_sb *sb)
42 {
43 return xfs_sb_is_v5(sb) &&
44 xfs_sb_has_incompat_feature(sb, XFS_SB_FEAT_INCOMPAT_PARENT);
45 }
46
47 /*
48 * Calculate the maximum length in bytes that would be required for a local
49 * attribute value as large attributes out of line are not logged.
50 */
51 STATIC int
52 xfs_log_calc_max_attrsetm_res(
53 struct xfs_mount *mp)
54 {
55 int size;
56 int nblks;
57
58 size = xfs_attr_leaf_entsize_local_max(mp->m_attr_geo->blksize) -
59 MAXNAMELEN - 1;
60 nblks = XFS_DAENTER_SPACE_RES(mp, XFS_ATTR_FORK);
61 nblks += XFS_B_TO_FSB(mp, size);
62
63 /*
64 * If the feature set is new enough, correct a unit conversion error in
65 * the xattr transaction reservation code that resulted in oversized
66 * minimum log size computations.
67 */
68 if (xfs_want_minlogsize_fixes(&mp->m_sb))
69 size = XFS_B_TO_FSB(mp, size);
70
71 nblks += XFS_NEXTENTADD_SPACE_RES(mp, size, XFS_ATTR_FORK);
72
73 return M_RES(mp)->tr_attrsetm.tr_logres +
74 M_RES(mp)->tr_attrsetrt.tr_logres * nblks;
75 }
76
77 /*
78 * Compute an alternate set of log reservation sizes for use exclusively with
79 * minimum log size calculations.
80 */
81 static void
82 xfs_log_calc_trans_resv_for_minlogblocks(
83 struct xfs_mount *mp,
84 struct xfs_trans_resv *resv)
85 {
86 unsigned int rmap_maxlevels = mp->m_rmap_maxlevels;
87
88 /*
89 * If the feature set is new enough, drop the oversized minimum log
90 * size computation introduced by the original reflink code.
91 */
92 if (xfs_want_minlogsize_fixes(&mp->m_sb)) {
93 xfs_trans_resv_calc(mp, resv);
94 return;
95 }
96
97 /*
98 * In the early days of rmap+reflink, we always set the rmap maxlevels
99 * to 9 even if the AG was small enough that it would never grow to
100 * that height. Transaction reservation sizes influence the minimum
101 * log size calculation, which influences the size of the log that mkfs
102 * creates. Use the old value here to ensure that newly formatted
103 * small filesystems will mount on older kernels.
104 */
105 if (xfs_has_rmapbt(mp) && xfs_has_reflink(mp))
106 mp->m_rmap_maxlevels = XFS_OLD_REFLINK_RMAP_MAXLEVELS;
107
108 xfs_trans_resv_calc(mp, resv);
109
110 if (xfs_has_reflink(mp)) {
111 /*
112 * In the early days of reflink, typical log operation counts
113 * were greatly overestimated.
114 */
115 resv->tr_write.tr_logcount = XFS_WRITE_LOG_COUNT_REFLINK;
116 resv->tr_itruncate.tr_logcount =
117 XFS_ITRUNCATE_LOG_COUNT_REFLINK;
118 resv->tr_qm_dqalloc.tr_logcount = XFS_WRITE_LOG_COUNT_REFLINK;
119 } else if (xfs_has_rmapbt(mp)) {
120 /*
121 * In the early days of non-reflink rmap, the impact of rmapbt
122 * updates on log counts were not taken into account at all.
123 */
124 resv->tr_write.tr_logcount = XFS_WRITE_LOG_COUNT;
125 resv->tr_itruncate.tr_logcount = XFS_ITRUNCATE_LOG_COUNT;
126 resv->tr_qm_dqalloc.tr_logcount = XFS_WRITE_LOG_COUNT;
127 }
128
129 /*
130 * In the early days of reflink, we did not use deferred refcount
131 * update log items, so log reservations must be recomputed using the
132 * old calculations.
133 */
134 resv->tr_write.tr_logres =
135 xfs_calc_write_reservation_minlogsize(mp);
136 resv->tr_itruncate.tr_logres =
137 xfs_calc_itruncate_reservation_minlogsize(mp);
138 resv->tr_qm_dqalloc.tr_logres =
139 xfs_calc_qm_dqalloc_reservation_minlogsize(mp);
140
141 /* Put everything back the way it was. This goes at the end. */
142 mp->m_rmap_maxlevels = rmap_maxlevels;
143 }
144
145 /*
146 * Iterate over the log space reservation table to figure out and return
147 * the maximum one in terms of the pre-calculated values which were done
148 * at mount time.
