| @@ -1,3184 +1,3235 @@ | | | @@ -1,3184 +1,3235 @@ |
1 | /* $NetBSD: vfs_wapbl.c,v 1.86 2016/11/10 20:56:32 jdolecek Exp $ */ | | 1 | /* $NetBSD: vfs_wapbl.c,v 1.87 2017/03/05 13:57:29 jdolecek Exp $ */ |
2 | | | 2 | |
3 | /*- | | 3 | /*- |
4 | * Copyright (c) 2003, 2008, 2009 The NetBSD Foundation, Inc. | | 4 | * Copyright (c) 2003, 2008, 2009 The NetBSD Foundation, Inc. |
5 | * All rights reserved. | | 5 | * All rights reserved. |
6 | * | | 6 | * |
7 | * This code is derived from software contributed to The NetBSD Foundation | | 7 | * This code is derived from software contributed to The NetBSD Foundation |
8 | * by Wasabi Systems, Inc. | | 8 | * by Wasabi Systems, Inc. |
9 | * | | 9 | * |
10 | * Redistribution and use in source and binary forms, with or without | | 10 | * Redistribution and use in source and binary forms, with or without |
11 | * modification, are permitted provided that the following conditions | | 11 | * modification, are permitted provided that the following conditions |
12 | * are met: | | 12 | * are met: |
13 | * 1. Redistributions of source code must retain the above copyright | | 13 | * 1. Redistributions of source code must retain the above copyright |
14 | * notice, this list of conditions and the following disclaimer. | | 14 | * notice, this list of conditions and the following disclaimer. |
15 | * 2. Redistributions in binary form must reproduce the above copyright | | 15 | * 2. Redistributions in binary form must reproduce the above copyright |
16 | * notice, this list of conditions and the following disclaimer in the | | 16 | * notice, this list of conditions and the following disclaimer in the |
17 | * documentation and/or other materials provided with the distribution. | | 17 | * documentation and/or other materials provided with the distribution. |
18 | * | | 18 | * |
19 | * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS | | 19 | * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS |
20 | * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED | | 20 | * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED |
21 | * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR | | 21 | * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR |
22 | * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS | | 22 | * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS |
23 | * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR | | 23 | * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR |
24 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF | | 24 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF |
25 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS | | 25 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS |
26 | * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN | | 26 | * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN |
27 | * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) | | 27 | * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
28 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE | | 28 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
29 | * POSSIBILITY OF SUCH DAMAGE. | | 29 | * POSSIBILITY OF SUCH DAMAGE. |
30 | */ | | 30 | */ |
31 | | | 31 | |
32 | /* | | 32 | /* |
33 | * This implements file system independent write ahead filesystem logging. | | 33 | * This implements file system independent write ahead filesystem logging. |
34 | */ | | 34 | */ |
35 | | | 35 | |
36 | #define WAPBL_INTERNAL | | 36 | #define WAPBL_INTERNAL |
37 | | | 37 | |
38 | #include <sys/cdefs.h> | | 38 | #include <sys/cdefs.h> |
39 | __KERNEL_RCSID(0, "$NetBSD: vfs_wapbl.c,v 1.86 2016/11/10 20:56:32 jdolecek Exp $"); | | 39 | __KERNEL_RCSID(0, "$NetBSD: vfs_wapbl.c,v 1.87 2017/03/05 13:57:29 jdolecek Exp $"); |
40 | | | 40 | |
41 | #include <sys/param.h> | | 41 | #include <sys/param.h> |
42 | #include <sys/bitops.h> | | 42 | #include <sys/bitops.h> |
43 | #include <sys/time.h> | | 43 | #include <sys/time.h> |
44 | #include <sys/wapbl.h> | | 44 | #include <sys/wapbl.h> |
45 | #include <sys/wapbl_replay.h> | | 45 | #include <sys/wapbl_replay.h> |
46 | | | 46 | |
47 | #ifdef _KERNEL | | 47 | #ifdef _KERNEL |
48 | | | 48 | |
49 | #include <sys/atomic.h> | | 49 | #include <sys/atomic.h> |
50 | #include <sys/conf.h> | | 50 | #include <sys/conf.h> |
51 | #include <sys/file.h> | | 51 | #include <sys/file.h> |
52 | #include <sys/kauth.h> | | 52 | #include <sys/kauth.h> |
53 | #include <sys/kernel.h> | | 53 | #include <sys/kernel.h> |
54 | #include <sys/module.h> | | 54 | #include <sys/module.h> |
55 | #include <sys/mount.h> | | 55 | #include <sys/mount.h> |
56 | #include <sys/mutex.h> | | 56 | #include <sys/mutex.h> |
57 | #include <sys/namei.h> | | 57 | #include <sys/namei.h> |
58 | #include <sys/proc.h> | | 58 | #include <sys/proc.h> |
59 | #include <sys/resourcevar.h> | | 59 | #include <sys/resourcevar.h> |
60 | #include <sys/sysctl.h> | | 60 | #include <sys/sysctl.h> |
61 | #include <sys/uio.h> | | 61 | #include <sys/uio.h> |
62 | #include <sys/vnode.h> | | 62 | #include <sys/vnode.h> |
63 | | | 63 | |
64 | #include <miscfs/specfs/specdev.h> | | 64 | #include <miscfs/specfs/specdev.h> |
65 | | | 65 | |
66 | #define wapbl_alloc(s) kmem_alloc((s), KM_SLEEP) | | 66 | #define wapbl_alloc(s) kmem_alloc((s), KM_SLEEP) |
67 | #define wapbl_free(a, s) kmem_free((a), (s)) | | 67 | #define wapbl_free(a, s) kmem_free((a), (s)) |
68 | #define wapbl_calloc(n, s) kmem_zalloc((n)*(s), KM_SLEEP) | | 68 | #define wapbl_calloc(n, s) kmem_zalloc((n)*(s), KM_SLEEP) |
69 | | | 69 | |
70 | static struct sysctllog *wapbl_sysctl; | | 70 | static struct sysctllog *wapbl_sysctl; |
71 | static int wapbl_flush_disk_cache = 1; | | 71 | static int wapbl_flush_disk_cache = 1; |
72 | static int wapbl_verbose_commit = 0; | | 72 | static int wapbl_verbose_commit = 0; |
73 | | | 73 | |
74 | static inline size_t wapbl_space_free(size_t, off_t, off_t); | | 74 | static inline size_t wapbl_space_free(size_t, off_t, off_t); |
75 | | | 75 | |
76 | #else /* !_KERNEL */ | | 76 | #else /* !_KERNEL */ |
77 | | | 77 | |
78 | #include <assert.h> | | 78 | #include <assert.h> |
79 | #include <errno.h> | | 79 | #include <errno.h> |
80 | #include <stdbool.h> | | 80 | #include <stdbool.h> |
81 | #include <stdio.h> | | 81 | #include <stdio.h> |
82 | #include <stdlib.h> | | 82 | #include <stdlib.h> |
83 | #include <string.h> | | 83 | #include <string.h> |
84 | | | 84 | |
85 | #define KDASSERT(x) assert(x) | | 85 | #define KDASSERT(x) assert(x) |
86 | #define KASSERT(x) assert(x) | | 86 | #define KASSERT(x) assert(x) |
87 | #define wapbl_alloc(s) malloc(s) | | 87 | #define wapbl_alloc(s) malloc(s) |
88 | #define wapbl_free(a, s) free(a) | | 88 | #define wapbl_free(a, s) free(a) |
89 | #define wapbl_calloc(n, s) calloc((n), (s)) | | 89 | #define wapbl_calloc(n, s) calloc((n), (s)) |
90 | | | 90 | |
91 | #endif /* !_KERNEL */ | | 91 | #endif /* !_KERNEL */ |
92 | | | 92 | |
93 | /* | | 93 | /* |
94 | * INTERNAL DATA STRUCTURES | | 94 | * INTERNAL DATA STRUCTURES |
95 | */ | | 95 | */ |
96 | | | 96 | |
97 | /* | | 97 | /* |
98 | * This structure holds per-mount log information. | | 98 | * This structure holds per-mount log information. |
99 | * | | 99 | * |
100 | * Legend: a = atomic access only | | 100 | * Legend: a = atomic access only |
101 | * r = read-only after init | | 101 | * r = read-only after init |
102 | * l = rwlock held | | 102 | * l = rwlock held |
103 | * m = mutex held | | 103 | * m = mutex held |
104 | * lm = rwlock held writing or mutex held | | 104 | * lm = rwlock held writing or mutex held |
105 | * u = unlocked access ok | | 105 | * u = unlocked access ok |
106 | * b = bufcache_lock held | | 106 | * b = bufcache_lock held |
107 | */ | | 107 | */ |
108 | LIST_HEAD(wapbl_ino_head, wapbl_ino); | | 108 | LIST_HEAD(wapbl_ino_head, wapbl_ino); |
109 | struct wapbl { | | 109 | struct wapbl { |
110 | struct vnode *wl_logvp; /* r: log here */ | | 110 | struct vnode *wl_logvp; /* r: log here */ |
111 | struct vnode *wl_devvp; /* r: log on this device */ | | 111 | struct vnode *wl_devvp; /* r: log on this device */ |
112 | struct mount *wl_mount; /* r: mountpoint wl is associated with */ | | 112 | struct mount *wl_mount; /* r: mountpoint wl is associated with */ |
113 | daddr_t wl_logpbn; /* r: Physical block number of start of log */ | | 113 | daddr_t wl_logpbn; /* r: Physical block number of start of log */ |
114 | int wl_log_dev_bshift; /* r: logarithm of device block size of log | | 114 | int wl_log_dev_bshift; /* r: logarithm of device block size of log |
115 | device */ | | 115 | device */ |
116 | int wl_fs_dev_bshift; /* r: logarithm of device block size of | | 116 | int wl_fs_dev_bshift; /* r: logarithm of device block size of |
117 | filesystem device */ | | 117 | filesystem device */ |
118 | | | 118 | |
119 | unsigned wl_lock_count; /* m: Count of transactions in progress */ | | 119 | unsigned wl_lock_count; /* m: Count of transactions in progress */ |
120 | | | 120 | |
121 | size_t wl_circ_size; /* r: Number of bytes in buffer of log */ | | 121 | size_t wl_circ_size; /* r: Number of bytes in buffer of log */ |
122 | size_t wl_circ_off; /* r: Number of bytes reserved at start */ | | 122 | size_t wl_circ_off; /* r: Number of bytes reserved at start */ |
123 | | | 123 | |
124 | size_t wl_bufcount_max; /* r: Number of buffers reserved for log */ | | 124 | size_t wl_bufcount_max; /* r: Number of buffers reserved for log */ |
125 | size_t wl_bufbytes_max; /* r: Number of buf bytes reserved for log */ | | 125 | size_t wl_bufbytes_max; /* r: Number of buf bytes reserved for log */ |
126 | | | 126 | |
127 | off_t wl_head; /* l: Byte offset of log head */ | | 127 | off_t wl_head; /* l: Byte offset of log head */ |
128 | off_t wl_tail; /* l: Byte offset of log tail */ | | 128 | off_t wl_tail; /* l: Byte offset of log tail */ |
129 | /* | | 129 | /* |
130 | * WAPBL log layout, stored on wl_devvp at wl_logpbn: | | 130 | * WAPBL log layout, stored on wl_devvp at wl_logpbn: |
131 | * | | 131 | * |
132 | * ___________________ wl_circ_size __________________ | | 132 | * ___________________ wl_circ_size __________________ |
133 | * / \ | | 133 | * / \ |
134 | * +---------+---------+-------+--------------+--------+ | | 134 | * +---------+---------+-------+--------------+--------+ |
135 | * [ commit0 | commit1 | CCWCW | EEEEEEEEEEEE | CCCWCW ] | | 135 | * [ commit0 | commit1 | CCWCW | EEEEEEEEEEEE | CCCWCW ] |
136 | * +---------+---------+-------+--------------+--------+ | | 136 | * +---------+---------+-------+--------------+--------+ |
137 | * wl_circ_off --^ ^-- wl_head ^-- wl_tail | | 137 | * wl_circ_off --^ ^-- wl_head ^-- wl_tail |
138 | * | | 138 | * |
139 | * commit0 and commit1 are commit headers. A commit header has | | 139 | * commit0 and commit1 are commit headers. A commit header has |
140 | * a generation number, indicating which of the two headers is | | 140 | * a generation number, indicating which of the two headers is |
141 | * more recent, and an assignment of head and tail pointers. | | 141 | * more recent, and an assignment of head and tail pointers. |
142 | * The rest is a circular queue of log records, starting at | | 142 | * The rest is a circular queue of log records, starting at |
143 | * the byte offset wl_circ_off. | | 143 | * the byte offset wl_circ_off. |
144 | * | | 144 | * |
145 | * E marks empty space for records. | | 145 | * E marks empty space for records. |
146 | * W marks records for block writes issued but waiting. | | 146 | * W marks records for block writes issued but waiting. |
147 | * C marks completed records. | | 147 | * C marks completed records. |
148 | * | | 148 | * |
149 | * wapbl_flush writes new records to empty `E' spaces after | | 149 | * wapbl_flush writes new records to empty `E' spaces after |
150 | * wl_head from the current transaction in memory. | | 150 | * wl_head from the current transaction in memory. |
151 | * | | 151 | * |
152 | * wapbl_truncate advances wl_tail past any completed `C' | | 152 | * wapbl_truncate advances wl_tail past any completed `C' |
153 | * records, freeing them up for use. | | 153 | * records, freeing them up for use. |
154 | * | | 154 | * |
155 | * head == tail == 0 means log is empty. | | 155 | * head == tail == 0 means log is empty. |
156 | * head == tail != 0 means log is full. | | 156 | * head == tail != 0 means log is full. |
157 | * | | 157 | * |
158 | * See assertions in wapbl_advance() for other boundary | | 158 | * See assertions in wapbl_advance() for other boundary |
159 | * conditions. | | 159 | * conditions. |
160 | * | | 160 | * |
161 | * Only wapbl_flush moves the head, except when wapbl_truncate | | 161 | * Only wapbl_flush moves the head, except when wapbl_truncate |
162 | * sets it to 0 to indicate that the log is empty. | | 162 | * sets it to 0 to indicate that the log is empty. |
163 | * | | 163 | * |
164 | * Only wapbl_truncate moves the tail, except when wapbl_flush | | 164 | * Only wapbl_truncate moves the tail, except when wapbl_flush |
165 | * sets it to wl_circ_off to indicate that the log is full. | | 165 | * sets it to wl_circ_off to indicate that the log is full. |
166 | */ | | 166 | */ |
167 | | | 167 | |
168 | struct wapbl_wc_header *wl_wc_header; /* l */ | | 168 | struct wapbl_wc_header *wl_wc_header; /* l */ |
169 | void *wl_wc_scratch; /* l: scratch space (XXX: por que?!?) */ | | 169 | void *wl_wc_scratch; /* l: scratch space (XXX: por que?!?) */ |
170 | | | 170 | |
171 | kmutex_t wl_mtx; /* u: short-term lock */ | | 171 | kmutex_t wl_mtx; /* u: short-term lock */ |
172 | krwlock_t wl_rwlock; /* u: File system transaction lock */ | | 172 | krwlock_t wl_rwlock; /* u: File system transaction lock */ |
173 | | | 173 | |
174 | /* | | 174 | /* |
175 | * Must be held while accessing | | 175 | * Must be held while accessing |
176 | * wl_count or wl_bufs or head or tail | | 176 | * wl_count or wl_bufs or head or tail |
177 | */ | | 177 | */ |
178 | | | 178 | |
| | | 179 | #if _KERNEL |
179 | /* | | 180 | /* |
180 | * Callback called from within the flush routine to flush any extra | | 181 | * Callback called from within the flush routine to flush any extra |
181 | * bits. Note that flush may be skipped without calling this if | | 182 | * bits. Note that flush may be skipped without calling this if |
182 | * there are no outstanding buffers in the transaction. | | 183 | * there are no outstanding buffers in the transaction. |
183 | */ | | 184 | */ |
184 | #if _KERNEL | | | |
185 | wapbl_flush_fn_t wl_flush; /* r */ | | 185 | wapbl_flush_fn_t wl_flush; /* r */ |
186 | wapbl_flush_fn_t wl_flush_abort;/* r */ | | 186 | wapbl_flush_fn_t wl_flush_abort;/* r */ |
| | | 187 | |
| | | 188 | /* Event counters */ |
| | | 189 | char wl_ev_group[EVCNT_STRING_MAX]; /* r */ |
| | | 190 | struct evcnt wl_ev_commit; /* l */ |
| | | 191 | struct evcnt wl_ev_journalwrite; /* l */ |
| | | 192 | struct evcnt wl_ev_metawrite; /* lm */ |
| | | 193 | struct evcnt wl_ev_cacheflush; /* l */ |
187 | #endif | | 194 | #endif |
188 | | | 195 | |
189 | size_t wl_bufbytes; /* m: Byte count of pages in wl_bufs */ | | 196 | size_t wl_bufbytes; /* m: Byte count of pages in wl_bufs */ |
190 | size_t wl_bufcount; /* m: Count of buffers in wl_bufs */ | | 197 | size_t wl_bufcount; /* m: Count of buffers in wl_bufs */ |
191 | size_t wl_bcount; /* m: Total bcount of wl_bufs */ | | 198 | size_t wl_bcount; /* m: Total bcount of wl_bufs */ |
192 | | | 199 | |
193 | LIST_HEAD(, buf) wl_bufs; /* m: Buffers in current transaction */ | | 200 | LIST_HEAD(, buf) wl_bufs; /* m: Buffers in current transaction */ |
194 | | | 201 | |
195 | kcondvar_t wl_reclaimable_cv; /* m (obviously) */ | | 202 | kcondvar_t wl_reclaimable_cv; /* m (obviously) */ |
196 | size_t wl_reclaimable_bytes; /* m: Amount of space available for | | 203 | size_t wl_reclaimable_bytes; /* m: Amount of space available for |
197 | reclamation by truncate */ | | 204 | reclamation by truncate */ |
198 | int wl_error_count; /* m: # of wl_entries with errors */ | | 205 | int wl_error_count; /* m: # of wl_entries with errors */ |
199 | size_t wl_reserved_bytes; /* never truncate log smaller than this */ | | 206 | size_t wl_reserved_bytes; /* never truncate log smaller than this */ |
200 | | | 207 | |
201 | #ifdef WAPBL_DEBUG_BUFBYTES | | 208 | #ifdef WAPBL_DEBUG_BUFBYTES |
202 | size_t wl_unsynced_bufbytes; /* Byte count of unsynced buffers */ | | 209 | size_t wl_unsynced_bufbytes; /* Byte count of unsynced buffers */ |
203 | #endif | | 210 | #endif |
204 | | | 211 | |
205 | #if _KERNEL | | 212 | #if _KERNEL |
206 | int wl_brperjblock; /* r Block records per journal block */ | | 213 | int wl_brperjblock; /* r Block records per journal block */ |
207 | #endif | | 214 | #endif |
208 | | | 215 | |
209 | TAILQ_HEAD(, wapbl_dealloc) wl_dealloclist; /* lm: list head */ | | 216 | TAILQ_HEAD(, wapbl_dealloc) wl_dealloclist; /* lm: list head */ |
210 | int wl_dealloccnt; /* lm: total count */ | | 217 | int wl_dealloccnt; /* lm: total count */ |
211 | int wl_dealloclim; /* r: max count */ | | 218 | int wl_dealloclim; /* r: max count */ |
212 | | | 219 | |
213 | /* hashtable of inode numbers for allocated but unlinked inodes */ | | 220 | /* hashtable of inode numbers for allocated but unlinked inodes */ |
214 | /* synch ??? */ | | 221 | /* synch ??? */ |
215 | struct wapbl_ino_head *wl_inohash; | | 222 | struct wapbl_ino_head *wl_inohash; |
216 | u_long wl_inohashmask; | | 223 | u_long wl_inohashmask; |
217 | int wl_inohashcnt; | | 224 | int wl_inohashcnt; |
218 | | | 225 | |
219 | SIMPLEQ_HEAD(, wapbl_entry) wl_entries; /* On disk transaction | | 226 | SIMPLEQ_HEAD(, wapbl_entry) wl_entries; /* On disk transaction |
220 | accounting */ | | 227 | accounting */ |
221 | | | 228 | |
222 | u_char *wl_buffer; /* l: buffer for wapbl_buffered_write() */ | | 229 | u_char *wl_buffer; /* l: buffer for wapbl_buffered_write() */ |
223 | daddr_t wl_buffer_dblk; /* l: buffer disk block address */ | | 230 | daddr_t wl_buffer_dblk; /* l: buffer disk block address */ |
224 | size_t wl_buffer_used; /* l: buffer current use */ | | 231 | size_t wl_buffer_used; /* l: buffer current use */ |
225 | }; | | 232 | }; |
226 | | | 233 | |
227 | #ifdef WAPBL_DEBUG_PRINT | | 234 | #ifdef WAPBL_DEBUG_PRINT |
228 | int wapbl_debug_print = WAPBL_DEBUG_PRINT; | | 235 | int wapbl_debug_print = WAPBL_DEBUG_PRINT; |
229 | #endif | | 236 | #endif |
230 | | | 237 | |
231 | /****************************************************************/ | | 238 | /****************************************************************/ |
232 | #ifdef _KERNEL | | 239 | #ifdef _KERNEL |
233 | | | 240 | |
234 | #ifdef WAPBL_DEBUG | | 241 | #ifdef WAPBL_DEBUG |
235 | struct wapbl *wapbl_debug_wl; | | 242 | struct wapbl *wapbl_debug_wl; |
236 | #endif | | 243 | #endif |
237 | | | 244 | |
238 | static int wapbl_write_commit(struct wapbl *wl, off_t head, off_t tail); | | 245 | static int wapbl_write_commit(struct wapbl *wl, off_t head, off_t tail); |
