1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2008-2010, 2013 Dave Chinner
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_inode.h"
14 #include "xfs_trans.h"
15 #include "xfs_trans_priv.h"
16 #include "xfs_icreate_item.h"
17 #include "xfs_log.h"
18 #include "xfs_log_priv.h"
19 #include "xfs_log_recover.h"
20 #include "xfs_ialloc.h"
21 #include "xfs_trace.h"
22
23 struct kmem_cache *xfs_icreate_cache; /* inode create item */
24
ICR_ITEM(struct xfs_log_item * lip)25 static inline struct xfs_icreate_item *ICR_ITEM(struct xfs_log_item *lip)
26 {
27 return container_of(lip, struct xfs_icreate_item, ic_item);
28 }
29
30 /*
31 * This returns the number of iovecs needed to log the given inode item.
32 *
33 * We only need one iovec for the icreate log structure.
34 */
35 STATIC void
xfs_icreate_item_size(struct xfs_log_item * lip,int * nvecs,int * nbytes)36 xfs_icreate_item_size(
37 struct xfs_log_item *lip,
38 int *nvecs,
39 int *nbytes)
40 {
41 *nvecs += 1;
42 *nbytes += sizeof(struct xfs_icreate_log);
43 }
44
45 /*
46 * This is called to fill in the vector of log iovecs for the
47 * given inode create log item.
48 */
49 STATIC void
xfs_icreate_item_format(struct xfs_log_item * lip,struct xfs_log_vec * lv)50 xfs_icreate_item_format(
51 struct xfs_log_item *lip,
52 struct xfs_log_vec *lv)
53 {
54 struct xfs_icreate_item *icp = ICR_ITEM(lip);
55 struct xfs_log_iovec *vecp = NULL;
56
57 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_ICREATE,
58 &icp->ic_format,
59 sizeof(struct xfs_icreate_log));
60 }
61
62 STATIC void
xfs_icreate_item_release(struct xfs_log_item * lip)63 xfs_icreate_item_release(
64 struct xfs_log_item *lip)
65 {
66 kmem_free(ICR_ITEM(lip)->ic_item.li_lv_shadow);
67 kmem_cache_free(xfs_icreate_cache, ICR_ITEM(lip));
68 }
69
70 static const struct xfs_item_ops xfs_icreate_item_ops = {
71 .flags = XFS_ITEM_RELEASE_WHEN_COMMITTED,
72 .iop_size = xfs_icreate_item_size,
73 .iop_format = xfs_icreate_item_format,
74 .iop_release = xfs_icreate_item_release,
75 };
76
77
78 /*
79 * Initialize the inode log item for a newly allocated (in-core) inode.
80 *
81 * Inode extents can only reside within an AG. Hence specify the starting
82 * block for the inode chunk by offset within an AG as well as the
83 * length of the allocated extent.
84 *
85 * This joins the item to the transaction and marks it dirty so
86 * that we don't need a separate call to do this, nor does the
87 * caller need to know anything about the icreate item.
88 */
89 void
xfs_icreate_log(struct xfs_trans * tp,xfs_agnumber_t agno,xfs_agblock_t agbno,unsigned int count,unsigned int inode_size,xfs_agblock_t length,unsigned int generation)90 xfs_icreate_log(
91 struct xfs_trans *tp,
92 xfs_agnumber_t agno,
93 xfs_agblock_t agbno,
94 unsigned int count,
95 unsigned int inode_size,
96 xfs_agblock_t length,
97 unsigned int generation)
98 {
99 struct xfs_icreate_item *icp;
100
101 icp = kmem_cache_zalloc(xfs_icreate_cache, GFP_KERNEL | __GFP_NOFAIL);
102
103 xfs_log_item_init(tp->t_mountp, &icp->ic_item, XFS_LI_ICREATE,
104 &xfs_icreate_item_ops);
105
106 icp->ic_format.icl_type = XFS_LI_ICREATE;
107 icp->ic_format.icl_size = 1; /* single vector */
108 icp->ic_format.icl_ag = cpu_to_be32(agno);
109 icp->ic_format.icl_agbno = cpu_to_be32(agbno);
110 icp->ic_format.icl_count = cpu_to_be32(count);
111 icp->ic_format.icl_isize = cpu_to_be32(inode_size);
112 icp->ic_format.icl_length = cpu_to_be32(length);
113 icp->ic_format.icl_gen = cpu_to_be32(generation);
114
115 xfs_trans_add_item(tp, &icp->ic_item);
116 tp->t_flags |= XFS_TRANS_DIRTY;
117 set_bit(XFS_LI_DIRTY, &icp->ic_item.li_flags);
118 }
119
120 static enum xlog_recover_reorder
xlog_recover_icreate_reorder(struct xlog_recover_item * item)121 xlog_recover_icreate_reorder(
122 struct xlog_recover_item *item)
123 {
124 /*
125 * Inode allocation buffers must be replayed before subsequent inode
126 * items try to modify those buffers. ICREATE items are the logical
127 * equivalent of logging a newly initialized inode buffer, so recover
128 * these at the same time that we recover logged buffers.