149 */
150 void
151 xfs_log_get_max_trans_res(
152 struct xfs_mount *mp,
153 struct xfs_trans_res *max_resp)
154 {
155 struct xfs_trans_resv resv = {};
156 struct xfs_trans_res *resp;
157 struct xfs_trans_res *end_resp;
158 unsigned int i;
159 int log_space = 0;
160 int attr_space;
161
162 attr_space = xfs_log_calc_max_attrsetm_res(mp);
163
164 xfs_log_calc_trans_resv_for_minlogblocks(mp, &resv);
165
166 resp = (struct xfs_trans_res *)&resv;
167 end_resp = (struct xfs_trans_res *)(&resv + 1);
168 for (i = 0; resp < end_resp; i++, resp++) {
169 int tmp = resp->tr_logcount > 1 ?
170 resp->tr_logres * resp->tr_logcount :
171 resp->tr_logres;
172
173 trace_xfs_trans_resv_calc_minlogsize(mp, i, resp);
174 if (log_space < tmp) {
175 log_space = tmp;
176 *max_resp = *resp; /* struct copy */
177 }
178 }
179
180 if (attr_space > log_space) {
181 *max_resp = resv.tr_attrsetm; /* struct copy */
182 max_resp->tr_logres = attr_space;
183 }
184 trace_xfs_log_get_max_trans_res(mp, max_resp);
185 }
186
187 /*
188 * Calculate the minimum valid log size for the given superblock configuration.
189 * Used to calculate the minimum log size at mkfs time, and to determine if
190 * the log is large enough or not at mount time. Returns the minimum size in
191 * filesystem block size units.
192 */
193 int
194 xfs_log_calc_minimum_size(
195 struct xfs_mount *mp)
196 {
197 struct xfs_trans_res tres = {0};
198 int max_logres;
199 int min_logblks = 0;
200 int lsunit = 0;
201
202 xfs_log_get_max_trans_res(mp, &tres);
203
204 max_logres = xfs_log_calc_unit_res(mp, tres.tr_logres);
205 if (tres.tr_logcount > 1)
206 max_logres *= tres.tr_logcount;
207
208 if (xfs_has_logv2(mp) && mp->m_sb.sb_logsunit > 1)
209 lsunit = BTOBB(mp->m_sb.sb_logsunit);
210
211 /*
212 * Two factors should be taken into account for calculating the minimum
213 * log space.
214 * 1) The fundamental limitation is that no single transaction can be
215 * larger than half size of the log.
216 *
217 * From mkfs.xfs, this is considered by the XFS_MIN_LOG_FACTOR
218 * define, which is set to 3. That means we can definitely fit
219 * maximally sized 2 transactions in the log. We'll use this same
220 * value here.
221 *
222 * 2) If the lsunit option is specified, a transaction requires 2 LSU
223 * for the reservation because there are two log writes that can
224 * require padding - the transaction data and the commit record which
225 * are written separately and both can require padding to the LSU.
226 * Consider that we can have an active CIL reservation holding 2*LSU,
227 * but the CIL is not over a push threshold, in this case, if we
228 * don't have enough log space for at one new transaction, which
229 * includes another 2*LSU in the reservation, we will run into dead
230 * loop situation in log space grant procedure. i.e.
231 * xlog_grant_head_wait().
232 *
233 * Hence the log size needs to be able to contain two maximally sized
234 * and padded transactions, which is (2 * (2 * LSU + maxlres)).
235 *
236 * Also, the log size should be a multiple of the log stripe unit, round
237 * it up to lsunit boundary if lsunit is specified.
238 */
239 if (lsunit) {
240 min_logblks = roundup_64(BTOBB(max_logres), lsunit) +
241 2 * lsunit;
242 } else
243 min_logblks = BTOBB(max_logres) + 2 * BBSIZE;
244 min_logblks *= XFS_MIN_LOG_FACTOR;
245
246 return XFS_BB_TO_FSB(mp, min_logblks);
247 }
248
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.