239 | static int wapbl_write_blocks(struct wapbl *wl, off_t *offp); | | 246 | static int wapbl_write_blocks(struct wapbl *wl, off_t *offp); |
240 | static int wapbl_write_revocations(struct wapbl *wl, off_t *offp); | | 247 | static int wapbl_write_revocations(struct wapbl *wl, off_t *offp); |
241 | static int wapbl_write_inodes(struct wapbl *wl, off_t *offp); | | 248 | static int wapbl_write_inodes(struct wapbl *wl, off_t *offp); |
242 | #endif /* _KERNEL */ | | 249 | #endif /* _KERNEL */ |
243 | | | 250 | |
244 | static int wapbl_replay_process(struct wapbl_replay *wr, off_t, off_t); | | 251 | static int wapbl_replay_process(struct wapbl_replay *wr, off_t, off_t); |
245 | | | 252 | |
246 | static inline size_t wapbl_space_used(size_t avail, off_t head, | | 253 | static inline size_t wapbl_space_used(size_t avail, off_t head, |
247 | off_t tail); | | 254 | off_t tail); |
248 | | | 255 | |
249 | #ifdef _KERNEL | | 256 | #ifdef _KERNEL |
250 | | | 257 | |
251 | static struct pool wapbl_entry_pool; | | 258 | static struct pool wapbl_entry_pool; |
252 | static struct pool wapbl_dealloc_pool; | | 259 | static struct pool wapbl_dealloc_pool; |
253 | | | 260 | |
254 | #define WAPBL_INODETRK_SIZE 83 | | 261 | #define WAPBL_INODETRK_SIZE 83 |
255 | static int wapbl_ino_pool_refcount; | | 262 | static int wapbl_ino_pool_refcount; |
256 | static struct pool wapbl_ino_pool; | | 263 | static struct pool wapbl_ino_pool; |
257 | struct wapbl_ino { | | 264 | struct wapbl_ino { |
258 | LIST_ENTRY(wapbl_ino) wi_hash; | | 265 | LIST_ENTRY(wapbl_ino) wi_hash; |
259 | ino_t wi_ino; | | 266 | ino_t wi_ino; |
260 | mode_t wi_mode; | | 267 | mode_t wi_mode; |
261 | }; | | 268 | }; |
262 | | | 269 | |
263 | static void wapbl_inodetrk_init(struct wapbl *wl, u_int size); | | 270 | static void wapbl_inodetrk_init(struct wapbl *wl, u_int size); |
264 | static void wapbl_inodetrk_free(struct wapbl *wl); | | 271 | static void wapbl_inodetrk_free(struct wapbl *wl); |
265 | static struct wapbl_ino *wapbl_inodetrk_get(struct wapbl *wl, ino_t ino); | | 272 | static struct wapbl_ino *wapbl_inodetrk_get(struct wapbl *wl, ino_t ino); |
266 | | | 273 | |
267 | static size_t wapbl_transaction_len(struct wapbl *wl); | | 274 | static size_t wapbl_transaction_len(struct wapbl *wl); |
268 | static inline size_t wapbl_transaction_inodes_len(struct wapbl *wl); | | 275 | static inline size_t wapbl_transaction_inodes_len(struct wapbl *wl); |
269 | | | 276 | |
270 | static void wapbl_deallocation_free(struct wapbl *, struct wapbl_dealloc *, | | 277 | static void wapbl_deallocation_free(struct wapbl *, struct wapbl_dealloc *, |
271 | bool); | | 278 | bool); |
272 | | | 279 | |
| | | 280 | static void wapbl_evcnt_init(struct wapbl *); |
| | | 281 | static void wapbl_evcnt_free(struct wapbl *); |
| | | 282 | |
273 | #if 0 | | 283 | #if 0 |
274 | int wapbl_replay_verify(struct wapbl_replay *, struct vnode *); | | 284 | int wapbl_replay_verify(struct wapbl_replay *, struct vnode *); |
275 | #endif | | 285 | #endif |
276 | | | 286 | |
277 | static int wapbl_replay_isopen1(struct wapbl_replay *); | | 287 | static int wapbl_replay_isopen1(struct wapbl_replay *); |
278 | | | 288 | |
279 | struct wapbl_ops wapbl_ops = { | | 289 | struct wapbl_ops wapbl_ops = { |
280 | .wo_wapbl_discard = wapbl_discard, | | 290 | .wo_wapbl_discard = wapbl_discard, |
281 | .wo_wapbl_replay_isopen = wapbl_replay_isopen1, | | 291 | .wo_wapbl_replay_isopen = wapbl_replay_isopen1, |
282 | .wo_wapbl_replay_can_read = wapbl_replay_can_read, | | 292 | .wo_wapbl_replay_can_read = wapbl_replay_can_read, |
283 | .wo_wapbl_replay_read = wapbl_replay_read, | | 293 | .wo_wapbl_replay_read = wapbl_replay_read, |
284 | .wo_wapbl_add_buf = wapbl_add_buf, | | 294 | .wo_wapbl_add_buf = wapbl_add_buf, |
285 | .wo_wapbl_remove_buf = wapbl_remove_buf, | | 295 | .wo_wapbl_remove_buf = wapbl_remove_buf, |
286 | .wo_wapbl_resize_buf = wapbl_resize_buf, | | 296 | .wo_wapbl_resize_buf = wapbl_resize_buf, |
287 | .wo_wapbl_begin = wapbl_begin, | | 297 | .wo_wapbl_begin = wapbl_begin, |
288 | .wo_wapbl_end = wapbl_end, | | 298 | .wo_wapbl_end = wapbl_end, |
289 | .wo_wapbl_junlock_assert= wapbl_junlock_assert, | | 299 | .wo_wapbl_junlock_assert= wapbl_junlock_assert, |
290 | | | 300 | |
291 | /* XXX: the following is only used to say "this is a wapbl buf" */ | | 301 | /* XXX: the following is only used to say "this is a wapbl buf" */ |
292 | .wo_wapbl_biodone = wapbl_biodone, | | 302 | .wo_wapbl_biodone = wapbl_biodone, |
293 | }; | | 303 | }; |
294 | | | 304 | |
295 | static int | | 305 | static int |
296 | wapbl_sysctl_init(void) | | 306 | wapbl_sysctl_init(void) |
297 | { | | 307 | { |
298 | int rv; | | 308 | int rv; |
299 | const struct sysctlnode *rnode, *cnode; | | 309 | const struct sysctlnode *rnode, *cnode; |
300 | | | 310 | |
301 | wapbl_sysctl = NULL; | | 311 | wapbl_sysctl = NULL; |
302 | | | 312 | |
303 | rv = sysctl_createv(&wapbl_sysctl, 0, NULL, &rnode, | | 313 | rv = sysctl_createv(&wapbl_sysctl, 0, NULL, &rnode, |
304 | CTLFLAG_PERMANENT, | | 314 | CTLFLAG_PERMANENT, |
305 | CTLTYPE_NODE, "wapbl", | | 315 | CTLTYPE_NODE, "wapbl", |
306 | SYSCTL_DESCR("WAPBL journaling options"), | | 316 | SYSCTL_DESCR("WAPBL journaling options"), |
307 | NULL, 0, NULL, 0, | | 317 | NULL, 0, NULL, 0, |
308 | CTL_VFS, CTL_CREATE, CTL_EOL); | | 318 | CTL_VFS, CTL_CREATE, CTL_EOL); |
309 | if (rv) | | 319 | if (rv) |
310 | return rv; | | 320 | return rv; |
311 | | | 321 | |
312 | rv = sysctl_createv(&wapbl_sysctl, 0, &rnode, &cnode, | | 322 | rv = sysctl_createv(&wapbl_sysctl, 0, &rnode, &cnode, |
313 | CTLFLAG_PERMANENT|CTLFLAG_READWRITE, | | 323 | CTLFLAG_PERMANENT|CTLFLAG_READWRITE, |
314 | CTLTYPE_INT, "flush_disk_cache", | | 324 | CTLTYPE_INT, "flush_disk_cache", |
315 | SYSCTL_DESCR("flush disk cache"), | | 325 | SYSCTL_DESCR("flush disk cache"), |
316 | NULL, 0, &wapbl_flush_disk_cache, 0, | | 326 | NULL, 0, &wapbl_flush_disk_cache, 0, |
317 | CTL_CREATE, CTL_EOL); | | 327 | CTL_CREATE, CTL_EOL); |
318 | if (rv) | | 328 | if (rv) |
319 | return rv; | | 329 | return rv; |
320 | | | 330 | |
321 | rv = sysctl_createv(&wapbl_sysctl, 0, &rnode, &cnode, | | 331 | rv = sysctl_createv(&wapbl_sysctl, 0, &rnode, &cnode, |
322 | CTLFLAG_PERMANENT|CTLFLAG_READWRITE, | | 332 | CTLFLAG_PERMANENT|CTLFLAG_READWRITE, |
323 | CTLTYPE_INT, "verbose_commit", | | 333 | CTLTYPE_INT, "verbose_commit", |
324 | SYSCTL_DESCR("show time and size of wapbl log commits"), | | 334 | SYSCTL_DESCR("show time and size of wapbl log commits"), |
325 | NULL, 0, &wapbl_verbose_commit, 0, | | 335 | NULL, 0, &wapbl_verbose_commit, 0, |
326 | CTL_CREATE, CTL_EOL); | | 336 | CTL_CREATE, CTL_EOL); |
327 | return rv; | | 337 | return rv; |
328 | } | | 338 | } |
329 | | | 339 | |
330 | static void | | 340 | static void |
331 | wapbl_init(void) | | 341 | wapbl_init(void) |
332 | { | | 342 | { |
333 | | | 343 | |
334 | pool_init(&wapbl_entry_pool, sizeof(struct wapbl_entry), 0, 0, 0, | | 344 | pool_init(&wapbl_entry_pool, sizeof(struct wapbl_entry), 0, 0, 0, |
335 | "wapblentrypl", &pool_allocator_kmem, IPL_VM); | | 345 | "wapblentrypl", &pool_allocator_kmem, IPL_VM); |
336 | pool_init(&wapbl_dealloc_pool, sizeof(struct wapbl_dealloc), 0, 0, 0, | | 346 | pool_init(&wapbl_dealloc_pool, sizeof(struct wapbl_dealloc), 0, 0, 0, |
337 | "wapbldealloc", &pool_allocator_nointr, IPL_NONE); | | 347 | "wapbldealloc", &pool_allocator_nointr, IPL_NONE); |
338 | | | 348 | |
339 | wapbl_sysctl_init(); | | 349 | wapbl_sysctl_init(); |
340 | } | | 350 | } |
341 | | | 351 | |
342 | static int | | 352 | static int |
343 | wapbl_fini(void) | | 353 | wapbl_fini(void) |
344 | { | | 354 | { |
345 | | | 355 | |
346 | if (wapbl_sysctl != NULL) | | 356 | if (wapbl_sysctl != NULL) |
347 | sysctl_teardown(&wapbl_sysctl); | | 357 | sysctl_teardown(&wapbl_sysctl); |
348 | | | 358 | |
349 | pool_destroy(&wapbl_dealloc_pool); | | 359 | pool_destroy(&wapbl_dealloc_pool); |
350 | pool_destroy(&wapbl_entry_pool); | | 360 | pool_destroy(&wapbl_entry_pool); |
351 | | | 361 | |
352 | return 0; | | 362 | return 0; |
353 | } | | 363 | } |
354 | | | 364 | |
| | | 365 | static void |
| | | 366 | wapbl_evcnt_init(struct wapbl *wl) |
| | | 367 | { |
| | | 368 | snprintf(wl->wl_ev_group, sizeof(wl->wl_ev_group), |
| | | 369 | "wapbl fsid 0x%x/0x%x", |
| | | 370 | wl->wl_mount->mnt_stat.f_fsidx.__fsid_val[0], |
| | | 371 | wl->wl_mount->mnt_stat.f_fsidx.__fsid_val[1] |
| | | 372 | ); |
| | | 373 | |
| | | 374 | evcnt_attach_dynamic(&wl->wl_ev_commit, EVCNT_TYPE_MISC, |
| | | 375 | NULL, wl->wl_ev_group, "commit"); |
| | | 376 | evcnt_attach_dynamic(&wl->wl_ev_journalwrite, EVCNT_TYPE_MISC, |
| | | 377 | NULL, wl->wl_ev_group, "journal sync block write"); |
| | | 378 | evcnt_attach_dynamic(&wl->wl_ev_metawrite, EVCNT_TYPE_MISC, |
| | | 379 | NULL, wl->wl_ev_group, "metadata finished block write"); |
| | | 380 | evcnt_attach_dynamic(&wl->wl_ev_cacheflush, EVCNT_TYPE_MISC, |
| | | 381 | NULL, wl->wl_ev_group, "cache flush"); |
| | | 382 | } |
| | | 383 | |
| | | 384 | static void |
| | | 385 | wapbl_evcnt_free(struct wapbl *wl) |
| | | 386 | { |
| | | 387 | evcnt_detach(&wl->wl_ev_commit); |
| | | 388 | evcnt_detach(&wl->wl_ev_journalwrite); |
| | | 389 | evcnt_detach(&wl->wl_ev_metawrite); |
| | | 390 | evcnt_detach(&wl->wl_ev_cacheflush); |
| | | 391 | } |
| | | 392 | |
355 | static int | | 393 | static int |
356 | wapbl_start_flush_inodes(struct wapbl *wl, struct wapbl_replay *wr) | | 394 | wapbl_start_flush_inodes(struct wapbl *wl, struct wapbl_replay *wr) |
357 | { | | 395 | { |
358 | int error, i; | | 396 | int error, i; |
359 | | | 397 | |
360 | WAPBL_PRINTF(WAPBL_PRINT_REPLAY, | | 398 | WAPBL_PRINTF(WAPBL_PRINT_REPLAY, |
361 | ("wapbl_start: reusing log with %d inodes\n", wr->wr_inodescnt)); | | 399 | ("wapbl_start: reusing log with %d inodes\n", wr->wr_inodescnt)); |
362 | | | 400 | |
363 | /* | | 401 | /* |
364 | * Its only valid to reuse the replay log if its | | 402 | * Its only valid to reuse the replay log if its |
365 | * the same as the new log we just opened. | | 403 | * the same as the new log we just opened. |
366 | */ | | 404 | */ |
367 | KDASSERT(!wapbl_replay_isopen(wr)); | | 405 | KDASSERT(!wapbl_replay_isopen(wr)); |
368 | KASSERT(wl->wl_devvp->v_type == VBLK); | | 406 | KASSERT(wl->wl_devvp->v_type == VBLK); |
369 | KASSERT(wr->wr_devvp->v_type == VBLK); | | 407 | KASSERT(wr->wr_devvp->v_type == VBLK); |
370 | KASSERT(wl->wl_devvp->v_rdev == wr->wr_devvp->v_rdev); | | 408 | KASSERT(wl->wl_devvp->v_rdev == wr->wr_devvp->v_rdev); |
371 | KASSERT(wl->wl_logpbn == wr->wr_logpbn); | | 409 | KASSERT(wl->wl_logpbn == wr->wr_logpbn); |
372 | KASSERT(wl->wl_circ_size == wr->wr_circ_size); | | 410 | KASSERT(wl->wl_circ_size == wr->wr_circ_size); |
373 | KASSERT(wl->wl_circ_off == wr->wr_circ_off); | | 411 | KASSERT(wl->wl_circ_off == wr->wr_circ_off); |
374 | KASSERT(wl->wl_log_dev_bshift == wr->wr_log_dev_bshift); | | 412 | KASSERT(wl->wl_log_dev_bshift == wr->wr_log_dev_bshift); |
375 | KASSERT(wl->wl_fs_dev_bshift == wr->wr_fs_dev_bshift); | | 413 | KASSERT(wl->wl_fs_dev_bshift == wr->wr_fs_dev_bshift); |
376 | | | 414 | |
377 | wl->wl_wc_header->wc_generation = wr->wr_generation + 1; | | 415 | wl->wl_wc_header->wc_generation = wr->wr_generation + 1; |
378 | | | 416 | |
379 | for (i = 0; i < wr->wr_inodescnt; i++) | | 417 | for (i = 0; i < wr->wr_inodescnt; i++) |
380 | wapbl_register_inode(wl, wr->wr_inodes[i].wr_inumber, | | 418 | wapbl_register_inode(wl, wr->wr_inodes[i].wr_inumber, |
381 | wr->wr_inodes[i].wr_imode); | | 419 | wr->wr_inodes[i].wr_imode); |
382 | | | 420 | |
383 | /* Make sure new transaction won't overwrite old inodes list */ | | 421 | /* Make sure new transaction won't overwrite old inodes list */ |
384 | KDASSERT(wapbl_transaction_len(wl) <= | | 422 | KDASSERT(wapbl_transaction_len(wl) <= |
385 | wapbl_space_free(wl->wl_circ_size, wr->wr_inodeshead, | | 423 | wapbl_space_free(wl->wl_circ_size, wr->wr_inodeshead, |
386 | wr->wr_inodestail)); | | 424 | wr->wr_inodestail)); |
387 | | | 425 | |
388 | wl->wl_head = wl->wl_tail = wr->wr_inodeshead; | | 426 | wl->wl_head = wl->wl_tail = wr->wr_inodeshead; |
389 | wl->wl_reclaimable_bytes = wl->wl_reserved_bytes = | | 427 | wl->wl_reclaimable_bytes = wl->wl_reserved_bytes = |
390 | wapbl_transaction_len(wl); | | 428 | wapbl_transaction_len(wl); |
391 | | | 429 | |
392 | error = wapbl_write_inodes(wl, &wl->wl_head); | | 430 | error = wapbl_write_inodes(wl, &wl->wl_head); |
393 | if (error) | | 431 | if (error) |
394 | return error; | | 432 | return error; |
395 | | | 433 | |
396 | KASSERT(wl->wl_head != wl->wl_tail); | | 434 | KASSERT(wl->wl_head != wl->wl_tail); |
397 | KASSERT(wl->wl_head != 0); | | 435 | KASSERT(wl->wl_head != 0); |
398 | | | 436 | |
399 | return 0; | | 437 | return 0; |
400 | } | | 438 | } |
401 | | | 439 | |
402 | int | | 440 | int |
403 | wapbl_start(struct wapbl ** wlp, struct mount *mp, struct vnode *vp, | | 441 | wapbl_start(struct wapbl ** wlp, struct mount *mp, struct vnode *vp, |
404 | daddr_t off, size_t count, size_t blksize, struct wapbl_replay *wr, | | 442 | daddr_t off, size_t count, size_t blksize, struct wapbl_replay *wr, |
405 | wapbl_flush_fn_t flushfn, wapbl_flush_fn_t flushabortfn) | | 443 | wapbl_flush_fn_t flushfn, wapbl_flush_fn_t flushabortfn) |
406 | { | | 444 | { |
407 | struct wapbl *wl; | | 445 | struct wapbl *wl; |
408 | struct vnode *devvp; | | 446 | struct vnode *devvp; |
409 | daddr_t logpbn; | | 447 | daddr_t logpbn; |
410 | int error; | | 448 | int error; |
411 | int log_dev_bshift = ilog2(blksize); | | 449 | int log_dev_bshift = ilog2(blksize); |
412 | int fs_dev_bshift = log_dev_bshift; | | 450 | int fs_dev_bshift = log_dev_bshift; |
413 | int run; | | 451 | int run; |
414 | | | 452 | |
415 | WAPBL_PRINTF(WAPBL_PRINT_OPEN, ("wapbl_start: vp=%p off=%" PRId64 | | 453 | WAPBL_PRINTF(WAPBL_PRINT_OPEN, ("wapbl_start: vp=%p off=%" PRId64 |
416 | " count=%zu blksize=%zu\n", vp, off, count, blksize)); | | 454 | " count=%zu blksize=%zu\n", vp, off, count, blksize)); |
417 | | | 455 | |
418 | if (log_dev_bshift > fs_dev_bshift) { | | 456 | if (log_dev_bshift > fs_dev_bshift) { |
419 | WAPBL_PRINTF(WAPBL_PRINT_OPEN, | | 457 | WAPBL_PRINTF(WAPBL_PRINT_OPEN, |
420 | ("wapbl: log device's block size cannot be larger " | | 458 | ("wapbl: log device's block size cannot be larger " |
421 | "than filesystem's\n")); | | 459 | "than filesystem's\n")); |
422 | /* | | 460 | /* |
423 | * Not currently implemented, although it could be if | | 461 | * Not currently implemented, although it could be if |
424 | * needed someday. | | 462 | * needed someday. |
425 | */ | | 463 | */ |
426 | return ENOSYS; | | 464 | return ENOSYS; |
427 | } | | 465 | } |
428 | | | 466 | |
429 | if (off < 0) | | 467 | if (off < 0) |
430 | return EINVAL; | | 468 | return EINVAL; |
431 | | | 469 | |
432 | if (blksize < DEV_BSIZE) | | 470 | if (blksize < DEV_BSIZE) |
433 | return EINVAL; | | 471 | return EINVAL; |
434 | if (blksize % DEV_BSIZE) | | 472 | if (blksize % DEV_BSIZE) |
435 | return EINVAL; | | 473 | return EINVAL; |
436 | | | 474 | |
437 | /* XXXTODO: verify that the full load is writable */ | | 475 | /* XXXTODO: verify that the full load is writable */ |
438 | | | 476 | |
439 | /* | | 477 | /* |
440 | * XXX check for minimum log size | | 478 | * XXX check for minimum log size |
441 | * minimum is governed by minimum amount of space | | 479 | * minimum is governed by minimum amount of space |
442 | * to complete a transaction. (probably truncate) | | 480 | * to complete a transaction. (probably truncate) |
443 | */ | | 481 | */ |
444 | /* XXX for now pick something minimal */ | | 482 | /* XXX for now pick something minimal */ |
445 | if ((count * blksize) < MAXPHYS) { | | 483 | if ((count * blksize) < MAXPHYS) { |
446 | return ENOSPC; | | 484 | return ENOSPC; |
447 | } | | 485 | } |
448 | | | 486 | |
449 | if ((error = VOP_BMAP(vp, off, &devvp, &logpbn, &run)) != 0) { | | 487 | if ((error = VOP_BMAP(vp, off, &devvp, &logpbn, &run)) != 0) { |
450 | return error; | | 488 | return error; |
451 | } | | 489 | } |
452 | | | 490 | |
453 | wl = wapbl_calloc(1, sizeof(*wl)); | | 491 | wl = wapbl_calloc(1, sizeof(*wl)); |
454 | rw_init(&wl->wl_rwlock); | | 492 | rw_init(&wl->wl_rwlock); |
455 | mutex_init(&wl->wl_mtx, MUTEX_DEFAULT, IPL_NONE); | | 493 | mutex_init(&wl->wl_mtx, MUTEX_DEFAULT, IPL_NONE); |
456 | cv_init(&wl->wl_reclaimable_cv, "wapblrec"); | | 494 | cv_init(&wl->wl_reclaimable_cv, "wapblrec"); |
457 | LIST_INIT(&wl->wl_bufs); | | 495 | LIST_INIT(&wl->wl_bufs); |
458 | SIMPLEQ_INIT(&wl->wl_entries); | | 496 | SIMPLEQ_INIT(&wl->wl_entries); |
459 | | | 497 | |
460 | wl->wl_logvp = vp; | | 498 | wl->wl_logvp = vp; |
461 | wl->wl_devvp = devvp; | | 499 | wl->wl_devvp = devvp; |
462 | wl->wl_mount = mp; | | 500 | wl->wl_mount = mp; |
463 | wl->wl_logpbn = logpbn; | | 501 | wl->wl_logpbn = logpbn; |
464 | wl->wl_log_dev_bshift = log_dev_bshift; | | 502 | wl->wl_log_dev_bshift = log_dev_bshift; |
465 | wl->wl_fs_dev_bshift = fs_dev_bshift; | | 503 | wl->wl_fs_dev_bshift = fs_dev_bshift; |
466 | | | 504 | |
467 | wl->wl_flush = flushfn; | | 505 | wl->wl_flush = flushfn; |
468 | wl->wl_flush_abort = flushabortfn; | | 506 | wl->wl_flush_abort = flushabortfn; |
469 | | | 507 | |
470 | /* Reserve two log device blocks for the commit headers */ | | 508 | /* Reserve two log device blocks for the commit headers */ |
471 | wl->wl_circ_off = 2<<wl->wl_log_dev_bshift; | | 509 | wl->wl_circ_off = 2<<wl->wl_log_dev_bshift; |
472 | wl->wl_circ_size = ((count * blksize) - wl->wl_circ_off); | | 510 | wl->wl_circ_size = ((count * blksize) - wl->wl_circ_off); |
473 | /* truncate the log usage to a multiple of log_dev_bshift */ | | 511 | /* truncate the log usage to a multiple of log_dev_bshift */ |
474 | wl->wl_circ_size >>= wl->wl_log_dev_bshift; | | 512 | wl->wl_circ_size >>= wl->wl_log_dev_bshift; |
475 | wl->wl_circ_size <<= wl->wl_log_dev_bshift; | | 513 | wl->wl_circ_size <<= wl->wl_log_dev_bshift; |
476 | | | 514 | |
477 | /* | | 515 | /* |
478 | * wl_bufbytes_max limits the size of the in memory transaction space. | | 516 | * wl_bufbytes_max limits the size of the in memory transaction space. |
479 | * - Since buffers are allocated and accounted for in units of | | 517 | * - Since buffers are allocated and accounted for in units of |
480 | * PAGE_SIZE it is required to be a multiple of PAGE_SIZE | | 518 | * PAGE_SIZE it is required to be a multiple of PAGE_SIZE |
481 | * (i.e. 1<<PAGE_SHIFT) | | 519 | * (i.e. 1<<PAGE_SHIFT) |
482 | * - Since the log device has to be written in units of | | 520 | * - Since the log device has to be written in units of |
483 | * 1<<wl_log_dev_bshift it is required to be a mulitple of | | 521 | * 1<<wl_log_dev_bshift it is required to be a mulitple of |
484 | * 1<<wl_log_dev_bshift. | | 522 | * 1<<wl_log_dev_bshift. |