129 */
130 return XLOG_REORDER_BUFFER_LIST;
131 }
132
133 /*
134 * This routine is called when an inode create format structure is found in a
135 * committed transaction in the log. It's purpose is to initialise the inodes
136 * being allocated on disk. This requires us to get inode cluster buffers that
137 * match the range to be initialised, stamped with inode templates and written
138 * by delayed write so that subsequent modifications will hit the cached buffer
139 * and only need writing out at the end of recovery.
140 */
141 STATIC int
xlog_recover_icreate_commit_pass2(struct xlog * log,struct list_head * buffer_list,struct xlog_recover_item * item,xfs_lsn_t lsn)142 xlog_recover_icreate_commit_pass2(
143 struct xlog *log,
144 struct list_head *buffer_list,
145 struct xlog_recover_item *item,
146 xfs_lsn_t lsn)
147 {
148 struct xfs_mount *mp = log->l_mp;
149 struct xfs_icreate_log *icl;
150 struct xfs_ino_geometry *igeo = M_IGEO(mp);
151 xfs_agnumber_t agno;
152 xfs_agblock_t agbno;
153 unsigned int count;
154 unsigned int isize;
155 xfs_agblock_t length;
156 int bb_per_cluster;
157 int cancel_count;
158 int nbufs;
159 int i;
160
161 icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
162 if (icl->icl_type != XFS_LI_ICREATE) {
163 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
164 return -EINVAL;
165 }
166
167 if (icl->icl_size != 1) {
168 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
169 return -EINVAL;
170 }
171
172 agno = be32_to_cpu(icl->icl_ag);
173 if (agno >= mp->m_sb.sb_agcount) {
174 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
175 return -EINVAL;
176 }
177 agbno = be32_to_cpu(icl->icl_agbno);
178 if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
179 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
180 return -EINVAL;
181 }
182 isize = be32_to_cpu(icl->icl_isize);
183 if (isize != mp->m_sb.sb_inodesize) {
184 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
185 return -EINVAL;
186 }
187 count = be32_to_cpu(icl->icl_count);
188 if (!count) {
189 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
190 return -EINVAL;
191 }
192 length = be32_to_cpu(icl->icl_length);
193 if (!length || length >= mp->m_sb.sb_agblocks) {
194 xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
195 return -EINVAL;
196 }
197
198 /*
199 * The inode chunk is either full or sparse and we only support
200 * m_ino_geo.ialloc_min_blks sized sparse allocations at this time.
201 */
202 if (length != igeo->ialloc_blks &&
203 length != igeo->ialloc_min_blks) {
204 xfs_warn(log->l_mp,
205 "%s: unsupported chunk length", __func__);
206 return -EINVAL;
207 }
208
209 /* verify inode count is consistent with extent length */
210 if ((count >> mp->m_sb.sb_inopblog) != length) {
211 xfs_warn(log->l_mp,
212 "%s: inconsistent inode count and chunk length",
213 __func__);
214 return -EINVAL;
215 }
216
217 /*
218 * The icreate transaction can cover multiple cluster buffers and these
219 * buffers could have been freed and reused. Check the individual
220 * buffers for cancellation so we don't overwrite anything written after
221 * a cancellation.
222 */
223 bb_per_cluster = XFS_FSB_TO_BB(mp, igeo->blocks_per_cluster);
224 nbufs = length / igeo->blocks_per_cluster;
225 for (i = 0, cancel_count = 0; i < nbufs; i++) {
226 xfs_daddr_t daddr;
227
228 daddr = XFS_AGB_TO_DADDR(mp, agno,
229 agbno + i * igeo->blocks_per_cluster);
230 if (xlog_is_buffer_cancelled(log, daddr, bb_per_cluster))
231 cancel_count++;
232 }
233
234 /*
235 * We currently only use icreate for a single allocation at a time. This
236 * means we should expect either all or none of the buffers to be
237 * cancelled. Be conservative and skip replay if at least one buffer is
238 * cancelled, but warn the user that something is awry if the buffers
239 * are not consistent.
240 *
241 * XXX: This must be refined to only skip cancelled clusters once we use
242 * icreate for multiple chunk allocations.
243 */
244 ASSERT(!cancel_count || cancel_count == nbufs);
245 if (cancel_count) {
246 if (cancel_count != nbufs)
247 xfs_warn(mp,
248 "WARNING: partial inode chunk cancellation, skipped icreate.");
249 trace_xfs_log_recover_icreate_cancel(log, icl);
250 return 0;
251 }
252
253 trace_xfs_log_recover_icreate_recover(log, icl);
254 return xfs_ialloc_inode_init(mp, NULL, buffer_list, count, agno, agbno,
255 length, be32_to_cpu(icl->icl_gen));
256 }
257
258 const struct xlog_recover_item_ops xlog_icreate_item_ops = {
259 .item_type = XFS_LI_ICREATE,
260 .reorder = xlog_recover_icreate_reorder,
261 .commit_pass2 = xlog_recover_icreate_commit_pass2,
262 };
263