485 | * - Since filesystem will provide data in units of 1<<wl_fs_dev_bshift, | | 523 | * - Since filesystem will provide data in units of 1<<wl_fs_dev_bshift, |
486 | * it is convenient to be a multiple of 1<<wl_fs_dev_bshift. | | 524 | * it is convenient to be a multiple of 1<<wl_fs_dev_bshift. |
487 | * Therefore it must be multiple of the least common multiple of those | | 525 | * Therefore it must be multiple of the least common multiple of those |
488 | * three quantities. Fortunately, all of those quantities are | | 526 | * three quantities. Fortunately, all of those quantities are |
489 | * guaranteed to be a power of two, and the least common multiple of | | 527 | * guaranteed to be a power of two, and the least common multiple of |
490 | * a set of numbers which are all powers of two is simply the maximum | | 528 | * a set of numbers which are all powers of two is simply the maximum |
491 | * of those numbers. Finally, the maximum logarithm of a power of two | | 529 | * of those numbers. Finally, the maximum logarithm of a power of two |
492 | * is the same as the log of the maximum power of two. So we can do | | 530 | * is the same as the log of the maximum power of two. So we can do |
493 | * the following operations to size wl_bufbytes_max: | | 531 | * the following operations to size wl_bufbytes_max: |
494 | */ | | 532 | */ |
495 | | | 533 | |
496 | /* XXX fix actual number of pages reserved per filesystem. */ | | 534 | /* XXX fix actual number of pages reserved per filesystem. */ |
497 | wl->wl_bufbytes_max = MIN(wl->wl_circ_size, buf_memcalc() / 2); | | 535 | wl->wl_bufbytes_max = MIN(wl->wl_circ_size, buf_memcalc() / 2); |
498 | | | 536 | |
499 | /* Round wl_bufbytes_max to the largest power of two constraint */ | | 537 | /* Round wl_bufbytes_max to the largest power of two constraint */ |
500 | wl->wl_bufbytes_max >>= PAGE_SHIFT; | | 538 | wl->wl_bufbytes_max >>= PAGE_SHIFT; |
501 | wl->wl_bufbytes_max <<= PAGE_SHIFT; | | 539 | wl->wl_bufbytes_max <<= PAGE_SHIFT; |
502 | wl->wl_bufbytes_max >>= wl->wl_log_dev_bshift; | | 540 | wl->wl_bufbytes_max >>= wl->wl_log_dev_bshift; |
503 | wl->wl_bufbytes_max <<= wl->wl_log_dev_bshift; | | 541 | wl->wl_bufbytes_max <<= wl->wl_log_dev_bshift; |
504 | wl->wl_bufbytes_max >>= wl->wl_fs_dev_bshift; | | 542 | wl->wl_bufbytes_max >>= wl->wl_fs_dev_bshift; |
505 | wl->wl_bufbytes_max <<= wl->wl_fs_dev_bshift; | | 543 | wl->wl_bufbytes_max <<= wl->wl_fs_dev_bshift; |
506 | | | 544 | |
507 | /* XXX maybe use filesystem fragment size instead of 1024 */ | | 545 | /* XXX maybe use filesystem fragment size instead of 1024 */ |
508 | /* XXX fix actual number of buffers reserved per filesystem. */ | | 546 | /* XXX fix actual number of buffers reserved per filesystem. */ |
509 | wl->wl_bufcount_max = (nbuf / 2) * 1024; | | 547 | wl->wl_bufcount_max = (nbuf / 2) * 1024; |
510 | | | 548 | |
511 | wl->wl_brperjblock = ((1<<wl->wl_log_dev_bshift) | | 549 | wl->wl_brperjblock = ((1<<wl->wl_log_dev_bshift) |
512 | - offsetof(struct wapbl_wc_blocklist, wc_blocks)) / | | 550 | - offsetof(struct wapbl_wc_blocklist, wc_blocks)) / |
513 | sizeof(((struct wapbl_wc_blocklist *)0)->wc_blocks[0]); | | 551 | sizeof(((struct wapbl_wc_blocklist *)0)->wc_blocks[0]); |
514 | KASSERT(wl->wl_brperjblock > 0); | | 552 | KASSERT(wl->wl_brperjblock > 0); |
515 | | | 553 | |
516 | /* XXX tie this into resource estimation */ | | 554 | /* XXX tie this into resource estimation */ |
517 | wl->wl_dealloclim = wl->wl_bufbytes_max / mp->mnt_stat.f_bsize / 2; | | 555 | wl->wl_dealloclim = wl->wl_bufbytes_max / mp->mnt_stat.f_bsize / 2; |
518 | TAILQ_INIT(&wl->wl_dealloclist); | | 556 | TAILQ_INIT(&wl->wl_dealloclist); |
519 | | | 557 | |
520 | wl->wl_buffer = wapbl_alloc(MAXPHYS); | | 558 | wl->wl_buffer = wapbl_alloc(MAXPHYS); |
521 | wl->wl_buffer_used = 0; | | 559 | wl->wl_buffer_used = 0; |
522 | | | 560 | |
523 | wapbl_inodetrk_init(wl, WAPBL_INODETRK_SIZE); | | 561 | wapbl_inodetrk_init(wl, WAPBL_INODETRK_SIZE); |
524 | | | 562 | |
| | | 563 | wapbl_evcnt_init(wl); |
| | | 564 | |
525 | /* Initialize the commit header */ | | 565 | /* Initialize the commit header */ |
526 | { | | 566 | { |
527 | struct wapbl_wc_header *wc; | | 567 | struct wapbl_wc_header *wc; |
528 | size_t len = 1 << wl->wl_log_dev_bshift; | | 568 | size_t len = 1 << wl->wl_log_dev_bshift; |
529 | wc = wapbl_calloc(1, len); | | 569 | wc = wapbl_calloc(1, len); |
530 | wc->wc_type = WAPBL_WC_HEADER; | | 570 | wc->wc_type = WAPBL_WC_HEADER; |
531 | wc->wc_len = len; | | 571 | wc->wc_len = len; |
532 | wc->wc_circ_off = wl->wl_circ_off; | | 572 | wc->wc_circ_off = wl->wl_circ_off; |
533 | wc->wc_circ_size = wl->wl_circ_size; | | 573 | wc->wc_circ_size = wl->wl_circ_size; |
534 | /* XXX wc->wc_fsid */ | | 574 | /* XXX wc->wc_fsid */ |
535 | wc->wc_log_dev_bshift = wl->wl_log_dev_bshift; | | 575 | wc->wc_log_dev_bshift = wl->wl_log_dev_bshift; |
536 | wc->wc_fs_dev_bshift = wl->wl_fs_dev_bshift; | | 576 | wc->wc_fs_dev_bshift = wl->wl_fs_dev_bshift; |
537 | wl->wl_wc_header = wc; | | 577 | wl->wl_wc_header = wc; |
538 | wl->wl_wc_scratch = wapbl_alloc(len); | | 578 | wl->wl_wc_scratch = wapbl_alloc(len); |
539 | } | | 579 | } |
540 | | | 580 | |
541 | /* | | 581 | /* |
542 | * if there was an existing set of unlinked but | | 582 | * if there was an existing set of unlinked but |
543 | * allocated inodes, preserve it in the new | | 583 | * allocated inodes, preserve it in the new |
544 | * log. | | 584 | * log. |
545 | */ | | 585 | */ |
546 | if (wr && wr->wr_inodescnt) { | | 586 | if (wr && wr->wr_inodescnt) { |
547 | error = wapbl_start_flush_inodes(wl, wr); | | 587 | error = wapbl_start_flush_inodes(wl, wr); |
548 | if (error) | | 588 | if (error) |
549 | goto errout; | | 589 | goto errout; |
550 | } | | 590 | } |
551 | | | 591 | |
552 | error = wapbl_write_commit(wl, wl->wl_head, wl->wl_tail); | | 592 | error = wapbl_write_commit(wl, wl->wl_head, wl->wl_tail); |
553 | if (error) { | | 593 | if (error) { |
554 | goto errout; | | 594 | goto errout; |
555 | } | | 595 | } |
556 | | | 596 | |
557 | *wlp = wl; | | 597 | *wlp = wl; |
558 | #if defined(WAPBL_DEBUG) | | 598 | #if defined(WAPBL_DEBUG) |
559 | wapbl_debug_wl = wl; | | 599 | wapbl_debug_wl = wl; |
560 | #endif | | 600 | #endif |
561 | | | 601 | |
562 | return 0; | | 602 | return 0; |
563 | errout: | | 603 | errout: |
564 | wapbl_discard(wl); | | 604 | wapbl_discard(wl); |
565 | wapbl_free(wl->wl_wc_scratch, wl->wl_wc_header->wc_len); | | 605 | wapbl_free(wl->wl_wc_scratch, wl->wl_wc_header->wc_len); |
566 | wapbl_free(wl->wl_wc_header, wl->wl_wc_header->wc_len); | | 606 | wapbl_free(wl->wl_wc_header, wl->wl_wc_header->wc_len); |
567 | wapbl_free(wl->wl_buffer, MAXPHYS); | | 607 | wapbl_free(wl->wl_buffer, MAXPHYS); |
568 | wapbl_inodetrk_free(wl); | | 608 | wapbl_inodetrk_free(wl); |
569 | wapbl_free(wl, sizeof(*wl)); | | 609 | wapbl_free(wl, sizeof(*wl)); |
570 | | | 610 | |
571 | return error; | | 611 | return error; |
572 | } | | 612 | } |
573 | | | 613 | |
574 | /* | | 614 | /* |
575 | * Like wapbl_flush, only discards the transaction | | 615 | * Like wapbl_flush, only discards the transaction |
576 | * completely | | 616 | * completely |
577 | */ | | 617 | */ |
578 | | | 618 | |
579 | void | | 619 | void |
580 | wapbl_discard(struct wapbl *wl) | | 620 | wapbl_discard(struct wapbl *wl) |
581 | { | | 621 | { |
582 | struct wapbl_entry *we; | | 622 | struct wapbl_entry *we; |
583 | struct wapbl_dealloc *wd; | | 623 | struct wapbl_dealloc *wd; |
584 | struct buf *bp; | | 624 | struct buf *bp; |
585 | int i; | | 625 | int i; |
586 | | | 626 | |
587 | /* | | 627 | /* |
588 | * XXX we may consider using upgrade here | | 628 | * XXX we may consider using upgrade here |
589 | * if we want to call flush from inside a transaction | | 629 | * if we want to call flush from inside a transaction |
590 | */ | | 630 | */ |
591 | rw_enter(&wl->wl_rwlock, RW_WRITER); | | 631 | rw_enter(&wl->wl_rwlock, RW_WRITER); |
592 | wl->wl_flush(wl->wl_mount, TAILQ_FIRST(&wl->wl_dealloclist)); | | 632 | wl->wl_flush(wl->wl_mount, TAILQ_FIRST(&wl->wl_dealloclist)); |
593 | | | 633 | |
594 | #ifdef WAPBL_DEBUG_PRINT | | 634 | #ifdef WAPBL_DEBUG_PRINT |
595 | { | | 635 | { |
596 | pid_t pid = -1; | | 636 | pid_t pid = -1; |
597 | lwpid_t lid = -1; | | 637 | lwpid_t lid = -1; |
598 | if (curproc) | | 638 | if (curproc) |
599 | pid = curproc->p_pid; | | 639 | pid = curproc->p_pid; |
600 | if (curlwp) | | 640 | if (curlwp) |
601 | lid = curlwp->l_lid; | | 641 | lid = curlwp->l_lid; |
602 | #ifdef WAPBL_DEBUG_BUFBYTES | | 642 | #ifdef WAPBL_DEBUG_BUFBYTES |
603 | WAPBL_PRINTF(WAPBL_PRINT_DISCARD, | | 643 | WAPBL_PRINTF(WAPBL_PRINT_DISCARD, |
604 | ("wapbl_discard: thread %d.%d discarding " | | 644 | ("wapbl_discard: thread %d.%d discarding " |
605 | "transaction\n" | | 645 | "transaction\n" |
606 | "\tbufcount=%zu bufbytes=%zu bcount=%zu " | | 646 | "\tbufcount=%zu bufbytes=%zu bcount=%zu " |
607 | "deallocs=%d inodes=%d\n" | | 647 | "deallocs=%d inodes=%d\n" |
608 | "\terrcnt = %u, reclaimable=%zu reserved=%zu " | | 648 | "\terrcnt = %u, reclaimable=%zu reserved=%zu " |
609 | "unsynced=%zu\n", | | 649 | "unsynced=%zu\n", |
610 | pid, lid, wl->wl_bufcount, wl->wl_bufbytes, | | 650 | pid, lid, wl->wl_bufcount, wl->wl_bufbytes, |
611 | wl->wl_bcount, wl->wl_dealloccnt, | | 651 | wl->wl_bcount, wl->wl_dealloccnt, |
612 | wl->wl_inohashcnt, wl->wl_error_count, | | 652 | wl->wl_inohashcnt, wl->wl_error_count, |
613 | wl->wl_reclaimable_bytes, wl->wl_reserved_bytes, | | 653 | wl->wl_reclaimable_bytes, wl->wl_reserved_bytes, |
614 | wl->wl_unsynced_bufbytes)); | | 654 | wl->wl_unsynced_bufbytes)); |
615 | SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { | | 655 | SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { |
616 | WAPBL_PRINTF(WAPBL_PRINT_DISCARD, | | 656 | WAPBL_PRINTF(WAPBL_PRINT_DISCARD, |
617 | ("\tentry: bufcount = %zu, reclaimable = %zu, " | | 657 | ("\tentry: bufcount = %zu, reclaimable = %zu, " |
618 | "error = %d, unsynced = %zu\n", | | 658 | "error = %d, unsynced = %zu\n", |
619 | we->we_bufcount, we->we_reclaimable_bytes, | | 659 | we->we_bufcount, we->we_reclaimable_bytes, |
620 | we->we_error, we->we_unsynced_bufbytes)); | | 660 | we->we_error, we->we_unsynced_bufbytes)); |
621 | } | | 661 | } |
622 | #else /* !WAPBL_DEBUG_BUFBYTES */ | | 662 | #else /* !WAPBL_DEBUG_BUFBYTES */ |
623 | WAPBL_PRINTF(WAPBL_PRINT_DISCARD, | | 663 | WAPBL_PRINTF(WAPBL_PRINT_DISCARD, |
624 | ("wapbl_discard: thread %d.%d discarding transaction\n" | | 664 | ("wapbl_discard: thread %d.%d discarding transaction\n" |
625 | "\tbufcount=%zu bufbytes=%zu bcount=%zu " | | 665 | "\tbufcount=%zu bufbytes=%zu bcount=%zu " |
626 | "deallocs=%d inodes=%d\n" | | 666 | "deallocs=%d inodes=%d\n" |
627 | "\terrcnt = %u, reclaimable=%zu reserved=%zu\n", | | 667 | "\terrcnt = %u, reclaimable=%zu reserved=%zu\n", |
628 | pid, lid, wl->wl_bufcount, wl->wl_bufbytes, | | 668 | pid, lid, wl->wl_bufcount, wl->wl_bufbytes, |
629 | wl->wl_bcount, wl->wl_dealloccnt, | | 669 | wl->wl_bcount, wl->wl_dealloccnt, |
630 | wl->wl_inohashcnt, wl->wl_error_count, | | 670 | wl->wl_inohashcnt, wl->wl_error_count, |
631 | wl->wl_reclaimable_bytes, wl->wl_reserved_bytes)); | | 671 | wl->wl_reclaimable_bytes, wl->wl_reserved_bytes)); |
632 | SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { | | 672 | SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { |
633 | WAPBL_PRINTF(WAPBL_PRINT_DISCARD, | | 673 | WAPBL_PRINTF(WAPBL_PRINT_DISCARD, |
634 | ("\tentry: bufcount = %zu, reclaimable = %zu, " | | 674 | ("\tentry: bufcount = %zu, reclaimable = %zu, " |
635 | "error = %d\n", | | 675 | "error = %d\n", |
636 | we->we_bufcount, we->we_reclaimable_bytes, | | 676 | we->we_bufcount, we->we_reclaimable_bytes, |
637 | we->we_error)); | | 677 | we->we_error)); |
638 | } | | 678 | } |
639 | #endif /* !WAPBL_DEBUG_BUFBYTES */ | | 679 | #endif /* !WAPBL_DEBUG_BUFBYTES */ |
640 | } | | 680 | } |
641 | #endif /* WAPBL_DEBUG_PRINT */ | | 681 | #endif /* WAPBL_DEBUG_PRINT */ |
642 | | | 682 | |
643 | for (i = 0; i <= wl->wl_inohashmask; i++) { | | 683 | for (i = 0; i <= wl->wl_inohashmask; i++) { |
644 | struct wapbl_ino_head *wih; | | 684 | struct wapbl_ino_head *wih; |
645 | struct wapbl_ino *wi; | | 685 | struct wapbl_ino *wi; |
646 | | | 686 | |
647 | wih = &wl->wl_inohash[i]; | | 687 | wih = &wl->wl_inohash[i]; |
648 | while ((wi = LIST_FIRST(wih)) != NULL) { | | 688 | while ((wi = LIST_FIRST(wih)) != NULL) { |
649 | LIST_REMOVE(wi, wi_hash); | | 689 | LIST_REMOVE(wi, wi_hash); |
650 | pool_put(&wapbl_ino_pool, wi); | | 690 | pool_put(&wapbl_ino_pool, wi); |
651 | KASSERT(wl->wl_inohashcnt > 0); | | 691 | KASSERT(wl->wl_inohashcnt > 0); |
652 | wl->wl_inohashcnt--; | | 692 | wl->wl_inohashcnt--; |
653 | } | | 693 | } |
654 | } | | 694 | } |
655 | | | 695 | |
656 | /* | | 696 | /* |
657 | * clean buffer list | | 697 | * clean buffer list |
658 | */ | | 698 | */ |
659 | mutex_enter(&bufcache_lock); | | 699 | mutex_enter(&bufcache_lock); |
660 | mutex_enter(&wl->wl_mtx); | | 700 | mutex_enter(&wl->wl_mtx); |
661 | while ((bp = LIST_FIRST(&wl->wl_bufs)) != NULL) { | | 701 | while ((bp = LIST_FIRST(&wl->wl_bufs)) != NULL) { |
662 | if (bbusy(bp, 0, 0, &wl->wl_mtx) == 0) { | | 702 | if (bbusy(bp, 0, 0, &wl->wl_mtx) == 0) { |
663 | /* | | 703 | /* |
664 | * The buffer will be unlocked and | | 704 | * The buffer will be unlocked and |
665 | * removed from the transaction in brelse | | 705 | * removed from the transaction in brelse |
666 | */ | | 706 | */ |
667 | mutex_exit(&wl->wl_mtx); | | 707 | mutex_exit(&wl->wl_mtx); |
668 | brelsel(bp, 0); | | 708 | brelsel(bp, 0); |
669 | mutex_enter(&wl->wl_mtx); | | 709 | mutex_enter(&wl->wl_mtx); |
670 | } | | 710 | } |
671 | } | | 711 | } |
672 | mutex_exit(&wl->wl_mtx); | | 712 | mutex_exit(&wl->wl_mtx); |
673 | mutex_exit(&bufcache_lock); | | 713 | mutex_exit(&bufcache_lock); |
674 | | | 714 | |
675 | /* | | 715 | /* |
676 | * Remove references to this wl from wl_entries, free any which | | 716 | * Remove references to this wl from wl_entries, free any which |
677 | * no longer have buffers, others will be freed in wapbl_biodone | | 717 | * no longer have buffers, others will be freed in wapbl_biodone |
678 | * when they no longer have any buffers. | | 718 | * when they no longer have any buffers. |
679 | */ | | 719 | */ |
680 | while ((we = SIMPLEQ_FIRST(&wl->wl_entries)) != NULL) { | | 720 | while ((we = SIMPLEQ_FIRST(&wl->wl_entries)) != NULL) { |
681 | SIMPLEQ_REMOVE_HEAD(&wl->wl_entries, we_entries); | | 721 | SIMPLEQ_REMOVE_HEAD(&wl->wl_entries, we_entries); |
682 | /* XXX should we be accumulating wl_error_count | | 722 | /* XXX should we be accumulating wl_error_count |
683 | * and increasing reclaimable bytes ? */ | | 723 | * and increasing reclaimable bytes ? */ |
684 | we->we_wapbl = NULL; | | 724 | we->we_wapbl = NULL; |
685 | if (we->we_bufcount == 0) { | | 725 | if (we->we_bufcount == 0) { |
686 | #ifdef WAPBL_DEBUG_BUFBYTES | | 726 | #ifdef WAPBL_DEBUG_BUFBYTES |
687 | KASSERT(we->we_unsynced_bufbytes == 0); | | 727 | KASSERT(we->we_unsynced_bufbytes == 0); |
688 | #endif | | 728 | #endif |
689 | pool_put(&wapbl_entry_pool, we); | | 729 | pool_put(&wapbl_entry_pool, we); |
690 | } | | 730 | } |
691 | } | | 731 | } |
692 | | | 732 | |
693 | /* Discard list of deallocs */ | | 733 | /* Discard list of deallocs */ |
694 | while ((wd = TAILQ_FIRST(&wl->wl_dealloclist)) != NULL) | | 734 | while ((wd = TAILQ_FIRST(&wl->wl_dealloclist)) != NULL) |
695 | wapbl_deallocation_free(wl, wd, true); | | 735 | wapbl_deallocation_free(wl, wd, true); |
696 | | | 736 | |
697 | /* XXX should we clear wl_reserved_bytes? */ | | 737 | /* XXX should we clear wl_reserved_bytes? */ |
698 | | | 738 | |
699 | KASSERT(wl->wl_bufbytes == 0); | | 739 | KASSERT(wl->wl_bufbytes == 0); |
700 | KASSERT(wl->wl_bcount == 0); | | 740 | KASSERT(wl->wl_bcount == 0); |
701 | KASSERT(wl->wl_bufcount == 0); | | 741 | KASSERT(wl->wl_bufcount == 0); |
702 | KASSERT(LIST_EMPTY(&wl->wl_bufs)); | | 742 | KASSERT(LIST_EMPTY(&wl->wl_bufs)); |
703 | KASSERT(SIMPLEQ_EMPTY(&wl->wl_entries)); | | 743 | KASSERT(SIMPLEQ_EMPTY(&wl->wl_entries)); |
704 | KASSERT(wl->wl_inohashcnt == 0); | | 744 | KASSERT(wl->wl_inohashcnt == 0); |
705 | KASSERT(TAILQ_EMPTY(&wl->wl_dealloclist)); | | 745 | KASSERT(TAILQ_EMPTY(&wl->wl_dealloclist)); |
706 | KASSERT(wl->wl_dealloccnt == 0); | | 746 | KASSERT(wl->wl_dealloccnt == 0); |
707 | | | 747 | |
708 | rw_exit(&wl->wl_rwlock); | | 748 | rw_exit(&wl->wl_rwlock); |
709 | } | | 749 | } |
710 | | | 750 | |
711 | int | | 751 | int |
712 | wapbl_stop(struct wapbl *wl, int force) | | 752 | wapbl_stop(struct wapbl *wl, int force) |
713 | { | | 753 | { |
714 | int error; | | 754 | int error; |
715 | | | 755 | |
716 | WAPBL_PRINTF(WAPBL_PRINT_OPEN, ("wapbl_stop called\n")); | | 756 | WAPBL_PRINTF(WAPBL_PRINT_OPEN, ("wapbl_stop called\n")); |
717 | error = wapbl_flush(wl, 1); | | 757 | error = wapbl_flush(wl, 1); |
718 | if (error) { | | 758 | if (error) { |
719 | if (force) | | 759 | if (force) |
720 | wapbl_discard(wl); | | 760 | wapbl_discard(wl); |
721 | else | | 761 | else |
722 | return error; | | 762 | return error; |
723 | } | | 763 | } |
724 | | | 764 | |
725 | /* Unlinked inodes persist after a flush */ | | 765 | /* Unlinked inodes persist after a flush */ |
726 | if (wl->wl_inohashcnt) { | | 766 | if (wl->wl_inohashcnt) { |
727 | if (force) { | | 767 | if (force) { |
728 | wapbl_discard(wl); | | 768 | wapbl_discard(wl); |
729 | } else { | | 769 | } else { |
730 | return EBUSY; | | 770 | return EBUSY; |
731 | } | | 771 | } |
732 | } | | 772 | } |
733 | | | 773 | |
734 | KASSERT(wl->wl_bufbytes == 0); | | 774 | KASSERT(wl->wl_bufbytes == 0); |
735 | KASSERT(wl->wl_bcount == 0); | | 775 | KASSERT(wl->wl_bcount == 0); |
736 | KASSERT(wl->wl_bufcount == 0); | | 776 | KASSERT(wl->wl_bufcount == 0); |
737 | KASSERT(LIST_EMPTY(&wl->wl_bufs)); | | 777 | KASSERT(LIST_EMPTY(&wl->wl_bufs)); |
738 | KASSERT(wl->wl_dealloccnt == 0); | | 778 | KASSERT(wl->wl_dealloccnt == 0); |
739 | KASSERT(SIMPLEQ_EMPTY(&wl->wl_entries)); | | 779 | KASSERT(SIMPLEQ_EMPTY(&wl->wl_entries)); |
740 | KASSERT(wl->wl_inohashcnt == 0); | | 780 | KASSERT(wl->wl_inohashcnt == 0); |
741 | KASSERT(TAILQ_EMPTY(&wl->wl_dealloclist)); | | 781 | KASSERT(TAILQ_EMPTY(&wl->wl_dealloclist)); |
742 | KASSERT(wl->wl_dealloccnt == 0); | | 782 | KASSERT(wl->wl_dealloccnt == 0); |
743 | | | 783 | |
744 | wapbl_free(wl->wl_wc_scratch, wl->wl_wc_header->wc_len); | | 784 | wapbl_free(wl->wl_wc_scratch, wl->wl_wc_header->wc_len); |
745 | wapbl_free(wl->wl_wc_header, wl->wl_wc_header->wc_len); | | 785 | wapbl_free(wl->wl_wc_header, wl->wl_wc_header->wc_len); |
746 | wapbl_free(wl->wl_buffer, MAXPHYS); | | 786 | wapbl_free(wl->wl_buffer, MAXPHYS); |
747 | wapbl_inodetrk_free(wl); | | 787 | wapbl_inodetrk_free(wl); |
748 | | | 788 | |
| | | 789 | wapbl_evcnt_free(wl); |
| | | 790 | |
749 | cv_destroy(&wl->wl_reclaimable_cv); | | 791 | cv_destroy(&wl->wl_reclaimable_cv); |
750 | mutex_destroy(&wl->wl_mtx); | | 792 | mutex_destroy(&wl->wl_mtx); |
751 | rw_destroy(&wl->wl_rwlock); | | 793 | rw_destroy(&wl->wl_rwlock); |
752 | wapbl_free(wl, sizeof(*wl)); | | 794 | wapbl_free(wl, sizeof(*wl)); |
753 | | | 795 | |
754 | return 0; | | 796 | return 0; |
755 | } | | 797 | } |
756 | | | 798 | |
757 | /****************************************************************/ | | 799 | /****************************************************************/ |
758 | /* | | 800 | /* |
759 | * Unbuffered disk I/O | | 801 | * Unbuffered disk I/O |
760 | */ | | 802 | */ |
761 | | | 803 | |
762 | static int | | 804 | static int |
763 | wapbl_doio(void *data, size_t len, struct vnode *devvp, daddr_t pbn, int flags) | | 805 | wapbl_doio(void *data, size_t len, struct vnode *devvp, daddr_t pbn, int flags) |
764 | { | | 806 | { |
765 | struct pstats *pstats = curlwp->l_proc->p_stats; | | 807 | struct pstats *pstats = curlwp->l_proc->p_stats; |
766 | struct buf *bp; | | 808 | struct buf *bp; |
767 | int error; | | 809 | int error; |
768 | | | 810 | |
769 | KASSERT((flags & ~(B_WRITE | B_READ)) == 0); | | 811 | KASSERT((flags & ~(B_WRITE | B_READ)) == 0); |
770 | KASSERT(devvp->v_type == VBLK); | | 812 | KASSERT(devvp->v_type == VBLK); |
771 | | | 813 | |
772 | if ((flags & (B_WRITE | B_READ)) == B_WRITE) { | | 814 | if ((flags & (B_WRITE | B_READ)) == B_WRITE) { |
773 | mutex_enter(devvp->v_interlock); | | 815 | mutex_enter(devvp->v_interlock); |
774 | devvp->v_numoutput++; | | 816 | devvp->v_numoutput++; |
775 | mutex_exit(devvp->v_interlock); | | 817 | mutex_exit(devvp->v_interlock); |
776 | pstats->p_ru.ru_oublock++; | | 818 | pstats->p_ru.ru_oublock++; |
777 | } else { | | 819 | } else { |
778 | pstats->p_ru.ru_inblock++; | | 820 | pstats->p_ru.ru_inblock++; |
779 | } | | 821 | } |
780 | | | 822 | |
781 | bp = getiobuf(devvp, true); | | 823 | bp = getiobuf(devvp, true); |
782 | bp->b_flags = flags; | | 824 | bp->b_flags = flags; |
783 | bp->b_cflags = BC_BUSY; /* silly & dubious */ | | 825 | bp->b_cflags = BC_BUSY; /* silly & dubious */ |
784 | bp->b_dev = devvp->v_rdev; | | 826 | bp->b_dev = devvp->v_rdev; |
785 | bp->b_data = data; | | 827 | bp->b_data = data; |
786 | bp->b_bufsize = bp->b_resid = bp->b_bcount = len; | | 828 | bp->b_bufsize = bp->b_resid = bp->b_bcount = len; |
787 | bp->b_blkno = pbn; | | 829 | bp->b_blkno = pbn; |
788 | BIO_SETPRIO(bp, BPRIO_TIMECRITICAL); | | 830 | BIO_SETPRIO(bp, BPRIO_TIMECRITICAL); |
789 | | | 831 | |
790 | WAPBL_PRINTF(WAPBL_PRINT_IO, | | 832 | WAPBL_PRINTF(WAPBL_PRINT_IO, |
791 | ("wapbl_doio: %s %d bytes at block %"PRId64" on dev 0x%"PRIx64"\n", | | 833 | ("wapbl_doio: %s %d bytes at block %"PRId64" on dev 0x%"PRIx64"\n", |
792 | BUF_ISWRITE(bp) ? "write" : "read", bp->b_bcount, | | 834 | BUF_ISWRITE(bp) ? "write" : "read", bp->b_bcount, |
793 | bp->b_blkno, bp->b_dev)); | | 835 | bp->b_blkno, bp->b_dev)); |
794 | | | 836 | |
795 | VOP_STRATEGY(devvp, bp); | | 837 | VOP_STRATEGY(devvp, bp); |
796 | | | 838 | |
797 | error = biowait(bp); | | 839 | error = biowait(bp); |
798 | putiobuf(bp); | | 840 | putiobuf(bp); |
799 | | | 841 | |
800 | if (error) { | | 842 | if (error) { |
801 | WAPBL_PRINTF(WAPBL_PRINT_ERROR, | | 843 | WAPBL_PRINTF(WAPBL_PRINT_ERROR, |
802 | ("wapbl_doio: %s %zu bytes at block %" PRId64 | | 844 | ("wapbl_doio: %s %zu bytes at block %" PRId64 |
803 | " on dev 0x%"PRIx64" failed with error %d\n", | | 845 | " on dev 0x%"PRIx64" failed with error %d\n", |
804 | (((flags & (B_WRITE | B_READ)) == B_WRITE) ? | | 846 | (((flags & (B_WRITE | B_READ)) == B_WRITE) ? |
805 | "write" : "read"), | | 847 | "write" : "read"), |
806 | len, pbn, devvp->v_rdev, error)); | | 848 | len, pbn, devvp->v_rdev, error)); |
807 | } | | 849 | } |
808 | | | 850 | |
809 | return error; | | 851 | return error; |
810 | } | | 852 | } |
811 | | | 853 | |
812 | /* | | 854 | /* |
813 | * wapbl_write(data, len, devvp, pbn) | | 855 | * wapbl_write(data, len, devvp, pbn) |
814 | * | | 856 | * |
815 | * Synchronously write len bytes from data to physical block pbn | | 857 | * Synchronously write len bytes from data to physical block pbn |
816 | * on devvp. | | 858 | * on devvp. |
817 | */ | | 859 | */ |
818 | int | | 860 | int |
819 | wapbl_write(void *data, size_t len, struct vnode *devvp, daddr_t pbn) | | 861 | wapbl_write(void *data, size_t len, struct vnode *devvp, daddr_t pbn) |
820 | { | | 862 | { |
821 | | | 863 | |
822 | return wapbl_doio(data, len, devvp, pbn, B_WRITE); | | 864 | return wapbl_doio(data, len, devvp, pbn, B_WRITE); |
823 | } | | 865 | } |
824 | | | 866 | |
825 | /* | | 867 | /* |
826 | * wapbl_read(data, len, devvp, pbn) | | 868 | * wapbl_read(data, len, devvp, pbn) |
827 | * | | 869 | * |
828 | * Synchronously read len bytes into data from physical block pbn | | 870 | * Synchronously read len bytes into data from physical block pbn |
829 | * on devvp. | | 871 | * on devvp. |
830 | */ | | 872 | */ |
831 | int | | 873 | int |
832 | wapbl_read(void *data, size_t len, struct vnode *devvp, daddr_t pbn) | | 874 | wapbl_read(void *data, size_t len, struct vnode *devvp, daddr_t pbn) |
833 | { | | 875 | { |
834 | | | 876 | |
835 | return wapbl_doio(data, len, devvp, pbn, B_READ); | | 877 | return wapbl_doio(data, len, devvp, pbn, B_READ); |
836 | } | | 878 | } |
837 | | | 879 | |
838 | /****************************************************************/ | | 880 | /****************************************************************/ |
839 | /* | | 881 | /* |
840 | * Buffered disk writes -- try to coalesce writes and emit | | 882 | * Buffered disk writes -- try to coalesce writes and emit |
841 | * MAXPHYS-aligned blocks. | | 883 | * MAXPHYS-aligned blocks. |
842 | */ | | 884 | */ |
843 | | | 885 | |
844 | /* | | 886 | /* |
845 | * wapbl_buffered_flush(wl) | | 887 | * wapbl_buffered_flush(wl) |
846 | * | | 888 | * |
847 | * Flush any buffered writes from wapbl_buffered_write. | | 889 | * Flush any buffered writes from wapbl_buffered_write. |
848 | */ | | 890 | */ |
849 | static int | | 891 | static int |
850 | wapbl_buffered_flush(struct wapbl *wl) | | 892 | wapbl_buffered_flush(struct wapbl *wl) |
851 | { | | 893 | { |
852 | int error; | | 894 | int error; |
853 | | | 895 | |
854 | if (wl->wl_buffer_used == 0) | | 896 | if (wl->wl_buffer_used == 0) |
855 | return 0; | | 897 | return 0; |
856 | | | 898 | |
857 | error = wapbl_doio(wl->wl_buffer, wl->wl_buffer_used, | | 899 | error = wapbl_doio(wl->wl_buffer, wl->wl_buffer_used, |
858 | wl->wl_devvp, wl->wl_buffer_dblk, B_WRITE); | | 900 | wl->wl_devvp, wl->wl_buffer_dblk, B_WRITE); |
859 | wl->wl_buffer_used = 0; | | 901 | wl->wl_buffer_used = 0; |
860 | | | 902 | |
| | | 903 | wl->wl_ev_journalwrite.ev_count++; |
| | | 904 | |
861 | return error; | | 905 | return error; |
862 | } | | 906 | } |
863 | | | 907 | |
864 | /* | | 908 | /* |
865 | * wapbl_buffered_write(data, len, wl, pbn) | | 909 | * wapbl_buffered_write(data, len, wl, pbn) |
866 | * | | 910 | * |
867 | * Write len bytes from data to physical block pbn on | | 911 | * Write len bytes from data to physical block pbn on |
868 | * wl->wl_devvp. The write may not complete until | | 912 | * wl->wl_devvp. The write may not complete until |
869 | * wapbl_buffered_flush. | | 913 | * wapbl_buffered_flush. |
870 | */ | | 914 | */ |
871 | static int | | 915 | static int |
872 | wapbl_buffered_write(void *data, size_t len, struct wapbl *wl, daddr_t pbn) | | 916 | wapbl_buffered_write(void *data, size_t len, struct wapbl *wl, daddr_t pbn) |
873 | { | | 917 | { |
874 | int error; | | 918 | int error; |
875 | size_t resid; | | 919 | size_t resid; |
876 | | | 920 | |
877 | /* | | 921 | /* |
878 | * If not adjacent to buffered data flush first. Disk block | | 922 | * If not adjacent to buffered data flush first. Disk block |
879 | * address is always valid for non-empty buffer. | | 923 | * address is always valid for non-empty buffer. |
880 | */ | | 924 | */ |
881 | if (wl->wl_buffer_used > 0 && | | 925 | if (wl->wl_buffer_used > 0 && |
882 | pbn != wl->wl_buffer_dblk + btodb(wl->wl_buffer_used)) { | | 926 | pbn != wl->wl_buffer_dblk + btodb(wl->wl_buffer_used)) { |
883 | error = wapbl_buffered_flush(wl); | | 927 | error = wapbl_buffered_flush(wl); |
884 | if (error) | | 928 | if (error) |
885 | return error; | | 929 | return error; |
886 | } | | 930 | } |
887 | /* | | 931 | /* |
888 | * If this write goes to an empty buffer we have to | | 932 | * If this write goes to an empty buffer we have to |
889 | * save the disk block address first. | | 933 | * save the disk block address first. |
890 | */ | | 934 | */ |
891 | if (wl->wl_buffer_used == 0) | | 935 | if (wl->wl_buffer_used == 0) |
892 | wl->wl_buffer_dblk = pbn; | | 936 | wl->wl_buffer_dblk = pbn; |
893 | /* | | 937 | /* |
894 | * Remaining space so this buffer ends on a MAXPHYS boundary. | | 938 | * Remaining space so this buffer ends on a MAXPHYS boundary. |
895 | * | | 939 | * |
896 | * Cannot become less or equal zero as the buffer would have been | | 940 | * Cannot become less or equal zero as the buffer would have been |
897 | * flushed on the last call then. | | 941 | * flushed on the last call then. |
898 | */ | | 942 | */ |
899 | resid = MAXPHYS - dbtob(wl->wl_buffer_dblk % btodb(MAXPHYS)) - | | 943 | resid = MAXPHYS - dbtob(wl->wl_buffer_dblk % btodb(MAXPHYS)) - |
900 | wl->wl_buffer_used; | | 944 | wl->wl_buffer_used; |
901 | KASSERT(resid > 0); | | 945 | KASSERT(resid > 0); |
902 | KASSERT(dbtob(btodb(resid)) == resid); | | 946 | KASSERT(dbtob(btodb(resid)) == resid); |
903 | if (len >= resid) { | | 947 | if (len >= resid) { |
904 | memcpy(wl->wl_buffer + wl->wl_buffer_used, data, resid); | | 948 | memcpy(wl->wl_buffer + wl->wl_buffer_used, data, resid); |
905 | wl->wl_buffer_used += resid; | | 949 | wl->wl_buffer_used += resid; |
906 | error = wapbl_doio(wl->wl_buffer, wl->wl_buffer_used, | | 950 | error = wapbl_doio(wl->wl_buffer, wl->wl_buffer_used, |
907 | wl->wl_devvp, wl->wl_buffer_dblk, B_WRITE); | | 951 | wl->wl_devvp, wl->wl_buffer_dblk, B_WRITE); |
908 | data = (uint8_t *)data + resid; | | 952 | data = (uint8_t *)data + resid; |
909 | len -= resid; | | 953 | len -= resid; |
910 | wl->wl_buffer_dblk = pbn + btodb(resid); | | 954 | wl->wl_buffer_dblk = pbn + btodb(resid); |
911 | wl->wl_buffer_used = 0; | | 955 | wl->wl_buffer_used = 0; |
912 | if (error) | | 956 | if (error) |
913 | return error; | | 957 | return error; |
914 | } | | 958 | } |
915 | KASSERT(len < MAXPHYS); | | 959 | KASSERT(len < MAXPHYS); |
916 | if (len > 0) { | | 960 | if (len > 0) { |
917 | memcpy(wl->wl_buffer + wl->wl_buffer_used, data, len); | | 961 | memcpy(wl->wl_buffer + wl->wl_buffer_used, data, len); |
918 | wl->wl_buffer_used += len; | | 962 | wl->wl_buffer_used += len; |
919 | } | | 963 | } |
920 | | | 964 | |
921 | return 0; | | 965 | return 0; |
922 | } | | 966 | } |
923 | | | 967 | |
924 | /* | | 968 | /* |
925 | * wapbl_circ_write(wl, data, len, offp) | | 969 | * wapbl_circ_write(wl, data, len, offp) |
926 | * | | 970 | * |
927 | * Write len bytes from data to the circular queue of wl, starting | | 971 | * Write len bytes from data to the circular queue of wl, starting |
928 | * at linear byte offset *offp, and returning the new linear byte | | 972 | * at linear byte offset *offp, and returning the new linear byte |
929 | * offset in *offp. | | 973 | * offset in *offp. |
930 | * | | 974 | * |
931 | * If the starting linear byte offset precedes wl->wl_circ_off, | | 975 | * If the starting linear byte offset precedes wl->wl_circ_off, |
932 | * the write instead begins at wl->wl_circ_off. XXX WTF? This | | 976 | * the write instead begins at wl->wl_circ_off. XXX WTF? This |
933 | * should be a KASSERT, not a conditional. | | 977 | * should be a KASSERT, not a conditional. |
934 | * | | 978 | * |
935 | * The write is buffered in wl and must be flushed with | | 979 | * The write is buffered in wl and must be flushed with |
936 | * wapbl_buffered_flush before it will be submitted to the disk. | | 980 | * wapbl_buffered_flush before it will be submitted to the disk. |
937 | */ | | 981 | */ |
938 | static int | | 982 | static int |
939 | wapbl_circ_write(struct wapbl *wl, void *data, size_t len, off_t *offp) | | 983 | wapbl_circ_write(struct wapbl *wl, void *data, size_t len, off_t *offp) |
940 | { | | 984 | { |
941 | size_t slen; | | 985 | size_t slen; |
942 | off_t off = *offp; | | 986 | off_t off = *offp; |
943 | int error; | | 987 | int error; |
944 | daddr_t pbn; | | 988 | daddr_t pbn; |
945 | | | 989 | |
946 | KDASSERT(((len >> wl->wl_log_dev_bshift) << | | 990 | KDASSERT(((len >> wl->wl_log_dev_bshift) << |
947 | wl->wl_log_dev_bshift) == len); | | 991 | wl->wl_log_dev_bshift) == len); |
948 | | | 992 | |
949 | if (off < wl->wl_circ_off) | | 993 | if (off < wl->wl_circ_off) |
950 | off = wl->wl_circ_off; | | 994 | off = wl->wl_circ_off; |
951 | slen = wl->wl_circ_off + wl->wl_circ_size - off; | | 995 | slen = wl->wl_circ_off + wl->wl_circ_size - off; |
952 | if (slen < len) { | | 996 | if (slen < len) { |
953 | pbn = wl->wl_logpbn + (off >> wl->wl_log_dev_bshift); | | 997 | pbn = wl->wl_logpbn + (off >> wl->wl_log_dev_bshift); |
954 | #ifdef _KERNEL | | 998 | #ifdef _KERNEL |
955 | pbn = btodb(pbn << wl->wl_log_dev_bshift); | | 999 | pbn = btodb(pbn << wl->wl_log_dev_bshift); |
956 | #endif | | 1000 | #endif |
957 | error = wapbl_buffered_write(data, slen, wl, pbn); | | 1001 | error = wapbl_buffered_write(data, slen, wl, pbn); |
958 | if (error) | | 1002 | if (error) |
959 | return error; | | 1003 | return error; |
960 | data = (uint8_t *)data + slen; | | 1004 | data = (uint8_t *)data + slen; |
961 | len -= slen; | | 1005 | len -= slen; |
962 | off = wl->wl_circ_off; | | 1006 | off = wl->wl_circ_off; |
963 | } | | 1007 | } |
964 | pbn = wl->wl_logpbn + (off >> wl->wl_log_dev_bshift); | | 1008 | pbn = wl->wl_logpbn + (off >> wl->wl_log_dev_bshift); |
965 | #ifdef _KERNEL | | 1009 | #ifdef _KERNEL |
966 | pbn = btodb(pbn << wl->wl_log_dev_bshift); | | 1010 | pbn = btodb(pbn << wl->wl_log_dev_bshift); |
967 | #endif | | 1011 | #endif |
968 | error = wapbl_buffered_write(data, len, wl, pbn); | | 1012 | error = wapbl_buffered_write(data, len, wl, pbn); |
969 | if (error) | | 1013 | if (error) |
970 | return error; | | 1014 | return error; |
971 | off += len; | | 1015 | off += len; |
972 | if (off >= wl->wl_circ_off + wl->wl_circ_size) | | 1016 | if (off >= wl->wl_circ_off + wl->wl_circ_size) |
973 | off = wl->wl_circ_off; | | 1017 | off = wl->wl_circ_off; |
974 | *offp = off; | | 1018 | *offp = off; |
975 | return 0; | | 1019 | return 0; |
976 | } | | 1020 | } |
977 | | | 1021 | |
978 | /****************************************************************/ | | 1022 | /****************************************************************/ |
979 | /* | | 1023 | /* |
980 | * WAPBL transactions: entering, adding/removing bufs, and exiting | | 1024 | * WAPBL transactions: entering, adding/removing bufs, and exiting |
981 | */ | | 1025 | */ |
982 | | | 1026 | |
983 | int | | 1027 | int |
984 | wapbl_begin(struct wapbl *wl, const char *file, int line) | | 1028 | wapbl_begin(struct wapbl *wl, const char *file, int line) |
985 | { | | 1029 | { |
986 | int doflush; | | 1030 | int doflush; |
987 | unsigned lockcount; | | 1031 | unsigned lockcount; |
988 | | | 1032 | |
989 | KDASSERT(wl); | | 1033 | KDASSERT(wl); |
990 | | | 1034 | |
991 | /* | | 1035 | /* |
992 | * XXX this needs to be made much more sophisticated. | | 1036 | * XXX this needs to be made much more sophisticated. |
993 | * perhaps each wapbl_begin could reserve a specified | | 1037 | * perhaps each wapbl_begin could reserve a specified |
994 | * number of buffers and bytes. | | 1038 | * number of buffers and bytes. |
995 | */ | | 1039 | */ |
996 | mutex_enter(&wl->wl_mtx); | | 1040 | mutex_enter(&wl->wl_mtx); |
997 | lockcount = wl->wl_lock_count; | | 1041 | lockcount = wl->wl_lock_count; |
998 | doflush = ((wl->wl_bufbytes + (lockcount * MAXPHYS)) > | | 1042 | doflush = ((wl->wl_bufbytes + (lockcount * MAXPHYS)) > |
999 | wl->wl_bufbytes_max / 2) || | | 1043 | wl->wl_bufbytes_max / 2) || |
1000 | ((wl->wl_bufcount + (lockcount * 10)) > | | 1044 | ((wl->wl_bufcount + (lockcount * 10)) > |
1001 | wl->wl_bufcount_max / 2) || | | 1045 | wl->wl_bufcount_max / 2) || |
1002 | (wapbl_transaction_len(wl) > wl->wl_circ_size / 2) || | | 1046 | (wapbl_transaction_len(wl) > wl->wl_circ_size / 2) || |
1003 | (wl->wl_dealloccnt >= (wl->wl_dealloclim / 2)); | | 1047 | (wl->wl_dealloccnt >= (wl->wl_dealloclim / 2)); |
1004 | mutex_exit(&wl->wl_mtx); | | 1048 | mutex_exit(&wl->wl_mtx); |
1005 | | | 1049 | |
1006 | if (doflush) { | | 1050 | if (doflush) { |
1007 | WAPBL_PRINTF(WAPBL_PRINT_FLUSH, | | 1051 | WAPBL_PRINTF(WAPBL_PRINT_FLUSH, |
1008 | ("force flush lockcnt=%d bufbytes=%zu " | | 1052 | ("force flush lockcnt=%d bufbytes=%zu " |
1009 | "(max=%zu) bufcount=%zu (max=%zu) " | | 1053 | "(max=%zu) bufcount=%zu (max=%zu) " |
1010 | "dealloccnt %d (lim=%d)\n", | | 1054 | "dealloccnt %d (lim=%d)\n", |
1011 | lockcount, wl->wl_bufbytes, | | 1055 | lockcount, wl->wl_bufbytes, |
1012 | wl->wl_bufbytes_max, wl->wl_bufcount, | | 1056 | wl->wl_bufbytes_max, wl->wl_bufcount, |
1013 | wl->wl_bufcount_max, | | 1057 | wl->wl_bufcount_max, |
1014 | wl->wl_dealloccnt, wl->wl_dealloclim)); | | 1058 | wl->wl_dealloccnt, wl->wl_dealloclim)); |
1015 | } | | 1059 | } |
1016 | | | 1060 | |
1017 | if (doflush) { | | 1061 | if (doflush) { |
1018 | int error = wapbl_flush(wl, 0); | | 1062 | int error = wapbl_flush(wl, 0); |
1019 | if (error) | | 1063 | if (error) |
1020 | return error; | | 1064 | return error; |
1021 | } | | 1065 | } |
1022 | | | 1066 | |
1023 | rw_enter(&wl->wl_rwlock, RW_READER); | | 1067 | rw_enter(&wl->wl_rwlock, RW_READER); |
1024 | mutex_enter(&wl->wl_mtx); | | 1068 | mutex_enter(&wl->wl_mtx); |
1025 | wl->wl_lock_count++; | | 1069 | wl->wl_lock_count++; |
1026 | mutex_exit(&wl->wl_mtx); | | 1070 | mutex_exit(&wl->wl_mtx); |
1027 | | | 1071 | |
1028 | #if defined(WAPBL_DEBUG_PRINT) | | 1072 | #if defined(WAPBL_DEBUG_PRINT) |
1029 | WAPBL_PRINTF(WAPBL_PRINT_TRANSACTION, | | 1073 | WAPBL_PRINTF(WAPBL_PRINT_TRANSACTION, |
1030 | ("wapbl_begin thread %d.%d with bufcount=%zu " | | 1074 | ("wapbl_begin thread %d.%d with bufcount=%zu " |
1031 | "bufbytes=%zu bcount=%zu at %s:%d\n", | | 1075 | "bufbytes=%zu bcount=%zu at %s:%d\n", |
1032 | curproc->p_pid, curlwp->l_lid, wl->wl_bufcount, | | 1076 | curproc->p_pid, curlwp->l_lid, wl->wl_bufcount, |
1033 | wl->wl_bufbytes, wl->wl_bcount, file, line)); | | 1077 | wl->wl_bufbytes, wl->wl_bcount, file, line)); |
1034 | #endif | | 1078 | #endif |
1035 | | | 1079 | |
1036 | return 0; | | 1080 | return 0; |
1037 | } | | 1081 | } |
1038 | | | 1082 | |
1039 | void | | 1083 | void |
1040 | wapbl_end(struct wapbl *wl) | | 1084 | wapbl_end(struct wapbl *wl) |
1041 | { | | 1085 | { |
1042 | | | 1086 | |
1043 | #if defined(WAPBL_DEBUG_PRINT) | | 1087 | #if defined(WAPBL_DEBUG_PRINT) |
1044 | WAPBL_PRINTF(WAPBL_PRINT_TRANSACTION, | | 1088 | WAPBL_PRINTF(WAPBL_PRINT_TRANSACTION, |
1045 | ("wapbl_end thread %d.%d with bufcount=%zu " | | 1089 | ("wapbl_end thread %d.%d with bufcount=%zu " |
1046 | "bufbytes=%zu bcount=%zu\n", | | 1090 | "bufbytes=%zu bcount=%zu\n", |
1047 | curproc->p_pid, curlwp->l_lid, wl->wl_bufcount, | | 1091 | curproc->p_pid, curlwp->l_lid, wl->wl_bufcount, |
1048 | wl->wl_bufbytes, wl->wl_bcount)); | | 1092 | wl->wl_bufbytes, wl->wl_bcount)); |
1049 | #endif | | 1093 | #endif |
1050 | | | 1094 | |
1051 | /* | | 1095 | /* |
1052 | * XXX this could be handled more gracefully, perhaps place | | 1096 | * XXX this could be handled more gracefully, perhaps place |
1053 | * only a partial transaction in the log and allow the | | 1097 | * only a partial transaction in the log and allow the |
1054 | * remaining to flush without the protection of the journal. | | 1098 | * remaining to flush without the protection of the journal. |
1055 | */ | | 1099 | */ |
1056 | KASSERTMSG((wapbl_transaction_len(wl) <= | | 1100 | KASSERTMSG((wapbl_transaction_len(wl) <= |
1057 | (wl->wl_circ_size - wl->wl_reserved_bytes)), | | 1101 | (wl->wl_circ_size - wl->wl_reserved_bytes)), |
1058 | "wapbl_end: current transaction too big to flush"); | | 1102 | "wapbl_end: current transaction too big to flush"); |
1059 | | | 1103 | |
1060 | mutex_enter(&wl->wl_mtx); | | 1104 | mutex_enter(&wl->wl_mtx); |
1061 | KASSERT(wl->wl_lock_count > 0); | | 1105 | KASSERT(wl->wl_lock_count > 0); |
1062 | wl->wl_lock_count--; | | 1106 | wl->wl_lock_count--; |
1063 | mutex_exit(&wl->wl_mtx); | | 1107 | mutex_exit(&wl->wl_mtx); |
1064 | | | 1108 | |
1065 | rw_exit(&wl->wl_rwlock); | | 1109 | rw_exit(&wl->wl_rwlock); |
1066 | } | | 1110 | } |
1067 | | | 1111 | |
1068 | void | | 1112 | void |
1069 | wapbl_add_buf(struct wapbl *wl, struct buf * bp) | | 1113 | wapbl_add_buf(struct wapbl *wl, struct buf * bp) |
1070 | { | | 1114 | { |
1071 | | | 1115 | |
1072 | KASSERT(bp->b_cflags & BC_BUSY); | | 1116 | KASSERT(bp->b_cflags & BC_BUSY); |
1073 | KASSERT(bp->b_vp); | | 1117 | KASSERT(bp->b_vp); |
1074 | | | 1118 | |
1075 | wapbl_jlock_assert(wl); | | 1119 | wapbl_jlock_assert(wl); |
1076 | | | 1120 | |
1077 | #if 0 | | 1121 | #if 0 |
1078 | /* | | 1122 | /* |
1079 | * XXX this might be an issue for swapfiles. | | 1123 | * XXX this might be an issue for swapfiles. |
1080 | * see uvm_swap.c:1702 | | 1124 | * see uvm_swap.c:1702 |
1081 | * | | 1125 | * |
1082 | * XXX2 why require it then? leap of semantics? | | 1126 | * XXX2 why require it then? leap of semantics? |
1083 | */ | | 1127 | */ |
1084 | KASSERT((bp->b_cflags & BC_NOCACHE) == 0); | | 1128 | KASSERT((bp->b_cflags & BC_NOCACHE) == 0); |
1085 | #endif | | 1129 | #endif |
1086 | | | 1130 | |
1087 | mutex_enter(&wl->wl_mtx); | | 1131 | mutex_enter(&wl->wl_mtx); |
1088 | if (bp->b_flags & B_LOCKED) { | | 1132 | if (bp->b_flags & B_LOCKED) { |
1089 | LIST_REMOVE(bp, b_wapbllist); | | 1133 | LIST_REMOVE(bp, b_wapbllist); |
1090 | WAPBL_PRINTF(WAPBL_PRINT_BUFFER2, | | 1134 | WAPBL_PRINTF(WAPBL_PRINT_BUFFER2, |
1091 | ("wapbl_add_buf thread %d.%d re-adding buf %p " | | 1135 | ("wapbl_add_buf thread %d.%d re-adding buf %p " |
1092 | "with %d bytes %d bcount\n", | | 1136 | "with %d bytes %d bcount\n", |
1093 | curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize, | | 1137 | curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize, |
1094 | bp->b_bcount)); | | 1138 | bp->b_bcount)); |
1095 | } else { | | 1139 | } else { |
1096 | /* unlocked by dirty buffers shouldn't exist */ | | 1140 | /* unlocked by dirty buffers shouldn't exist */ |
1097 | KASSERT(!(bp->b_oflags & BO_DELWRI)); | | 1141 | KASSERT(!(bp->b_oflags & BO_DELWRI)); |
1098 | wl->wl_bufbytes += bp->b_bufsize; | | 1142 | wl->wl_bufbytes += bp->b_bufsize; |
1099 | wl->wl_bcount += bp->b_bcount; | | 1143 | wl->wl_bcount += bp->b_bcount; |
1100 | wl->wl_bufcount++; | | 1144 | wl->wl_bufcount++; |
1101 | WAPBL_PRINTF(WAPBL_PRINT_BUFFER, | | 1145 | WAPBL_PRINTF(WAPBL_PRINT_BUFFER, |
1102 | ("wapbl_add_buf thread %d.%d adding buf %p " | | 1146 | ("wapbl_add_buf thread %d.%d adding buf %p " |
1103 | "with %d bytes %d bcount\n", | | 1147 | "with %d bytes %d bcount\n", |
1104 | curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize, | | 1148 | curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize, |
1105 | bp->b_bcount)); | | 1149 | bp->b_bcount)); |
1106 | } | | 1150 | } |
1107 | LIST_INSERT_HEAD(&wl->wl_bufs, bp, b_wapbllist); | | 1151 | LIST_INSERT_HEAD(&wl->wl_bufs, bp, b_wapbllist); |
1108 | mutex_exit(&wl->wl_mtx); | | 1152 | mutex_exit(&wl->wl_mtx); |
1109 | | | 1153 | |
1110 | bp->b_flags |= B_LOCKED; | | 1154 | bp->b_flags |= B_LOCKED; |
1111 | } | | 1155 | } |
1112 | | | 1156 | |
1113 | static void | | 1157 | static void |
1114 | wapbl_remove_buf_locked(struct wapbl * wl, struct buf *bp) | | 1158 | wapbl_remove_buf_locked(struct wapbl * wl, struct buf *bp) |
1115 | { | | 1159 | { |
1116 | | | 1160 | |
1117 | KASSERT(mutex_owned(&wl->wl_mtx)); | | 1161 | KASSERT(mutex_owned(&wl->wl_mtx)); |
1118 | KASSERT(bp->b_cflags & BC_BUSY); | | 1162 | KASSERT(bp->b_cflags & BC_BUSY); |
1119 | wapbl_jlock_assert(wl); | | 1163 | wapbl_jlock_assert(wl); |
1120 | | | 1164 | |
1121 | #if 0 | | 1165 | #if 0 |
1122 | /* | | 1166 | /* |
1123 | * XXX this might be an issue for swapfiles. | | 1167 | * XXX this might be an issue for swapfiles. |
1124 | * see uvm_swap.c:1725 | | 1168 | * see uvm_swap.c:1725 |
1125 | * | | 1169 | * |
1126 | * XXXdeux: see above | | 1170 | * XXXdeux: see above |
1127 | */ | | 1171 | */ |
1128 | KASSERT((bp->b_flags & BC_NOCACHE) == 0); | | 1172 | KASSERT((bp->b_flags & BC_NOCACHE) == 0); |
1129 | #endif | | 1173 | #endif |
1130 | KASSERT(bp->b_flags & B_LOCKED); | | 1174 | KASSERT(bp->b_flags & B_LOCKED); |
1131 | | | 1175 | |
1132 | WAPBL_PRINTF(WAPBL_PRINT_BUFFER, | | 1176 | WAPBL_PRINTF(WAPBL_PRINT_BUFFER, |
1133 | ("wapbl_remove_buf thread %d.%d removing buf %p with " | | 1177 | ("wapbl_remove_buf thread %d.%d removing buf %p with " |
1134 | "%d bytes %d bcount\n", | | 1178 | "%d bytes %d bcount\n", |
1135 | curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize, bp->b_bcount)); | | 1179 | curproc->p_pid, curlwp->l_lid, bp, bp->b_bufsize, bp->b_bcount)); |
1136 | | | 1180 | |
1137 | KASSERT(wl->wl_bufbytes >= bp->b_bufsize); | | 1181 | KASSERT(wl->wl_bufbytes >= bp->b_bufsize); |
1138 | wl->wl_bufbytes -= bp->b_bufsize; | | 1182 | wl->wl_bufbytes -= bp->b_bufsize; |
1139 | KASSERT(wl->wl_bcount >= bp->b_bcount); | | 1183 | KASSERT(wl->wl_bcount >= bp->b_bcount); |
1140 | wl->wl_bcount -= bp->b_bcount; | | 1184 | wl->wl_bcount -= bp->b_bcount; |
1141 | KASSERT(wl->wl_bufcount > 0); | | 1185 | KASSERT(wl->wl_bufcount > 0); |
1142 | wl->wl_bufcount--; | | 1186 | wl->wl_bufcount--; |
1143 | KASSERT((wl->wl_bufcount == 0) == (wl->wl_bufbytes == 0)); | | 1187 | KASSERT((wl->wl_bufcount == 0) == (wl->wl_bufbytes == 0)); |
1144 | KASSERT((wl->wl_bufcount == 0) == (wl->wl_bcount == 0)); | | 1188 | KASSERT((wl->wl_bufcount == 0) == (wl->wl_bcount == 0)); |
1145 | LIST_REMOVE(bp, b_wapbllist); | | 1189 | LIST_REMOVE(bp, b_wapbllist); |
1146 | | | 1190 | |
1147 | bp->b_flags &= ~B_LOCKED; | | 1191 | bp->b_flags &= ~B_LOCKED; |
1148 | } | | 1192 | } |
1149 | | | 1193 | |
1150 | /* called from brelsel() in vfs_bio among other places */ | | 1194 | /* called from brelsel() in vfs_bio among other places */ |
1151 | void | | 1195 | void |
1152 | wapbl_remove_buf(struct wapbl * wl, struct buf *bp) | | 1196 | wapbl_remove_buf(struct wapbl * wl, struct buf *bp) |
1153 | { | | 1197 | { |
1154 | | | 1198 | |
1155 | mutex_enter(&wl->wl_mtx); | | 1199 | mutex_enter(&wl->wl_mtx); |
1156 | wapbl_remove_buf_locked(wl, bp); | | 1200 | wapbl_remove_buf_locked(wl, bp); |
1157 | mutex_exit(&wl->wl_mtx); | | 1201 | mutex_exit(&wl->wl_mtx); |
1158 | } | | 1202 | } |
1159 | | | 1203 | |
1160 | void | | 1204 | void |
1161 | wapbl_resize_buf(struct wapbl *wl, struct buf *bp, long oldsz, long oldcnt) | | 1205 | wapbl_resize_buf(struct wapbl *wl, struct buf *bp, long oldsz, long oldcnt) |
1162 | { | | 1206 | { |
1163 | | | 1207 | |
1164 | KASSERT(bp->b_cflags & BC_BUSY); | | 1208 | KASSERT(bp->b_cflags & BC_BUSY); |
1165 | | | 1209 | |
1166 | /* | | 1210 | /* |
1167 | * XXX: why does this depend on B_LOCKED? otherwise the buf | | 1211 | * XXX: why does this depend on B_LOCKED? otherwise the buf |
1168 | * is not for a transaction? if so, why is this called in the | | 1212 | * is not for a transaction? if so, why is this called in the |
1169 | * first place? | | 1213 | * first place? |
1170 | */ | | 1214 | */ |
1171 | if (bp->b_flags & B_LOCKED) { | | 1215 | if (bp->b_flags & B_LOCKED) { |
1172 | mutex_enter(&wl->wl_mtx); | | 1216 | mutex_enter(&wl->wl_mtx); |
1173 | wl->wl_bufbytes += bp->b_bufsize - oldsz; | | 1217 | wl->wl_bufbytes += bp->b_bufsize - oldsz; |
1174 | wl->wl_bcount += bp->b_bcount - oldcnt; | | 1218 | wl->wl_bcount += bp->b_bcount - oldcnt; |
1175 | mutex_exit(&wl->wl_mtx); | | 1219 | mutex_exit(&wl->wl_mtx); |
1176 | } | | 1220 | } |
1177 | } | | 1221 | } |
1178 | | | 1222 | |
1179 | #endif /* _KERNEL */ | | 1223 | #endif /* _KERNEL */ |
1180 | | | 1224 | |
1181 | /****************************************************************/ | | 1225 | /****************************************************************/ |
1182 | /* Some utility inlines */ | | 1226 | /* Some utility inlines */ |
1183 | | | 1227 | |
1184 | /* | | 1228 | /* |
1185 | * wapbl_space_used(avail, head, tail) | | 1229 | * wapbl_space_used(avail, head, tail) |
1186 | * | | 1230 | * |
1187 | * Number of bytes used in a circular queue of avail total bytes, | | 1231 | * Number of bytes used in a circular queue of avail total bytes, |
1188 | * from tail to head. | | 1232 | * from tail to head. |
1189 | */ | | 1233 | */ |
1190 | static inline size_t | | 1234 | static inline size_t |
1191 | wapbl_space_used(size_t avail, off_t head, off_t tail) | | 1235 | wapbl_space_used(size_t avail, off_t head, off_t tail) |
1192 | { | | 1236 | { |
1193 | | | 1237 | |
1194 | if (tail == 0) { | | 1238 | if (tail == 0) { |
1195 | KASSERT(head == 0); | | 1239 | KASSERT(head == 0); |
1196 | return 0; | | 1240 | return 0; |
1197 | } | | 1241 | } |
1198 | return ((head + (avail - 1) - tail) % avail) + 1; | | 1242 | return ((head + (avail - 1) - tail) % avail) + 1; |
1199 | } | | 1243 | } |
1200 | | | 1244 | |
1201 | #ifdef _KERNEL | | 1245 | #ifdef _KERNEL |
1202 | /* | | 1246 | /* |
1203 | * wapbl_advance(size, off, oldoff, delta) | | 1247 | * wapbl_advance(size, off, oldoff, delta) |
1204 | * | | 1248 | * |
1205 | * Given a byte offset oldoff into a circular queue of size bytes | | 1249 | * Given a byte offset oldoff into a circular queue of size bytes |
1206 | * starting at off, return a new byte offset oldoff + delta into | | 1250 | * starting at off, return a new byte offset oldoff + delta into |
1207 | * the circular queue. | | 1251 | * the circular queue. |
1208 | */ | | 1252 | */ |
1209 | static inline off_t | | 1253 | static inline off_t |
1210 | wapbl_advance(size_t size, size_t off, off_t oldoff, size_t delta) | | 1254 | wapbl_advance(size_t size, size_t off, off_t oldoff, size_t delta) |
1211 | { | | 1255 | { |
1212 | off_t newoff; | | 1256 | off_t newoff; |
1213 | | | 1257 | |
1214 | /* Define acceptable ranges for inputs. */ | | 1258 | /* Define acceptable ranges for inputs. */ |
1215 | KASSERT(delta <= (size_t)size); | | 1259 | KASSERT(delta <= (size_t)size); |
1216 | KASSERT((oldoff == 0) || ((size_t)oldoff >= off)); | | 1260 | KASSERT((oldoff == 0) || ((size_t)oldoff >= off)); |
1217 | KASSERT(oldoff < (off_t)(size + off)); | | 1261 | KASSERT(oldoff < (off_t)(size + off)); |
1218 | | | 1262 | |
1219 | if ((oldoff == 0) && (delta != 0)) | | 1263 | if ((oldoff == 0) && (delta != 0)) |
1220 | newoff = off + delta; | | 1264 | newoff = off + delta; |
1221 | else if ((oldoff + delta) < (size + off)) | | 1265 | else if ((oldoff + delta) < (size + off)) |
1222 | newoff = oldoff + delta; | | 1266 | newoff = oldoff + delta; |
1223 | else | | 1267 | else |
1224 | newoff = (oldoff + delta) - size; | | 1268 | newoff = (oldoff + delta) - size; |
1225 | | | 1269 | |
1226 | /* Note some interesting axioms */ | | 1270 | /* Note some interesting axioms */ |
1227 | KASSERT((delta != 0) || (newoff == oldoff)); | | 1271 | KASSERT((delta != 0) || (newoff == oldoff)); |
1228 | KASSERT((delta == 0) || (newoff != 0)); | | 1272 | KASSERT((delta == 0) || (newoff != 0)); |
1229 | KASSERT((delta != (size)) || (newoff == oldoff)); | | 1273 | KASSERT((delta != (size)) || (newoff == oldoff)); |
1230 | | | 1274 | |
1231 | /* Define acceptable ranges for output. */ | | 1275 | /* Define acceptable ranges for output. */ |
1232 | KASSERT((newoff == 0) || ((size_t)newoff >= off)); | | 1276 | KASSERT((newoff == 0) || ((size_t)newoff >= off)); |
1233 | KASSERT((size_t)newoff < (size + off)); | | 1277 | KASSERT((size_t)newoff < (size + off)); |
1234 | return newoff; | | 1278 | return newoff; |
1235 | } | | 1279 | } |
1236 | | | 1280 | |
1237 | /* | | 1281 | /* |
1238 | * wapbl_space_free(avail, head, tail) | | 1282 | * wapbl_space_free(avail, head, tail) |
1239 | * | | 1283 | * |
1240 | * Number of bytes free in a circular queue of avail total bytes, | | 1284 | * Number of bytes free in a circular queue of avail total bytes, |
1241 | * in which everything from tail to head is used. | | 1285 | * in which everything from tail to head is used. |
1242 | */ | | 1286 | */ |
1243 | static inline size_t | | 1287 | static inline size_t |
1244 | wapbl_space_free(size_t avail, off_t head, off_t tail) | | 1288 | wapbl_space_free(size_t avail, off_t head, off_t tail) |
1245 | { | | 1289 | { |
1246 | | | 1290 | |
1247 | return avail - wapbl_space_used(avail, head, tail); | | 1291 | return avail - wapbl_space_used(avail, head, tail); |
1248 | } | | 1292 | } |
1249 | | | 1293 | |
1250 | /* | | 1294 | /* |
1251 | * wapbl_advance_head(size, off, delta, headp, tailp) | | 1295 | * wapbl_advance_head(size, off, delta, headp, tailp) |
1252 | * | | 1296 | * |
1253 | * In a circular queue of size bytes starting at off, given the | | 1297 | * In a circular queue of size bytes starting at off, given the |
1254 | * old head and tail offsets *headp and *tailp, store the new head | | 1298 | * old head and tail offsets *headp and *tailp, store the new head |
1255 | * and tail offsets in *headp and *tailp resulting from adding | | 1299 | * and tail offsets in *headp and *tailp resulting from adding |
1256 | * delta bytes of data to the head. | | 1300 | * delta bytes of data to the head. |
1257 | */ | | 1301 | */ |
1258 | static inline void | | 1302 | static inline void |
1259 | wapbl_advance_head(size_t size, size_t off, size_t delta, off_t *headp, | | 1303 | wapbl_advance_head(size_t size, size_t off, size_t delta, off_t *headp, |
1260 | off_t *tailp) | | 1304 | off_t *tailp) |
1261 | { | | 1305 | { |
1262 | off_t head = *headp; | | 1306 | off_t head = *headp; |
1263 | off_t tail = *tailp; | | 1307 | off_t tail = *tailp; |
1264 | | | 1308 | |
1265 | KASSERT(delta <= wapbl_space_free(size, head, tail)); | | 1309 | KASSERT(delta <= wapbl_space_free(size, head, tail)); |
1266 | head = wapbl_advance(size, off, head, delta); | | 1310 | head = wapbl_advance(size, off, head, delta); |
1267 | if ((tail == 0) && (head != 0)) | | 1311 | if ((tail == 0) && (head != 0)) |
1268 | tail = off; | | 1312 | tail = off; |
1269 | *headp = head; | | 1313 | *headp = head; |
1270 | *tailp = tail; | | 1314 | *tailp = tail; |
1271 | } | | 1315 | } |
1272 | | | 1316 | |
1273 | /* | | 1317 | /* |
1274 | * wapbl_advance_tail(size, off, delta, headp, tailp) | | 1318 | * wapbl_advance_tail(size, off, delta, headp, tailp) |
1275 | * | | 1319 | * |
1276 | * In a circular queue of size bytes starting at off, given the | | 1320 | * In a circular queue of size bytes starting at off, given the |
1277 | * old head and tail offsets *headp and *tailp, store the new head | | 1321 | * old head and tail offsets *headp and *tailp, store the new head |
1278 | * and tail offsets in *headp and *tailp resulting from removing | | 1322 | * and tail offsets in *headp and *tailp resulting from removing |
1279 | * delta bytes of data from the tail. | | 1323 | * delta bytes of data from the tail. |
1280 | */ | | 1324 | */ |
1281 | static inline void | | 1325 | static inline void |
1282 | wapbl_advance_tail(size_t size, size_t off, size_t delta, off_t *headp, | | 1326 | wapbl_advance_tail(size_t size, size_t off, size_t delta, off_t *headp, |
1283 | off_t *tailp) | | 1327 | off_t *tailp) |
1284 | { | | 1328 | { |
1285 | off_t head = *headp; | | 1329 | off_t head = *headp; |
1286 | off_t tail = *tailp; | | 1330 | off_t tail = *tailp; |
1287 | | | 1331 | |
1288 | KASSERT(delta <= wapbl_space_used(size, head, tail)); | | 1332 | KASSERT(delta <= wapbl_space_used(size, head, tail)); |
1289 | tail = wapbl_advance(size, off, tail, delta); | | 1333 | tail = wapbl_advance(size, off, tail, delta); |
1290 | if (head == tail) { | | 1334 | if (head == tail) { |
1291 | head = tail = 0; | | 1335 | head = tail = 0; |
1292 | } | | 1336 | } |
1293 | *headp = head; | | 1337 | *headp = head; |
1294 | *tailp = tail; | | 1338 | *tailp = tail; |
1295 | } | | 1339 | } |
1296 | | | 1340 | |
1297 | | | 1341 | |
1298 | /****************************************************************/ | | 1342 | /****************************************************************/ |
1299 | | | 1343 | |
1300 | /* | | 1344 | /* |
1301 | * wapbl_truncate(wl, minfree) | | 1345 | * wapbl_truncate(wl, minfree) |
1302 | * | | 1346 | * |
1303 | * Wait until at least minfree bytes are available in the log. | | 1347 | * Wait until at least minfree bytes are available in the log. |
1304 | * | | 1348 | * |
1305 | * If it was necessary to wait for writes to complete, | | 1349 | * If it was necessary to wait for writes to complete, |
1306 | * advance the circular queue tail to reflect the new write | | 1350 | * advance the circular queue tail to reflect the new write |
1307 | * completions and issue a write commit to the log. | | 1351 | * completions and issue a write commit to the log. |
1308 | * | | 1352 | * |
1309 | * => Caller must hold wl->wl_rwlock writer lock. | | 1353 | * => Caller must hold wl->wl_rwlock writer lock. |
1310 | */ | | 1354 | */ |
1311 | static int | | 1355 | static int |
1312 | wapbl_truncate(struct wapbl *wl, size_t minfree) | | 1356 | wapbl_truncate(struct wapbl *wl, size_t minfree) |
1313 | { | | 1357 | { |
1314 | size_t delta; | | 1358 | size_t delta; |
1315 | size_t avail; | | 1359 | size_t avail; |
1316 | off_t head; | | 1360 | off_t head; |
1317 | off_t tail; | | 1361 | off_t tail; |
1318 | int error = 0; | | 1362 | int error = 0; |
1319 | | | 1363 | |
1320 | KASSERT(minfree <= (wl->wl_circ_size - wl->wl_reserved_bytes)); | | 1364 | KASSERT(minfree <= (wl->wl_circ_size - wl->wl_reserved_bytes)); |
1321 | KASSERT(rw_write_held(&wl->wl_rwlock)); | | 1365 | KASSERT(rw_write_held(&wl->wl_rwlock)); |
1322 | | | 1366 | |
1323 | mutex_enter(&wl->wl_mtx); | | 1367 | mutex_enter(&wl->wl_mtx); |
1324 | | | 1368 | |
1325 | /* | | 1369 | /* |
1326 | * First check to see if we have to do a commit | | 1370 | * First check to see if we have to do a commit |
1327 | * at all. | | 1371 | * at all. |
1328 | */ | | 1372 | */ |
1329 | avail = wapbl_space_free(wl->wl_circ_size, wl->wl_head, wl->wl_tail); | | 1373 | avail = wapbl_space_free(wl->wl_circ_size, wl->wl_head, wl->wl_tail); |
1330 | if (minfree < avail) { | | 1374 | if (minfree < avail) { |
1331 | mutex_exit(&wl->wl_mtx); | | 1375 | mutex_exit(&wl->wl_mtx); |
1332 | return 0; | | 1376 | return 0; |
1333 | } | | 1377 | } |
1334 | minfree -= avail; | | 1378 | minfree -= avail; |
1335 | while ((wl->wl_error_count == 0) && | | 1379 | while ((wl->wl_error_count == 0) && |
1336 | (wl->wl_reclaimable_bytes < minfree)) { | | 1380 | (wl->wl_reclaimable_bytes < minfree)) { |
1337 | WAPBL_PRINTF(WAPBL_PRINT_TRUNCATE, | | 1381 | WAPBL_PRINTF(WAPBL_PRINT_TRUNCATE, |
1338 | ("wapbl_truncate: sleeping on %p wl=%p bytes=%zd " | | 1382 | ("wapbl_truncate: sleeping on %p wl=%p bytes=%zd " |
1339 | "minfree=%zd\n", | | 1383 | "minfree=%zd\n", |
1340 | &wl->wl_reclaimable_bytes, wl, wl->wl_reclaimable_bytes, | | 1384 | &wl->wl_reclaimable_bytes, wl, wl->wl_reclaimable_bytes, |
1341 | minfree)); | | 1385 | minfree)); |
1342 | | | 1386 | |
1343 | cv_wait(&wl->wl_reclaimable_cv, &wl->wl_mtx); | | 1387 | cv_wait(&wl->wl_reclaimable_cv, &wl->wl_mtx); |
1344 | } | | 1388 | } |
1345 | if (wl->wl_reclaimable_bytes < minfree) { | | 1389 | if (wl->wl_reclaimable_bytes < minfree) { |
1346 | KASSERT(wl->wl_error_count); | | 1390 | KASSERT(wl->wl_error_count); |
1347 | /* XXX maybe get actual error from buffer instead someday? */ | | 1391 | /* XXX maybe get actual error from buffer instead someday? */ |
1348 | error = EIO; | | 1392 | error = EIO; |
1349 | } | | 1393 | } |
1350 | head = wl->wl_head; | | 1394 | head = wl->wl_head; |
1351 | tail = wl->wl_tail; | | 1395 | tail = wl->wl_tail; |
1352 | delta = wl->wl_reclaimable_bytes; | | 1396 | delta = wl->wl_reclaimable_bytes; |
1353 | | | 1397 | |
1354 | /* If all of of the entries are flushed, then be sure to keep | | 1398 | /* If all of of the entries are flushed, then be sure to keep |
1355 | * the reserved bytes reserved. Watch out for discarded transactions, | | 1399 | * the reserved bytes reserved. Watch out for discarded transactions, |
1356 | * which could leave more bytes reserved than are reclaimable. | | 1400 | * which could leave more bytes reserved than are reclaimable. |
1357 | */ | | 1401 | */ |
1358 | if (SIMPLEQ_EMPTY(&wl->wl_entries) && | | 1402 | if (SIMPLEQ_EMPTY(&wl->wl_entries) && |
1359 | (delta >= wl->wl_reserved_bytes)) { | | 1403 | (delta >= wl->wl_reserved_bytes)) { |
1360 | delta -= wl->wl_reserved_bytes; | | 1404 | delta -= wl->wl_reserved_bytes; |
1361 | } | | 1405 | } |
1362 | wapbl_advance_tail(wl->wl_circ_size, wl->wl_circ_off, delta, &head, | | 1406 | wapbl_advance_tail(wl->wl_circ_size, wl->wl_circ_off, delta, &head, |
1363 | &tail); | | 1407 | &tail); |
1364 | KDASSERT(wl->wl_reserved_bytes <= | | 1408 | KDASSERT(wl->wl_reserved_bytes <= |
1365 | wapbl_space_used(wl->wl_circ_size, head, tail)); | | 1409 | wapbl_space_used(wl->wl_circ_size, head, tail)); |
1366 | mutex_exit(&wl->wl_mtx); | | 1410 | mutex_exit(&wl->wl_mtx); |
1367 | | | 1411 | |
1368 | if (error) | | 1412 | if (error) |
1369 | return error; | | 1413 | return error; |
1370 | | | 1414 | |
1371 | /* | | 1415 | /* |
1372 | * This is where head, tail and delta are unprotected | | 1416 | * This is where head, tail and delta are unprotected |
1373 | * from races against itself or flush. This is ok since | | 1417 | * from races against itself or flush. This is ok since |
1374 | * we only call this routine from inside flush itself. | | 1418 | * we only call this routine from inside flush itself. |
1375 | * | | 1419 | * |
1376 | * XXX: how can it race against itself when accessed only | | 1420 | * XXX: how can it race against itself when accessed only |
1377 | * from behind the write-locked rwlock? | | 1421 | * from behind the write-locked rwlock? |
1378 | */ | | 1422 | */ |
1379 | error = wapbl_write_commit(wl, head, tail); | | 1423 | error = wapbl_write_commit(wl, head, tail); |
1380 | if (error) | | 1424 | if (error) |
1381 | return error; | | 1425 | return error; |
1382 | | | 1426 | |
1383 | wl->wl_head = head; | | 1427 | wl->wl_head = head; |
1384 | wl->wl_tail = tail; | | 1428 | wl->wl_tail = tail; |
1385 | | | 1429 | |
1386 | mutex_enter(&wl->wl_mtx); | | 1430 | mutex_enter(&wl->wl_mtx); |
1387 | KASSERT(wl->wl_reclaimable_bytes >= delta); | | 1431 | KASSERT(wl->wl_reclaimable_bytes >= delta); |
1388 | wl->wl_reclaimable_bytes -= delta; | | 1432 | wl->wl_reclaimable_bytes -= delta; |
1389 | mutex_exit(&wl->wl_mtx); | | 1433 | mutex_exit(&wl->wl_mtx); |
1390 | WAPBL_PRINTF(WAPBL_PRINT_TRUNCATE, | | 1434 | WAPBL_PRINTF(WAPBL_PRINT_TRUNCATE, |
1391 | ("wapbl_truncate thread %d.%d truncating %zu bytes\n", | | 1435 | ("wapbl_truncate thread %d.%d truncating %zu bytes\n", |
1392 | curproc->p_pid, curlwp->l_lid, delta)); | | 1436 | curproc->p_pid, curlwp->l_lid, delta)); |
1393 | | | 1437 | |
1394 | return 0; | | 1438 | return 0; |
1395 | } | | 1439 | } |
1396 | | | 1440 | |
1397 | /****************************************************************/ | | 1441 | /****************************************************************/ |
1398 | | | 1442 | |
1399 | void | | 1443 | void |
1400 | wapbl_biodone(struct buf *bp) | | 1444 | wapbl_biodone(struct buf *bp) |
1401 | { | | 1445 | { |
1402 | struct wapbl_entry *we = bp->b_private; | | 1446 | struct wapbl_entry *we = bp->b_private; |
1403 | struct wapbl *wl = we->we_wapbl; | | 1447 | struct wapbl *wl = we->we_wapbl; |
1404 | #ifdef WAPBL_DEBUG_BUFBYTES | | 1448 | #ifdef WAPBL_DEBUG_BUFBYTES |
1405 | const int bufsize = bp->b_bufsize; | | 1449 | const int bufsize = bp->b_bufsize; |
1406 | #endif | | 1450 | #endif |
1407 | | | 1451 | |
1408 | /* | | 1452 | /* |
1409 | * Handle possible flushing of buffers after log has been | | 1453 | * Handle possible flushing of buffers after log has been |
1410 | * decomissioned. | | 1454 | * decomissioned. |
1411 | */ | | 1455 | */ |
1412 | if (!wl) { | | 1456 | if (!wl) { |
1413 | KASSERT(we->we_bufcount > 0); | | 1457 | KASSERT(we->we_bufcount > 0); |
1414 | we->we_bufcount--; | | 1458 | we->we_bufcount--; |
1415 | #ifdef WAPBL_DEBUG_BUFBYTES | | 1459 | #ifdef WAPBL_DEBUG_BUFBYTES |
1416 | KASSERT(we->we_unsynced_bufbytes >= bufsize); | | 1460 | KASSERT(we->we_unsynced_bufbytes >= bufsize); |
1417 | we->we_unsynced_bufbytes -= bufsize; | | 1461 | we->we_unsynced_bufbytes -= bufsize; |
1418 | #endif | | 1462 | #endif |
1419 | | | 1463 | |
1420 | if (we->we_bufcount == 0) { | | 1464 | if (we->we_bufcount == 0) { |
1421 | #ifdef WAPBL_DEBUG_BUFBYTES | | 1465 | #ifdef WAPBL_DEBUG_BUFBYTES |
1422 | KASSERT(we->we_unsynced_bufbytes == 0); | | 1466 | KASSERT(we->we_unsynced_bufbytes == 0); |
1423 | #endif | | 1467 | #endif |
1424 | pool_put(&wapbl_entry_pool, we); | | 1468 | pool_put(&wapbl_entry_pool, we); |
1425 | } | | 1469 | } |
1426 | | | 1470 | |
1427 | brelse(bp, 0); | | 1471 | brelse(bp, 0); |
1428 | return; | | 1472 | return; |
1429 | } | | 1473 | } |
1430 | | | 1474 | |
1431 | #ifdef ohbother | | 1475 | #ifdef ohbother |
1432 | KDASSERT(bp->b_oflags & BO_DONE); | | 1476 | KDASSERT(bp->b_oflags & BO_DONE); |
1433 | KDASSERT(!(bp->b_oflags & BO_DELWRI)); | | 1477 | KDASSERT(!(bp->b_oflags & BO_DELWRI)); |
1434 | KDASSERT(bp->b_flags & B_ASYNC); | | 1478 | KDASSERT(bp->b_flags & B_ASYNC); |
1435 | KDASSERT(bp->b_cflags & BC_BUSY); | | 1479 | KDASSERT(bp->b_cflags & BC_BUSY); |
1436 | KDASSERT(!(bp->b_flags & B_LOCKED)); | | 1480 | KDASSERT(!(bp->b_flags & B_LOCKED)); |
1437 | KDASSERT(!(bp->b_flags & B_READ)); | | 1481 | KDASSERT(!(bp->b_flags & B_READ)); |
1438 | KDASSERT(!(bp->b_cflags & BC_INVAL)); | | 1482 | KDASSERT(!(bp->b_cflags & BC_INVAL)); |
1439 | KDASSERT(!(bp->b_cflags & BC_NOCACHE)); | | 1483 | KDASSERT(!(bp->b_cflags & BC_NOCACHE)); |
1440 | #endif | | 1484 | #endif |
1441 | | | 1485 | |
1442 | if (bp->b_error) { | | 1486 | if (bp->b_error) { |
1443 | /* | | 1487 | /* |
1444 | * If an error occurs, it would be nice to leave the buffer | | 1488 | * If an error occurs, it would be nice to leave the buffer |
1445 | * as a delayed write on the LRU queue so that we can retry | | 1489 | * as a delayed write on the LRU queue so that we can retry |
1446 | * it later. But buffercache(9) can't handle dirty buffer | | 1490 | * it later. But buffercache(9) can't handle dirty buffer |
1447 | * reuse, so just mark the log permanently errored out. | | 1491 | * reuse, so just mark the log permanently errored out. |
1448 | */ | | 1492 | */ |
1449 | mutex_enter(&wl->wl_mtx); | | 1493 | mutex_enter(&wl->wl_mtx); |
1450 | if (wl->wl_error_count == 0) { | | 1494 | if (wl->wl_error_count == 0) { |
1451 | wl->wl_error_count++; | | 1495 | wl->wl_error_count++; |
1452 | cv_broadcast(&wl->wl_reclaimable_cv); | | 1496 | cv_broadcast(&wl->wl_reclaimable_cv); |
1453 | } | | 1497 | } |
1454 | mutex_exit(&wl->wl_mtx); | | 1498 | mutex_exit(&wl->wl_mtx); |
1455 | } | | 1499 | } |
1456 | | | 1500 | |
1457 | /* | | 1501 | /* |
1458 | * Release the buffer here. wapbl_flush() may wait for the | | 1502 | * Release the buffer here. wapbl_flush() may wait for the |
1459 | * log to become empty and we better unbusy the buffer before | | 1503 | * log to become empty and we better unbusy the buffer before |
1460 | * wapbl_flush() returns. | | 1504 | * wapbl_flush() returns. |
1461 | */ | | 1505 | */ |
1462 | brelse(bp, 0); | | 1506 | brelse(bp, 0); |
1463 | | | 1507 | |
1464 | mutex_enter(&wl->wl_mtx); | | 1508 | mutex_enter(&wl->wl_mtx); |
1465 | | | 1509 | |
1466 | KASSERT(we->we_bufcount > 0); | | 1510 | KASSERT(we->we_bufcount > 0); |
1467 | we->we_bufcount--; | | 1511 | we->we_bufcount--; |
1468 | #ifdef WAPBL_DEBUG_BUFBYTES | | 1512 | #ifdef WAPBL_DEBUG_BUFBYTES |
1469 | KASSERT(we->we_unsynced_bufbytes >= bufsize); | | 1513 | KASSERT(we->we_unsynced_bufbytes >= bufsize); |
1470 | we->we_unsynced_bufbytes -= bufsize; | | 1514 | we->we_unsynced_bufbytes -= bufsize; |
1471 | KASSERT(wl->wl_unsynced_bufbytes >= bufsize); | | 1515 | KASSERT(wl->wl_unsynced_bufbytes >= bufsize); |
1472 | wl->wl_unsynced_bufbytes -= bufsize; | | 1516 | wl->wl_unsynced_bufbytes -= bufsize; |
1473 | #endif | | 1517 | #endif |
| | | 1518 | wl->wl_ev_metawrite.ev_count++; |
1474 | | | 1519 | |
1475 | /* | | 1520 | /* |
1476 | * If the current transaction can be reclaimed, start | | 1521 | * If the current transaction can be reclaimed, start |
1477 | * at the beginning and reclaim any consecutive reclaimable | | 1522 | * at the beginning and reclaim any consecutive reclaimable |
1478 | * transactions. If we successfully reclaim anything, | | 1523 | * transactions. If we successfully reclaim anything, |
1479 | * then wakeup anyone waiting for the reclaim. | | 1524 | * then wakeup anyone waiting for the reclaim. |
1480 | */ | | 1525 | */ |
1481 | if (we->we_bufcount == 0) { | | 1526 | if (we->we_bufcount == 0) { |
1482 | size_t delta = 0; | | 1527 | size_t delta = 0; |
1483 | int errcnt = 0; | | 1528 | int errcnt = 0; |
1484 | #ifdef WAPBL_DEBUG_BUFBYTES | | 1529 | #ifdef WAPBL_DEBUG_BUFBYTES |
1485 | KDASSERT(we->we_unsynced_bufbytes == 0); | | 1530 | KDASSERT(we->we_unsynced_bufbytes == 0); |
1486 | #endif | | 1531 | #endif |
1487 | /* | | 1532 | /* |
1488 | * clear any posted error, since the buffer it came from | | 1533 | * clear any posted error, since the buffer it came from |
1489 | * has successfully flushed by now | | 1534 | * has successfully flushed by now |
1490 | */ | | 1535 | */ |
1491 | while ((we = SIMPLEQ_FIRST(&wl->wl_entries)) && | | 1536 | while ((we = SIMPLEQ_FIRST(&wl->wl_entries)) && |
1492 | (we->we_bufcount == 0)) { | | 1537 | (we->we_bufcount == 0)) { |
1493 | delta += we->we_reclaimable_bytes; | | 1538 | delta += we->we_reclaimable_bytes; |
1494 | if (we->we_error) | | 1539 | if (we->we_error) |
1495 | errcnt++; | | 1540 | errcnt++; |
1496 | SIMPLEQ_REMOVE_HEAD(&wl->wl_entries, we_entries); | | 1541 | SIMPLEQ_REMOVE_HEAD(&wl->wl_entries, we_entries); |
1497 | pool_put(&wapbl_entry_pool, we); | | 1542 | pool_put(&wapbl_entry_pool, we); |
1498 | } | | 1543 | } |
1499 | | | 1544 | |
1500 | if (delta) { | | 1545 | if (delta) { |
1501 | wl->wl_reclaimable_bytes += delta; | | 1546 | wl->wl_reclaimable_bytes += delta; |
1502 | KASSERT(wl->wl_error_count >= errcnt); | | 1547 | KASSERT(wl->wl_error_count >= errcnt); |
1503 | wl->wl_error_count -= errcnt; | | 1548 | wl->wl_error_count -= errcnt; |
1504 | cv_broadcast(&wl->wl_reclaimable_cv); | | 1549 | cv_broadcast(&wl->wl_reclaimable_cv); |
1505 | } | | 1550 | } |
1506 | } | | 1551 | } |
1507 | | | 1552 | |
1508 | mutex_exit(&wl->wl_mtx); | | 1553 | mutex_exit(&wl->wl_mtx); |
1509 | } | | 1554 | } |
1510 | | | 1555 | |
1511 | /* | | 1556 | /* |
1512 | * wapbl_flush(wl, wait) | | 1557 | * wapbl_flush(wl, wait) |
1513 | * | | 1558 | * |
1514 | * Flush pending block writes, deallocations, and inodes from | | 1559 | * Flush pending block writes, deallocations, and inodes from |
1515 | * the current transaction in memory to the log on disk: | | 1560 | * the current transaction in memory to the log on disk: |
1516 | * | | 1561 | * |
1517 | * 1. Call the file system's wl_flush callback to flush any | | 1562 | * 1. Call the file system's wl_flush callback to flush any |
1518 | * per-file-system pending updates. | | 1563 | * per-file-system pending updates. |
1519 | * 2. Wait for enough space in the log for the current transaction. | | 1564 | * 2. Wait for enough space in the log for the current transaction. |
1520 | * 3. Synchronously write the new log records, advancing the | | 1565 | * 3. Synchronously write the new log records, advancing the |
1521 | * circular queue head. | | 1566 | * circular queue head. |
1522 | * 4. Issue the pending block writes asynchronously, now that they | | 1567 | * 4. Issue the pending block writes asynchronously, now that they |
1523 | * are recorded in the log and can be replayed after crash. | | 1568 | * are recorded in the log and can be replayed after crash. |
1524 | * 5. If wait is true, wait for all writes to complete and for the | | 1569 | * 5. If wait is true, wait for all writes to complete and for the |
1525 | * log to become empty. | | 1570 | * log to become empty. |
1526 | * | | 1571 | * |
1527 | * On failure, call the file system's wl_flush_abort callback. | | 1572 | * On failure, call the file system's wl_flush_abort callback. |
1528 | */ | | 1573 | */ |
1529 | int | | 1574 | int |
1530 | wapbl_flush(struct wapbl *wl, int waitfor) | | 1575 | wapbl_flush(struct wapbl *wl, int waitfor) |
1531 | { | | 1576 | { |
1532 | struct buf *bp; | | 1577 | struct buf *bp; |
1533 | struct wapbl_entry *we; | | 1578 | struct wapbl_entry *we; |
1534 | off_t off; | | 1579 | off_t off; |
1535 | off_t head; | | 1580 | off_t head; |
1536 | off_t tail; | | 1581 | off_t tail; |
1537 | size_t delta = 0; | | 1582 | size_t delta = 0; |
1538 | size_t flushsize; | | 1583 | size_t flushsize; |
1539 | size_t reserved; | | 1584 | size_t reserved; |
1540 | int error = 0; | | 1585 | int error = 0; |
1541 | | | 1586 | |
1542 | /* | | 1587 | /* |
1543 | * Do a quick check to see if a full flush can be skipped | | 1588 | * Do a quick check to see if a full flush can be skipped |
1544 | * This assumes that the flush callback does not need to be called | | 1589 | * This assumes that the flush callback does not need to be called |
1545 | * unless there are other outstanding bufs. | | 1590 | * unless there are other outstanding bufs. |
1546 | */ | | 1591 | */ |
1547 | if (!waitfor) { | | 1592 | if (!waitfor) { |
1548 | size_t nbufs; | | 1593 | size_t nbufs; |
1549 | mutex_enter(&wl->wl_mtx); /* XXX need mutex here to | | 1594 | mutex_enter(&wl->wl_mtx); /* XXX need mutex here to |
1550 | protect the KASSERTS */ | | 1595 | protect the KASSERTS */ |
1551 | nbufs = wl->wl_bufcount; | | 1596 | nbufs = wl->wl_bufcount; |
1552 | KASSERT((wl->wl_bufcount == 0) == (wl->wl_bufbytes == 0)); | | 1597 | KASSERT((wl->wl_bufcount == 0) == (wl->wl_bufbytes == 0)); |
1553 | KASSERT((wl->wl_bufcount == 0) == (wl->wl_bcount == 0)); | | 1598 | KASSERT((wl->wl_bufcount == 0) == (wl->wl_bcount == 0)); |
1554 | mutex_exit(&wl->wl_mtx); | | 1599 | mutex_exit(&wl->wl_mtx); |
1555 | if (nbufs == 0) | | 1600 | if (nbufs == 0) |
1556 | return 0; | | 1601 | return 0; |
1557 | } | | 1602 | } |
1558 | | | 1603 | |
1559 | /* | | 1604 | /* |
1560 | * XXX we may consider using LK_UPGRADE here | | 1605 | * XXX we may consider using LK_UPGRADE here |
1561 | * if we want to call flush from inside a transaction | | 1606 | * if we want to call flush from inside a transaction |
1562 | */ | | 1607 | */ |
1563 | rw_enter(&wl->wl_rwlock, RW_WRITER); | | 1608 | rw_enter(&wl->wl_rwlock, RW_WRITER); |
1564 | wl->wl_flush(wl->wl_mount, TAILQ_FIRST(&wl->wl_dealloclist)); | | 1609 | wl->wl_flush(wl->wl_mount, TAILQ_FIRST(&wl->wl_dealloclist)); |
1565 | | | 1610 | |
1566 | /* | | 1611 | /* |
1567 | * Now that we are exclusively locked and the file system has | | 1612 | * Now that we are exclusively locked and the file system has |
1568 | * issued any deferred block writes for this transaction, check | | 1613 | * issued any deferred block writes for this transaction, check |
1569 | * whether there are any blocks to write to the log. If not, | | 1614 | * whether there are any blocks to write to the log. If not, |
1570 | * skip waiting for space or writing any log entries. | | 1615 | * skip waiting for space or writing any log entries. |
1571 | * | | 1616 | * |
1572 | * XXX Shouldn't this also check wl_dealloccnt and | | 1617 | * XXX Shouldn't this also check wl_dealloccnt and |
1573 | * wl_inohashcnt? Perhaps wl_dealloccnt doesn't matter if the | | 1618 | * wl_inohashcnt? Perhaps wl_dealloccnt doesn't matter if the |
1574 | * file system didn't produce any blocks as a consequence of | | 1619 | * file system didn't produce any blocks as a consequence of |
1575 | * it, but the same does not seem to be so of wl_inohashcnt. | | 1620 | * it, but the same does not seem to be so of wl_inohashcnt. |
1576 | */ | | 1621 | */ |
1577 | if (wl->wl_bufcount == 0) { | | 1622 | if (wl->wl_bufcount == 0) { |
1578 | goto wait_out; | | 1623 | goto wait_out; |
1579 | } | | 1624 | } |
1580 | | | 1625 | |
1581 | #if 0 | | 1626 | #if 0 |
1582 | WAPBL_PRINTF(WAPBL_PRINT_FLUSH, | | 1627 | WAPBL_PRINTF(WAPBL_PRINT_FLUSH, |
1583 | ("wapbl_flush thread %d.%d flushing entries with " | | 1628 | ("wapbl_flush thread %d.%d flushing entries with " |
1584 | "bufcount=%zu bufbytes=%zu\n", | | 1629 | "bufcount=%zu bufbytes=%zu\n", |
1585 | curproc->p_pid, curlwp->l_lid, wl->wl_bufcount, | | 1630 | curproc->p_pid, curlwp->l_lid, wl->wl_bufcount, |
1586 | wl->wl_bufbytes)); | | 1631 | wl->wl_bufbytes)); |
1587 | #endif | | 1632 | #endif |
1588 | | | 1633 | |
1589 | /* Calculate amount of space needed to flush */ | | 1634 | /* Calculate amount of space needed to flush */ |
1590 | flushsize = wapbl_transaction_len(wl); | | 1635 | flushsize = wapbl_transaction_len(wl); |
1591 | if (wapbl_verbose_commit) { | | 1636 | if (wapbl_verbose_commit) { |
1592 | struct timespec ts; | | 1637 | struct timespec ts; |
1593 | getnanotime(&ts); | | 1638 | getnanotime(&ts); |
1594 | printf("%s: %lld.%09ld this transaction = %zu bytes\n", | | 1639 | printf("%s: %lld.%09ld this transaction = %zu bytes\n", |
1595 | __func__, (long long)ts.tv_sec, | | 1640 | __func__, (long long)ts.tv_sec, |
1596 | (long)ts.tv_nsec, flushsize); | | 1641 | (long)ts.tv_nsec, flushsize); |
1597 | } | | 1642 | } |
1598 | | | 1643 | |
1599 | if (flushsize > (wl->wl_circ_size - wl->wl_reserved_bytes)) { | | 1644 | if (flushsize > (wl->wl_circ_size - wl->wl_reserved_bytes)) { |
1600 | /* | | 1645 | /* |
1601 | * XXX this could be handled more gracefully, perhaps place | | 1646 | * XXX this could be handled more gracefully, perhaps place |
1602 | * only a partial transaction in the log and allow the | | 1647 | * only a partial transaction in the log and allow the |
1603 | * remaining to flush without the protection of the journal. | | 1648 | * remaining to flush without the protection of the journal. |
1604 | */ | | 1649 | */ |
1605 | panic("wapbl_flush: current transaction too big to flush"); | | 1650 | panic("wapbl_flush: current transaction too big to flush"); |
1606 | } | | 1651 | } |
1607 | | | 1652 | |
1608 | error = wapbl_truncate(wl, flushsize); | | 1653 | error = wapbl_truncate(wl, flushsize); |
1609 | if (error) | | 1654 | if (error) |
1610 | goto out; | | 1655 | goto out; |
1611 | | | 1656 | |
1612 | off = wl->wl_head; | | 1657 | off = wl->wl_head; |
1613 | KASSERT((off == 0) || (off >= wl->wl_circ_off)); | | 1658 | KASSERT((off == 0) || (off >= wl->wl_circ_off)); |
1614 | KASSERT((off == 0) || (off < wl->wl_circ_off + wl->wl_circ_size)); | | 1659 | KASSERT((off == 0) || (off < wl->wl_circ_off + wl->wl_circ_size)); |
1615 | error = wapbl_write_blocks(wl, &off); | | 1660 | error = wapbl_write_blocks(wl, &off); |
1616 | if (error) | | 1661 | if (error) |
1617 | goto out; | | 1662 | goto out; |
1618 | error = wapbl_write_revocations(wl, &off); | | 1663 | error = wapbl_write_revocations(wl, &off); |
1619 | if (error) | | 1664 | if (error) |
1620 | goto out; | | 1665 | goto out; |
1621 | error = wapbl_write_inodes(wl, &off); | | 1666 | error = wapbl_write_inodes(wl, &off); |
1622 | if (error) | | 1667 | if (error) |
1623 | goto out; | | 1668 | goto out; |
1624 | | | 1669 | |
1625 | reserved = 0; | | 1670 | reserved = 0; |
1626 | if (wl->wl_inohashcnt) | | 1671 | if (wl->wl_inohashcnt) |
1627 | reserved = wapbl_transaction_inodes_len(wl); | | 1672 | reserved = wapbl_transaction_inodes_len(wl); |
1628 | | | 1673 | |
1629 | head = wl->wl_head; | | 1674 | head = wl->wl_head; |
1630 | tail = wl->wl_tail; | | 1675 | tail = wl->wl_tail; |
1631 | | | 1676 | |
1632 | wapbl_advance_head(wl->wl_circ_size, wl->wl_circ_off, flushsize, | | 1677 | wapbl_advance_head(wl->wl_circ_size, wl->wl_circ_off, flushsize, |
1633 | &head, &tail); | | 1678 | &head, &tail); |
1634 | | | 1679 | |
1635 | KASSERTMSG(head == off, | | 1680 | KASSERTMSG(head == off, |
1636 | "lost head! head=%"PRIdMAX" tail=%" PRIdMAX | | 1681 | "lost head! head=%"PRIdMAX" tail=%" PRIdMAX |
1637 | " off=%"PRIdMAX" flush=%zu", | | 1682 | " off=%"PRIdMAX" flush=%zu", |
1638 | (intmax_t)head, (intmax_t)tail, (intmax_t)off, | | 1683 | (intmax_t)head, (intmax_t)tail, (intmax_t)off, |
1639 | flushsize); | | 1684 | flushsize); |
1640 | | | 1685 | |
1641 | /* Opportunistically move the tail forward if we can */ | | 1686 | /* Opportunistically move the tail forward if we can */ |
1642 | mutex_enter(&wl->wl_mtx); | | 1687 | mutex_enter(&wl->wl_mtx); |
1643 | delta = wl->wl_reclaimable_bytes; | | 1688 | delta = wl->wl_reclaimable_bytes; |
1644 | mutex_exit(&wl->wl_mtx); | | 1689 | mutex_exit(&wl->wl_mtx); |
1645 | wapbl_advance_tail(wl->wl_circ_size, wl->wl_circ_off, delta, | | 1690 | wapbl_advance_tail(wl->wl_circ_size, wl->wl_circ_off, delta, |
1646 | &head, &tail); | | 1691 | &head, &tail); |
1647 | | | 1692 | |
1648 | error = wapbl_write_commit(wl, head, tail); | | 1693 | error = wapbl_write_commit(wl, head, tail); |
1649 | if (error) | | 1694 | if (error) |
1650 | goto out; | | 1695 | goto out; |
1651 | | | 1696 | |
1652 | we = pool_get(&wapbl_entry_pool, PR_WAITOK); | | 1697 | we = pool_get(&wapbl_entry_pool, PR_WAITOK); |
1653 | | | 1698 | |
1654 | #ifdef WAPBL_DEBUG_BUFBYTES | | 1699 | #ifdef WAPBL_DEBUG_BUFBYTES |
1655 | WAPBL_PRINTF(WAPBL_PRINT_FLUSH, | | 1700 | WAPBL_PRINTF(WAPBL_PRINT_FLUSH, |
1656 | ("wapbl_flush: thread %d.%d head+=%zu tail+=%zu used=%zu" | | 1701 | ("wapbl_flush: thread %d.%d head+=%zu tail+=%zu used=%zu" |
1657 | " unsynced=%zu" | | 1702 | " unsynced=%zu" |
1658 | "\n\tbufcount=%zu bufbytes=%zu bcount=%zu deallocs=%d " | | 1703 | "\n\tbufcount=%zu bufbytes=%zu bcount=%zu deallocs=%d " |
1659 | "inodes=%d\n", | | 1704 | "inodes=%d\n", |
1660 | curproc->p_pid, curlwp->l_lid, flushsize, delta, | | 1705 | curproc->p_pid, curlwp->l_lid, flushsize, delta, |
1661 | wapbl_space_used(wl->wl_circ_size, head, tail), | | 1706 | wapbl_space_used(wl->wl_circ_size, head, tail), |
1662 | wl->wl_unsynced_bufbytes, wl->wl_bufcount, | | 1707 | wl->wl_unsynced_bufbytes, wl->wl_bufcount, |
1663 | wl->wl_bufbytes, wl->wl_bcount, wl->wl_dealloccnt, | | 1708 | wl->wl_bufbytes, wl->wl_bcount, wl->wl_dealloccnt, |
1664 | wl->wl_inohashcnt)); | | 1709 | wl->wl_inohashcnt)); |
1665 | #else | | 1710 | #else |
1666 | WAPBL_PRINTF(WAPBL_PRINT_FLUSH, | | 1711 | WAPBL_PRINTF(WAPBL_PRINT_FLUSH, |
1667 | ("wapbl_flush: thread %d.%d head+=%zu tail+=%zu used=%zu" | | 1712 | ("wapbl_flush: thread %d.%d head+=%zu tail+=%zu used=%zu" |
1668 | "\n\tbufcount=%zu bufbytes=%zu bcount=%zu deallocs=%d " | | 1713 | "\n\tbufcount=%zu bufbytes=%zu bcount=%zu deallocs=%d " |
1669 | "inodes=%d\n", | | 1714 | "inodes=%d\n", |
1670 | curproc->p_pid, curlwp->l_lid, flushsize, delta, | | 1715 | curproc->p_pid, curlwp->l_lid, flushsize, delta, |
1671 | wapbl_space_used(wl->wl_circ_size, head, tail), | | 1716 | wapbl_space_used(wl->wl_circ_size, head, tail), |
1672 | wl->wl_bufcount, wl->wl_bufbytes, wl->wl_bcount, | | 1717 | wl->wl_bufcount, wl->wl_bufbytes, wl->wl_bcount, |
1673 | wl->wl_dealloccnt, wl->wl_inohashcnt)); | | 1718 | wl->wl_dealloccnt, wl->wl_inohashcnt)); |
1674 | #endif | | 1719 | #endif |
1675 | | | 1720 | |
1676 | | | 1721 | |
1677 | mutex_enter(&bufcache_lock); | | 1722 | mutex_enter(&bufcache_lock); |
1678 | mutex_enter(&wl->wl_mtx); | | 1723 | mutex_enter(&wl->wl_mtx); |
1679 | | | 1724 | |
1680 | wl->wl_reserved_bytes = reserved; | | 1725 | wl->wl_reserved_bytes = reserved; |
1681 | wl->wl_head = head; | | 1726 | wl->wl_head = head; |
1682 | wl->wl_tail = tail; | | 1727 | wl->wl_tail = tail; |
1683 | KASSERT(wl->wl_reclaimable_bytes >= delta); | | 1728 | KASSERT(wl->wl_reclaimable_bytes >= delta); |
1684 | wl->wl_reclaimable_bytes -= delta; | | 1729 | wl->wl_reclaimable_bytes -= delta; |
1685 | KDASSERT(wl->wl_dealloccnt == 0); | | 1730 | KDASSERT(wl->wl_dealloccnt == 0); |
1686 | #ifdef WAPBL_DEBUG_BUFBYTES | | 1731 | #ifdef WAPBL_DEBUG_BUFBYTES |
1687 | wl->wl_unsynced_bufbytes += wl->wl_bufbytes; | | 1732 | wl->wl_unsynced_bufbytes += wl->wl_bufbytes; |
1688 | #endif | | 1733 | #endif |
1689 | | | 1734 | |
1690 | we->we_wapbl = wl; | | 1735 | we->we_wapbl = wl; |
1691 | we->we_bufcount = wl->wl_bufcount; | | 1736 | we->we_bufcount = wl->wl_bufcount; |
1692 | #ifdef WAPBL_DEBUG_BUFBYTES | | 1737 | #ifdef WAPBL_DEBUG_BUFBYTES |
1693 | we->we_unsynced_bufbytes = wl->wl_bufbytes; | | 1738 | we->we_unsynced_bufbytes = wl->wl_bufbytes; |
1694 | #endif | | 1739 | #endif |
1695 | we->we_reclaimable_bytes = flushsize; | | 1740 | we->we_reclaimable_bytes = flushsize; |
1696 | we->we_error = 0; | | 1741 | we->we_error = 0; |
1697 | SIMPLEQ_INSERT_TAIL(&wl->wl_entries, we, we_entries); | | 1742 | SIMPLEQ_INSERT_TAIL(&wl->wl_entries, we, we_entries); |
1698 | | | 1743 | |
1699 | /* | | 1744 | /* |
1700 | * this flushes bufs in reverse order than they were queued | | 1745 | * this flushes bufs in reverse order than they were queued |
1701 | * it shouldn't matter, but if we care we could use TAILQ instead. | | 1746 | * it shouldn't matter, but if we care we could use TAILQ instead. |
1702 | * XXX Note they will get put on the lru queue when they flush | | 1747 | * XXX Note they will get put on the lru queue when they flush |
1703 | * so we might actually want to change this to preserve order. | | 1748 | * so we might actually want to change this to preserve order. |
1704 | */ | | 1749 | */ |
1705 | while ((bp = LIST_FIRST(&wl->wl_bufs)) != NULL) { | | 1750 | while ((bp = LIST_FIRST(&wl->wl_bufs)) != NULL) { |
1706 | if (bbusy(bp, 0, 0, &wl->wl_mtx)) { | | 1751 | if (bbusy(bp, 0, 0, &wl->wl_mtx)) { |
1707 | continue; | | 1752 | continue; |
1708 | } | | 1753 | } |
1709 | bp->b_iodone = wapbl_biodone; | | 1754 | bp->b_iodone = wapbl_biodone; |
1710 | bp->b_private = we; | | 1755 | bp->b_private = we; |
1711 | bremfree(bp); | | 1756 | bremfree(bp); |
1712 | wapbl_remove_buf_locked(wl, bp); | | 1757 | wapbl_remove_buf_locked(wl, bp); |
1713 | mutex_exit(&wl->wl_mtx); | | 1758 | mutex_exit(&wl->wl_mtx); |
1714 | mutex_exit(&bufcache_lock); | | 1759 | mutex_exit(&bufcache_lock); |
1715 | bawrite(bp); | | 1760 | bawrite(bp); |
1716 | mutex_enter(&bufcache_lock); | | 1761 | mutex_enter(&bufcache_lock); |
1717 | mutex_enter(&wl->wl_mtx); | | 1762 | mutex_enter(&wl->wl_mtx); |
1718 | } | | 1763 | } |
1719 | mutex_exit(&wl->wl_mtx); | | 1764 | mutex_exit(&wl->wl_mtx); |
1720 | mutex_exit(&bufcache_lock); | | 1765 | mutex_exit(&bufcache_lock); |
1721 | | | 1766 | |
1722 | #if 0 | | 1767 | #if 0 |
1723 | WAPBL_PRINTF(WAPBL_PRINT_FLUSH, | | 1768 | WAPBL_PRINTF(WAPBL_PRINT_FLUSH, |
1724 | ("wapbl_flush thread %d.%d done flushing entries...\n", | | 1769 | ("wapbl_flush thread %d.%d done flushing entries...\n", |
1725 | curproc->p_pid, curlwp->l_lid)); | | 1770 | curproc->p_pid, curlwp->l_lid)); |
1726 | #endif | | 1771 | #endif |
1727 | | | 1772 | |
1728 | wait_out: | | 1773 | wait_out: |
1729 | | | 1774 | |
1730 | /* | | 1775 | /* |
1731 | * If the waitfor flag is set, don't return until everything is | | 1776 | * If the waitfor flag is set, don't return until everything is |
1732 | * fully flushed and the on disk log is empty. | | 1777 | * fully flushed and the on disk log is empty. |
1733 | */ | | 1778 | */ |
1734 | if (waitfor) { | | 1779 | if (waitfor) { |
1735 | error = wapbl_truncate(wl, wl->wl_circ_size - | | 1780 | error = wapbl_truncate(wl, wl->wl_circ_size - |
1736 | wl->wl_reserved_bytes); | | 1781 | wl->wl_reserved_bytes); |
1737 | } | | 1782 | } |
1738 | | | 1783 | |
1739 | out: | | 1784 | out: |
1740 | if (error) { | | 1785 | if (error) { |
1741 | wl->wl_flush_abort(wl->wl_mount, | | 1786 | wl->wl_flush_abort(wl->wl_mount, |
1742 | TAILQ_FIRST(&wl->wl_dealloclist)); | | 1787 | TAILQ_FIRST(&wl->wl_dealloclist)); |
1743 | } | | 1788 | } |
1744 | | | 1789 | |
1745 | #ifdef WAPBL_DEBUG_PRINT | | 1790 | #ifdef WAPBL_DEBUG_PRINT |
1746 | if (error) { | | 1791 | if (error) { |
1747 | pid_t pid = -1; | | 1792 | pid_t pid = -1; |
1748 | lwpid_t lid = -1; | | 1793 | lwpid_t lid = -1; |
1749 | if (curproc) | | 1794 | if (curproc) |
1750 | pid = curproc->p_pid; | | 1795 | pid = curproc->p_pid; |
1751 | if (curlwp) | | 1796 | if (curlwp) |
1752 | lid = curlwp->l_lid; | | 1797 | lid = curlwp->l_lid; |
1753 | mutex_enter(&wl->wl_mtx); | | 1798 | mutex_enter(&wl->wl_mtx); |
1754 | #ifdef WAPBL_DEBUG_BUFBYTES | | 1799 | #ifdef WAPBL_DEBUG_BUFBYTES |
1755 | WAPBL_PRINTF(WAPBL_PRINT_ERROR, | | 1800 | WAPBL_PRINTF(WAPBL_PRINT_ERROR, |
1756 | ("wapbl_flush: thread %d.%d aborted flush: " | | 1801 | ("wapbl_flush: thread %d.%d aborted flush: " |
1757 | "error = %d\n" | | 1802 | "error = %d\n" |
1758 | "\tbufcount=%zu bufbytes=%zu bcount=%zu " | | 1803 | "\tbufcount=%zu bufbytes=%zu bcount=%zu " |
1759 | "deallocs=%d inodes=%d\n" | | 1804 | "deallocs=%d inodes=%d\n" |
1760 | "\terrcnt = %d, reclaimable=%zu reserved=%zu " | | 1805 | "\terrcnt = %d, reclaimable=%zu reserved=%zu " |
1761 | "unsynced=%zu\n", | | 1806 | "unsynced=%zu\n", |
1762 | pid, lid, error, wl->wl_bufcount, | | 1807 | pid, lid, error, wl->wl_bufcount, |
1763 | wl->wl_bufbytes, wl->wl_bcount, | | 1808 | wl->wl_bufbytes, wl->wl_bcount, |
1764 | wl->wl_dealloccnt, wl->wl_inohashcnt, | | 1809 | wl->wl_dealloccnt, wl->wl_inohashcnt, |
1765 | wl->wl_error_count, wl->wl_reclaimable_bytes, | | 1810 | wl->wl_error_count, wl->wl_reclaimable_bytes, |
1766 | wl->wl_reserved_bytes, wl->wl_unsynced_bufbytes)); | | 1811 | wl->wl_reserved_bytes, wl->wl_unsynced_bufbytes)); |
1767 | SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { | | 1812 | SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { |
1768 | WAPBL_PRINTF(WAPBL_PRINT_ERROR, | | 1813 | WAPBL_PRINTF(WAPBL_PRINT_ERROR, |
1769 | ("\tentry: bufcount = %zu, reclaimable = %zu, " | | 1814 | ("\tentry: bufcount = %zu, reclaimable = %zu, " |
1770 | "error = %d, unsynced = %zu\n", | | 1815 | "error = %d, unsynced = %zu\n", |
1771 | we->we_bufcount, we->we_reclaimable_bytes, | | 1816 | we->we_bufcount, we->we_reclaimable_bytes, |
1772 | we->we_error, we->we_unsynced_bufbytes)); | | 1817 | we->we_error, we->we_unsynced_bufbytes)); |
1773 | } | | 1818 | } |
1774 | #else | | 1819 | #else |
1775 | WAPBL_PRINTF(WAPBL_PRINT_ERROR, | | 1820 | WAPBL_PRINTF(WAPBL_PRINT_ERROR, |
1776 | ("wapbl_flush: thread %d.%d aborted flush: " | | 1821 | ("wapbl_flush: thread %d.%d aborted flush: " |
1777 | "error = %d\n" | | 1822 | "error = %d\n" |
1778 | "\tbufcount=%zu bufbytes=%zu bcount=%zu " | | 1823 | "\tbufcount=%zu bufbytes=%zu bcount=%zu " |
1779 | "deallocs=%d inodes=%d\n" | | 1824 | "deallocs=%d inodes=%d\n" |
1780 | "\terrcnt = %d, reclaimable=%zu reserved=%zu\n", | | 1825 | "\terrcnt = %d, reclaimable=%zu reserved=%zu\n", |
1781 | pid, lid, error, wl->wl_bufcount, | | 1826 | pid, lid, error, wl->wl_bufcount, |
1782 | wl->wl_bufbytes, wl->wl_bcount, | | 1827 | wl->wl_bufbytes, wl->wl_bcount, |
1783 | wl->wl_dealloccnt, wl->wl_inohashcnt, | | 1828 | wl->wl_dealloccnt, wl->wl_inohashcnt, |
1784 | wl->wl_error_count, wl->wl_reclaimable_bytes, | | 1829 | wl->wl_error_count, wl->wl_reclaimable_bytes, |
1785 | wl->wl_reserved_bytes)); | | 1830 | wl->wl_reserved_bytes)); |
1786 | SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { | | 1831 | SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { |
1787 | WAPBL_PRINTF(WAPBL_PRINT_ERROR, | | 1832 | WAPBL_PRINTF(WAPBL_PRINT_ERROR, |
1788 | ("\tentry: bufcount = %zu, reclaimable = %zu, " | | 1833 | ("\tentry: bufcount = %zu, reclaimable = %zu, " |
1789 | "error = %d\n", we->we_bufcount, | | 1834 | "error = %d\n", we->we_bufcount, |
1790 | we->we_reclaimable_bytes, we->we_error)); | | 1835 | we->we_reclaimable_bytes, we->we_error)); |
1791 | } | | 1836 | } |
1792 | #endif | | 1837 | #endif |
1793 | mutex_exit(&wl->wl_mtx); | | 1838 | mutex_exit(&wl->wl_mtx); |
1794 | } | | 1839 | } |
1795 | #endif | | 1840 | #endif |
1796 | | | 1841 | |
1797 | rw_exit(&wl->wl_rwlock); | | 1842 | rw_exit(&wl->wl_rwlock); |
1798 | return error; | | 1843 | return error; |
1799 | } | | 1844 | } |
1800 | | | 1845 | |
1801 | /****************************************************************/ | | 1846 | /****************************************************************/ |
1802 | | | 1847 | |
1803 | void | | 1848 | void |
1804 | wapbl_jlock_assert(struct wapbl *wl) | | 1849 | wapbl_jlock_assert(struct wapbl *wl) |
1805 | { | | 1850 | { |
1806 | | | 1851 | |
1807 | KASSERT(rw_lock_held(&wl->wl_rwlock)); | | 1852 | KASSERT(rw_lock_held(&wl->wl_rwlock)); |
1808 | } | | 1853 | } |
1809 | | | 1854 | |
1810 | void | | 1855 | void |
1811 | wapbl_junlock_assert(struct wapbl *wl) | | 1856 | wapbl_junlock_assert(struct wapbl *wl) |
1812 | { | | 1857 | { |
1813 | | | 1858 | |
1814 | KASSERT(!rw_write_held(&wl->wl_rwlock)); | | 1859 | KASSERT(!rw_write_held(&wl->wl_rwlock)); |
1815 | } | | 1860 | } |
1816 | | | 1861 | |
1817 | /****************************************************************/ | | 1862 | /****************************************************************/ |
1818 | | | 1863 | |
1819 | /* locks missing */ | | 1864 | /* locks missing */ |
1820 | void | | 1865 | void |
1821 | wapbl_print(struct wapbl *wl, | | 1866 | wapbl_print(struct wapbl *wl, |
1822 | int full, | | 1867 | int full, |
1823 | void (*pr)(const char *, ...)) | | 1868 | void (*pr)(const char *, ...)) |
1824 | { | | 1869 | { |
1825 | struct buf *bp; | | 1870 | struct buf *bp; |
1826 | struct wapbl_entry *we; | | 1871 | struct wapbl_entry *we; |
1827 | (*pr)("wapbl %p", wl); | | 1872 | (*pr)("wapbl %p", wl); |
1828 | (*pr)("\nlogvp = %p, devvp = %p, logpbn = %"PRId64"\n", | | 1873 | (*pr)("\nlogvp = %p, devvp = %p, logpbn = %"PRId64"\n", |
1829 | wl->wl_logvp, wl->wl_devvp, wl->wl_logpbn); | | 1874 | wl->wl_logvp, wl->wl_devvp, wl->wl_logpbn); |
1830 | (*pr)("circ = %zu, header = %zu, head = %"PRIdMAX" tail = %"PRIdMAX"\n", | | 1875 | (*pr)("circ = %zu, header = %zu, head = %"PRIdMAX" tail = %"PRIdMAX"\n", |
1831 | wl->wl_circ_size, wl->wl_circ_off, | | 1876 | wl->wl_circ_size, wl->wl_circ_off, |
1832 | (intmax_t)wl->wl_head, (intmax_t)wl->wl_tail); | | 1877 | (intmax_t)wl->wl_head, (intmax_t)wl->wl_tail); |
1833 | (*pr)("fs_dev_bshift = %d, log_dev_bshift = %d\n", | | 1878 | (*pr)("fs_dev_bshift = %d, log_dev_bshift = %d\n", |
1834 | wl->wl_log_dev_bshift, wl->wl_fs_dev_bshift); | | 1879 | wl->wl_log_dev_bshift, wl->wl_fs_dev_bshift); |
1835 | #ifdef WAPBL_DEBUG_BUFBYTES | | 1880 | #ifdef WAPBL_DEBUG_BUFBYTES |
1836 | (*pr)("bufcount = %zu, bufbytes = %zu bcount = %zu reclaimable = %zu " | | 1881 | (*pr)("bufcount = %zu, bufbytes = %zu bcount = %zu reclaimable = %zu " |
1837 | "reserved = %zu errcnt = %d unsynced = %zu\n", | | 1882 | "reserved = %zu errcnt = %d unsynced = %zu\n", |
1838 | wl->wl_bufcount, wl->wl_bufbytes, wl->wl_bcount, | | 1883 | wl->wl_bufcount, wl->wl_bufbytes, wl->wl_bcount, |
1839 | wl->wl_reclaimable_bytes, wl->wl_reserved_bytes, | | 1884 | wl->wl_reclaimable_bytes, wl->wl_reserved_bytes, |
1840 | wl->wl_error_count, wl->wl_unsynced_bufbytes); | | 1885 | wl->wl_error_count, wl->wl_unsynced_bufbytes); |
1841 | #else | | 1886 | #else |
1842 | (*pr)("bufcount = %zu, bufbytes = %zu bcount = %zu reclaimable = %zu " | | 1887 | (*pr)("bufcount = %zu, bufbytes = %zu bcount = %zu reclaimable = %zu " |
1843 | "reserved = %zu errcnt = %d\n", wl->wl_bufcount, wl->wl_bufbytes, | | 1888 | "reserved = %zu errcnt = %d\n", wl->wl_bufcount, wl->wl_bufbytes, |
1844 | wl->wl_bcount, wl->wl_reclaimable_bytes, wl->wl_reserved_bytes, | | 1889 | wl->wl_bcount, wl->wl_reclaimable_bytes, wl->wl_reserved_bytes, |
1845 | wl->wl_error_count); | | 1890 | wl->wl_error_count); |
1846 | #endif | | 1891 | #endif |
1847 | (*pr)("\tdealloccnt = %d, dealloclim = %d\n", | | 1892 | (*pr)("\tdealloccnt = %d, dealloclim = %d\n", |
1848 | wl->wl_dealloccnt, wl->wl_dealloclim); | | 1893 | wl->wl_dealloccnt, wl->wl_dealloclim); |
1849 | (*pr)("\tinohashcnt = %d, inohashmask = 0x%08x\n", | | 1894 | (*pr)("\tinohashcnt = %d, inohashmask = 0x%08x\n", |
1850 | wl->wl_inohashcnt, wl->wl_inohashmask); | | 1895 | wl->wl_inohashcnt, wl->wl_inohashmask); |
1851 | (*pr)("entries:\n"); | | 1896 | (*pr)("entries:\n"); |
1852 | SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { | | 1897 | SIMPLEQ_FOREACH(we, &wl->wl_entries, we_entries) { |
1853 | #ifdef WAPBL_DEBUG_BUFBYTES | | 1898 | #ifdef WAPBL_DEBUG_BUFBYTES |
1854 | (*pr)("\tbufcount = %zu, reclaimable = %zu, error = %d, " | | 1899 | (*pr)("\tbufcount = %zu, reclaimable = %zu, error = %d, " |
1855 | "unsynced = %zu\n", | | 1900 | "unsynced = %zu\n", |
1856 | we->we_bufcount, we->we_reclaimable_bytes, | | 1901 | we->we_bufcount, we->we_reclaimable_bytes, |
1857 | we->we_error, we->we_unsynced_bufbytes); | | 1902 | we->we_error, we->we_unsynced_bufbytes); |
1858 | #else | | 1903 | #else |
1859 | (*pr)("\tbufcount = %zu, reclaimable = %zu, error = %d\n", | | 1904 | (*pr)("\tbufcount = %zu, reclaimable = %zu, error = %d\n", |
1860 | we->we_bufcount, we->we_reclaimable_bytes, we->we_error); | | 1905 | we->we_bufcount, we->we_reclaimable_bytes, we->we_error); |
1861 | #endif | | 1906 | #endif |
1862 | } | | 1907 | } |
1863 | if (full) { | | 1908 | if (full) { |
1864 | int cnt = 0; | | 1909 | int cnt = 0; |
1865 | (*pr)("bufs ="); | | 1910 | (*pr)("bufs ="); |
1866 | LIST_FOREACH(bp, &wl->wl_bufs, b_wapbllist) { | | 1911 | LIST_FOREACH(bp, &wl->wl_bufs, b_wapbllist) { |
1867 | if (!LIST_NEXT(bp, b_wapbllist)) { | | 1912 | if (!LIST_NEXT(bp, b_wapbllist)) { |
1868 | (*pr)(" %p", bp); | | 1913 | (*pr)(" %p", bp); |
1869 | } else if ((++cnt % 6) == 0) { | | 1914 | } else if ((++cnt % 6) == 0) { |
1870 | (*pr)(" %p,\n\t", bp); | | 1915 | (*pr)(" %p,\n\t", bp); |
1871 | } else { | | 1916 | } else { |
1872 | (*pr)(" %p,", bp); | | 1917 | (*pr)(" %p,", bp); |
1873 | } | | 1918 | } |
1874 | } | | 1919 | } |
1875 | (*pr)("\n"); | | 1920 | (*pr)("\n"); |
1876 | | | 1921 | |
1877 | (*pr)("dealloced blks = "); | | 1922 | (*pr)("dealloced blks = "); |
1878 | { | | 1923 | { |
1879 | struct wapbl_dealloc *wd; | | 1924 | struct wapbl_dealloc *wd; |
1880 | cnt = 0; | | 1925 | cnt = 0; |
1881 | TAILQ_FOREACH(wd, &wl->wl_dealloclist, wd_entries) { | | 1926 | TAILQ_FOREACH(wd, &wl->wl_dealloclist, wd_entries) { |
1882 | (*pr)(" %"PRId64":%d,", | | 1927 | (*pr)(" %"PRId64":%d,", |
1883 | wd->wd_blkno, | | 1928 | wd->wd_blkno, |
1884 | wd->wd_len); | | 1929 | wd->wd_len); |
1885 | if ((++cnt % 4) == 0) { | | 1930 | if ((++cnt % 4) == 0) { |
1886 | (*pr)("\n\t"); | | 1931 | (*pr)("\n\t"); |
1887 | } | | 1932 | } |
1888 | } | | 1933 | } |
1889 | } | | 1934 | } |
1890 | (*pr)("\n"); | | 1935 | (*pr)("\n"); |
1891 | | | 1936 | |
1892 | (*pr)("registered inodes = "); | | 1937 | (*pr)("registered inodes = "); |
1893 | { | | 1938 | { |
1894 | int i; | | 1939 | int i; |
1895 | cnt = 0; | | 1940 | cnt = 0; |
1896 | for (i = 0; i <= wl->wl_inohashmask; i++) { | | 1941 | for (i = 0; i <= wl->wl_inohashmask; i++) { |
1897 | struct wapbl_ino_head *wih; | | 1942 | struct wapbl_ino_head *wih; |
1898 | struct wapbl_ino *wi; | | 1943 | struct wapbl_ino *wi; |
1899 | | | 1944 | |
1900 | wih = &wl->wl_inohash[i]; | | 1945 | wih = &wl->wl_inohash[i]; |
1901 | LIST_FOREACH(wi, wih, wi_hash) { | | 1946 | LIST_FOREACH(wi, wih, wi_hash) { |
1902 | if (wi->wi_ino == 0) | | 1947 | if (wi->wi_ino == 0) |
1903 | continue; | | 1948 | continue; |
1904 | (*pr)(" %"PRIu64"/0%06"PRIo32",", | | 1949 | (*pr)(" %"PRIu64"/0%06"PRIo32",", |
1905 | wi->wi_ino, wi->wi_mode); | | 1950 | wi->wi_ino, wi->wi_mode); |
1906 | if ((++cnt % 4) == 0) { | | 1951 | if ((++cnt % 4) == 0) { |
1907 | (*pr)("\n\t"); | | 1952 | (*pr)("\n\t"); |
1908 | } | | 1953 | } |
1909 | } | | 1954 | } |
1910 | } | | 1955 | } |
1911 | (*pr)("\n"); | | 1956 | (*pr)("\n"); |
1912 | } | | 1957 | } |
1913 | } | | 1958 | } |
1914 | } | | 1959 | } |
1915 | | | 1960 | |
1916 | #if defined(WAPBL_DEBUG) || defined(DDB) | | 1961 | #if defined(WAPBL_DEBUG) || defined(DDB) |
1917 | void | | 1962 | void |
1918 | wapbl_dump(struct wapbl *wl) | | 1963 | wapbl_dump(struct wapbl *wl) |
1919 | { | | 1964 | { |
1920 | #if defined(WAPBL_DEBUG) | | 1965 | #if defined(WAPBL_DEBUG) |
1921 | if (!wl) | | 1966 | if (!wl) |
1922 | wl = wapbl_debug_wl; | | 1967 | wl = wapbl_debug_wl; |
1923 | #endif | | 1968 | #endif |
1924 | if (!wl) | | 1969 | if (!wl) |
1925 | return; | | 1970 | return; |
1926 | wapbl_print(wl, 1, printf); | | 1971 | wapbl_print(wl, 1, printf); |
1927 | } | | 1972 | } |
1928 | #endif | | 1973 | #endif |
1929 | | | 1974 | |
1930 | /****************************************************************/ | | 1975 | /****************************************************************/ |
1931 | | | 1976 | |
1932 | int | | 1977 | int |
1933 | wapbl_register_deallocation(struct wapbl *wl, daddr_t blk, int len, bool force, | | 1978 | wapbl_register_deallocation(struct wapbl *wl, daddr_t blk, int len, bool force, |
1934 | void **cookiep) | | 1979 | void **cookiep) |
1935 | { | | 1980 | { |
1936 | struct wapbl_dealloc *wd; | | 1981 | struct wapbl_dealloc *wd; |
1937 | int error = 0; | | 1982 | int error = 0; |
1938 | | | 1983 | |
1939 | wapbl_jlock_assert(wl); | | 1984 | wapbl_jlock_assert(wl); |
1940 | | | 1985 | |
1941 | mutex_enter(&wl->wl_mtx); | | 1986 | mutex_enter(&wl->wl_mtx); |
1942 | | | 1987 | |
1943 | if (__predict_false(wl->wl_dealloccnt >= wl->wl_dealloclim)) { | | 1988 | if (__predict_false(wl->wl_dealloccnt >= wl->wl_dealloclim)) { |
1944 | if (!force) { | | 1989 | if (!force) { |
1945 | error = EAGAIN; | | 1990 | error = EAGAIN; |
1946 | goto out; | | 1991 | goto out; |
1947 | } | | 1992 | } |
1948 | | | 1993 | |
1949 | /* | | 1994 | /* |
1950 | * Forced registration can only be used when: | | 1995 | * Forced registration can only be used when: |
1951 | * 1) the caller can't cope with failure | | 1996 | * 1) the caller can't cope with failure |
1952 | * 2) the path can be triggered only bounded, small | | 1997 | * 2) the path can be triggered only bounded, small |
1953 | * times per transaction | | 1998 | * times per transaction |
1954 | * If this is not fullfilled, and the path would be triggered | | 1999 | * If this is not fullfilled, and the path would be triggered |
1955 | * many times, this could overflow maximum transaction size | | 2000 | * many times, this could overflow maximum transaction size |
1956 | * and panic later. | | 2001 | * and panic later. |
1957 | */ | | 2002 | */ |
1958 | printf("%s: forced dealloc registration over limit: %d >= %d\n", | | 2003 | printf("%s: forced dealloc registration over limit: %d >= %d\n", |
1959 | wl->wl_mount->mnt_stat.f_mntonname, | | 2004 | wl->wl_mount->mnt_stat.f_mntonname, |
1960 | wl->wl_dealloccnt, wl->wl_dealloclim); | | 2005 | wl->wl_dealloccnt, wl->wl_dealloclim); |
1961 | } | | 2006 | } |
1962 | | | 2007 | |
1963 | wl->wl_dealloccnt++; | | 2008 | wl->wl_dealloccnt++; |
1964 | mutex_exit(&wl->wl_mtx); | | 2009 | mutex_exit(&wl->wl_mtx); |
1965 | | | 2010 | |
1966 | wd = pool_get(&wapbl_dealloc_pool, PR_WAITOK); | | 2011 | wd = pool_get(&wapbl_dealloc_pool, PR_WAITOK); |
1967 | wd->wd_blkno = blk; | | 2012 | wd->wd_blkno = blk; |
1968 | wd->wd_len = len; | | 2013 | wd->wd_len = len; |
1969 | | | 2014 | |
1970 | mutex_enter(&wl->wl_mtx); | | 2015 | mutex_enter(&wl->wl_mtx); |
1971 | TAILQ_INSERT_TAIL(&wl->wl_dealloclist, wd, wd_entries); | | 2016 | TAILQ_INSERT_TAIL(&wl->wl_dealloclist, wd, wd_entries); |
1972 | | | 2017 | |
1973 | if (cookiep) | | 2018 | if (cookiep) |
1974 | *cookiep = wd; | | 2019 | *cookiep = wd; |
1975 | | | 2020 | |
1976 | out: | | 2021 | out: |
1977 | mutex_exit(&wl->wl_mtx); | | 2022 | mutex_exit(&wl->wl_mtx); |
1978 | | | 2023 | |
1979 | WAPBL_PRINTF(WAPBL_PRINT_ALLOC, | | 2024 | WAPBL_PRINTF(WAPBL_PRINT_ALLOC, |
1980 | ("wapbl_register_deallocation: blk=%"PRId64" len=%d error=%d\n", | | 2025 | ("wapbl_register_deallocation: blk=%"PRId64" len=%d error=%d\n", |
1981 | blk, len, error)); | | 2026 | blk, len, error)); |
1982 | | | 2027 | |
1983 | return error; | | 2028 | return error; |
1984 | } | | 2029 | } |
1985 | | | 2030 | |
1986 | static void | | 2031 | static void |
1987 | wapbl_deallocation_free(struct wapbl *wl, struct wapbl_dealloc *wd, | | 2032 | wapbl_deallocation_free(struct wapbl *wl, struct wapbl_dealloc *wd, |
1988 | bool locked) | | 2033 | bool locked) |
1989 | { | | 2034 | { |
1990 | KASSERT(!locked | | 2035 | KASSERT(!locked |
1991 | || rw_lock_held(&wl->wl_rwlock) || mutex_owned(&wl->wl_mtx)); | | 2036 | || rw_lock_held(&wl->wl_rwlock) || mutex_owned(&wl->wl_mtx)); |
1992 | | | 2037 | |
1993 | if (!locked) | | 2038 | if (!locked) |
1994 | mutex_enter(&wl->wl_mtx); | | 2039 | mutex_enter(&wl->wl_mtx); |
1995 | | | 2040 | |
1996 | TAILQ_REMOVE(&wl->wl_dealloclist, wd, wd_entries); | | 2041 | TAILQ_REMOVE(&wl->wl_dealloclist, wd, wd_entries); |
1997 | wl->wl_dealloccnt--; | | 2042 | wl->wl_dealloccnt--; |
1998 | | | 2043 | |
1999 | if (!locked) | | 2044 | if (!locked) |
2000 | mutex_exit(&wl->wl_mtx); | | 2045 | mutex_exit(&wl->wl_mtx); |
2001 | | | 2046 | |
2002 | pool_put(&wapbl_dealloc_pool, wd); | | 2047 | pool_put(&wapbl_dealloc_pool, wd); |
2003 | } | | 2048 | } |
2004 | | | 2049 | |
2005 | void | | 2050 | void |
2006 | wapbl_unregister_deallocation(struct wapbl *wl, void *cookie) | | 2051 | wapbl_unregister_deallocation(struct wapbl *wl, void *cookie) |
2007 | { | | 2052 | { |
2008 | KASSERT(cookie != NULL); | | 2053 | KASSERT(cookie != NULL); |
2009 | wapbl_deallocation_free(wl, cookie, false); | | 2054 | wapbl_deallocation_free(wl, cookie, false); |
2010 | } | | 2055 | } |
2011 | | | 2056 | |
2012 | /****************************************************************/ | | 2057 | /****************************************************************/ |
2013 | | | 2058 | |
2014 | static void | | 2059 | static void |
2015 | wapbl_inodetrk_init(struct wapbl *wl, u_int size) | | 2060 | wapbl_inodetrk_init(struct wapbl *wl, u_int size) |
2016 | { | | 2061 | { |
2017 | | | 2062 | |
2018 | wl->wl_inohash = hashinit(size, HASH_LIST, true, &wl->wl_inohashmask); | | 2063 | wl->wl_inohash = hashinit(size, HASH_LIST, true, &wl->wl_inohashmask); |
2019 | if (atomic_inc_uint_nv(&wapbl_ino_pool_refcount) == 1) { | | 2064 | if (atomic_inc_uint_nv(&wapbl_ino_pool_refcount) == 1) { |
2020 | pool_init(&wapbl_ino_pool, sizeof(struct wapbl_ino), 0, 0, 0, | | 2065 | pool_init(&wapbl_ino_pool, sizeof(struct wapbl_ino), 0, 0, 0, |
2021 | "wapblinopl", &pool_allocator_nointr, IPL_NONE); | | 2066 | "wapblinopl", &pool_allocator_nointr, IPL_NONE); |
2022 | } | | 2067 | } |
2023 | } | | 2068 | } |
2024 | | | 2069 | |
2025 | static void | | 2070 | static void |
2026 | wapbl_inodetrk_free(struct wapbl *wl) | | 2071 | wapbl_inodetrk_free(struct wapbl *wl) |
2027 | { | | 2072 | { |
2028 | | | 2073 | |
2029 | /* XXX this KASSERT needs locking/mutex analysis */ | | 2074 | /* XXX this KASSERT needs locking/mutex analysis */ |
2030 | KASSERT(wl->wl_inohashcnt == 0); | | 2075 | KASSERT(wl->wl_inohashcnt == 0); |
2031 | hashdone(wl->wl_inohash, HASH_LIST, wl->wl_inohashmask); | | 2076 | hashdone(wl->wl_inohash, HASH_LIST, wl->wl_inohashmask); |
2032 | if (atomic_dec_uint_nv(&wapbl_ino_pool_refcount) == 0) { | | 2077 | if (atomic_dec_uint_nv(&wapbl_ino_pool_refcount) == 0) { |
2033 | pool_destroy(&wapbl_ino_pool); | | 2078 | pool_destroy(&wapbl_ino_pool); |
2034 | } | | 2079 | } |
2035 | } | | 2080 | } |
2036 | | | 2081 | |
2037 | static struct wapbl_ino * | | 2082 | static struct wapbl_ino * |
2038 | wapbl_inodetrk_get(struct wapbl *wl, ino_t ino) | | 2083 | wapbl_inodetrk_get(struct wapbl *wl, ino_t ino) |
2039 | { | | 2084 | { |
2040 | struct wapbl_ino_head *wih; | | 2085 | struct wapbl_ino_head *wih; |
2041 | struct wapbl_ino *wi; | | 2086 | struct wapbl_ino *wi; |
2042 | | | 2087 | |
2043 | KASSERT(mutex_owned(&wl->wl_mtx)); | | 2088 | KASSERT(mutex_owned(&wl->wl_mtx)); |
2044 | | | 2089 | |
2045 | wih = &wl->wl_inohash[ino & wl->wl_inohashmask]; | | 2090 | wih = &wl->wl_inohash[ino & wl->wl_inohashmask]; |
2046 | LIST_FOREACH(wi, wih, wi_hash) { | | 2091 | LIST_FOREACH(wi, wih, wi_hash) { |
2047 | if (ino == wi->wi_ino) | | 2092 | if (ino == wi->wi_ino) |
2048 | return wi; | | 2093 | return wi; |
2049 | } | | 2094 | } |
2050 | return 0; | | 2095 | return 0; |
2051 | } | | 2096 | } |
2052 | | | 2097 | |
2053 | void | | 2098 | void |
2054 | wapbl_register_inode(struct wapbl *wl, ino_t ino, mode_t mode) | | 2099 | wapbl_register_inode(struct wapbl *wl, ino_t ino, mode_t mode) |
2055 | { | | 2100 | { |
2056 | struct wapbl_ino_head *wih; | | 2101 | struct wapbl_ino_head *wih; |
2057 | struct wapbl_ino *wi; | | 2102 | struct wapbl_ino *wi; |
2058 | | | 2103 | |
2059 | wi = pool_get(&wapbl_ino_pool, PR_WAITOK); | | 2104 | wi = pool_get(&wapbl_ino_pool, PR_WAITOK); |
2060 | | | 2105 | |
2061 | mutex_enter(&wl->wl_mtx); | | 2106 | mutex_enter(&wl->wl_mtx); |
2062 | if (wapbl_inodetrk_get(wl, ino) == NULL) { | | 2107 | if (wapbl_inodetrk_get(wl, ino) == NULL) { |
2063 | wi->wi_ino = ino; | | 2108 | wi->wi_ino = ino; |
2064 | wi->wi_mode = mode; | | 2109 | wi->wi_mode = mode; |
2065 | wih = &wl->wl_inohash[ino & wl->wl_inohashmask]; | | 2110 | wih = &wl->wl_inohash[ino & wl->wl_inohashmask]; |
2066 | LIST_INSERT_HEAD(wih, wi, wi_hash); | | 2111 | LIST_INSERT_HEAD(wih, wi, wi_hash); |
2067 | wl->wl_inohashcnt++; | | 2112 | wl->wl_inohashcnt++; |
2068 | WAPBL_PRINTF(WAPBL_PRINT_INODE, | | 2113 | WAPBL_PRINTF(WAPBL_PRINT_INODE, |
2069 | ("wapbl_register_inode: ino=%"PRId64"\n", ino)); | | 2114 | ("wapbl_register_inode: ino=%"PRId64"\n", ino)); |
2070 | mutex_exit(&wl->wl_mtx); | | 2115 | mutex_exit(&wl->wl_mtx); |
2071 | } else { | | 2116 | } else { |
2072 | mutex_exit(&wl->wl_mtx); | | 2117 | mutex_exit(&wl->wl_mtx); |
2073 | pool_put(&wapbl_ino_pool, wi); | | 2118 | pool_put(&wapbl_ino_pool, wi); |
2074 | } | | 2119 | } |
2075 | } | | 2120 | } |
2076 | | | 2121 | |
2077 | void | | 2122 | void |
2078 | wapbl_unregister_inode(struct wapbl *wl, ino_t ino, mode_t mode) | | 2123 | wapbl_unregister_inode(struct wapbl *wl, ino_t ino, mode_t mode) |
2079 | { | | 2124 | { |
2080 | struct wapbl_ino *wi; | | 2125 | struct wapbl_ino *wi; |
2081 | | | 2126 | |
2082 | mutex_enter(&wl->wl_mtx); | | 2127 | mutex_enter(&wl->wl_mtx); |
2083 | wi = wapbl_inodetrk_get(wl, ino); | | 2128 | wi = wapbl_inodetrk_get(wl, ino); |
2084 | if (wi) { | | 2129 | if (wi) { |
2085 | WAPBL_PRINTF(WAPBL_PRINT_INODE, | | 2130 | WAPBL_PRINTF(WAPBL_PRINT_INODE, |
2086 | ("wapbl_unregister_inode: ino=%"PRId64"\n", ino)); | | 2131 | ("wapbl_unregister_inode: ino=%"PRId64"\n", ino)); |
2087 | KASSERT(wl->wl_inohashcnt > 0); | | 2132 | KASSERT(wl->wl_inohashcnt > 0); |
2088 | wl->wl_inohashcnt--; | | 2133 | wl->wl_inohashcnt--; |
2089 | LIST_REMOVE(wi, wi_hash); | | 2134 | LIST_REMOVE(wi, wi_hash); |
2090 | mutex_exit(&wl->wl_mtx); | | 2135 | mutex_exit(&wl->wl_mtx); |
2091 | | | 2136 | |
2092 | pool_put(&wapbl_ino_pool, wi); | | 2137 | pool_put(&wapbl_ino_pool, wi); |
2093 | } else { | | 2138 | } else { |
2094 | mutex_exit(&wl->wl_mtx); | | 2139 | mutex_exit(&wl->wl_mtx); |
2095 | } | | 2140 | } |
2096 | } | | 2141 | } |
2097 | | | 2142 | |
2098 | /****************************************************************/ | | 2143 | /****************************************************************/ |
2099 | | | 2144 | |
2100 | /* | | 2145 | /* |
2101 | * wapbl_transaction_inodes_len(wl) | | 2146 | * wapbl_transaction_inodes_len(wl) |
2102 | * | | 2147 | * |
2103 | * Calculate the number of bytes required for inode registration | | 2148 | * Calculate the number of bytes required for inode registration |
2104 | * log records in wl. | | 2149 | * log records in wl. |
2105 | */ | | 2150 | */ |
2106 | static inline size_t | | 2151 | static inline size_t |
2107 | wapbl_transaction_inodes_len(struct wapbl *wl) | | 2152 | wapbl_transaction_inodes_len(struct wapbl *wl) |
2108 | { | | 2153 | { |
2109 | int blocklen = 1<<wl->wl_log_dev_bshift; | | 2154 | int blocklen = 1<<wl->wl_log_dev_bshift; |
2110 | int iph; | | 2155 | int iph; |
2111 | | | 2156 | |
2112 | /* Calculate number of inodes described in a inodelist header */ | | 2157 | /* Calculate number of inodes described in a inodelist header */ |
2113 | iph = (blocklen - offsetof(struct wapbl_wc_inodelist, wc_inodes)) / | | 2158 | iph = (blocklen - offsetof(struct wapbl_wc_inodelist, wc_inodes)) / |
2114 | sizeof(((struct wapbl_wc_inodelist *)0)->wc_inodes[0]); | | 2159 | sizeof(((struct wapbl_wc_inodelist *)0)->wc_inodes[0]); |
2115 | | | 2160 | |
2116 | KASSERT(iph > 0); | | 2161 | KASSERT(iph > 0); |
2117 | | | 2162 | |
2118 | return MAX(1, howmany(wl->wl_inohashcnt, iph)) * blocklen; | | 2163 | return MAX(1, howmany(wl->wl_inohashcnt, iph)) * blocklen; |
2119 | } | | 2164 | } |
2120 | | | 2165 | |
2121 | | | 2166 | |
2122 | /* | | 2167 | /* |
2123 | * wapbl_transaction_len(wl) | | 2168 | * wapbl_transaction_len(wl) |
2124 | * | | 2169 | * |
2125 | * Calculate number of bytes required for all log records in wl. | | 2170 | * Calculate number of bytes required for all log records in wl. |
2126 | */ | | 2171 | */ |
2127 | static size_t | | 2172 | static size_t |
2128 | wapbl_transaction_len(struct wapbl *wl) | | 2173 | wapbl_transaction_len(struct wapbl *wl) |
2129 | { | | 2174 | { |
2130 | int blocklen = 1<<wl->wl_log_dev_bshift; | | 2175 | int blocklen = 1<<wl->wl_log_dev_bshift; |
2131 | size_t len; | | 2176 | size_t len; |
2132 | | | 2177 | |
2133 | /* Calculate number of blocks described in a blocklist header */ | | 2178 | /* Calculate number of blocks described in a blocklist header */ |
2134 | len = wl->wl_bcount; | | 2179 | len = wl->wl_bcount; |
2135 | len += howmany(wl->wl_bufcount, wl->wl_brperjblock) * blocklen; | | 2180 | len += howmany(wl->wl_bufcount, wl->wl_brperjblock) * blocklen; |
2136 | len += howmany(wl->wl_dealloccnt, wl->wl_brperjblock) * blocklen; | | 2181 | len += howmany(wl->wl_dealloccnt, wl->wl_brperjblock) * blocklen; |
2137 | len += wapbl_transaction_inodes_len(wl); | | 2182 | len += wapbl_transaction_inodes_len(wl); |
2138 | | | 2183 | |
2139 | return len; | | 2184 | return len; |
2140 | } | | 2185 | } |
2141 | | | 2186 | |
2142 | /* | | 2187 | /* |
2143 | * wapbl_cache_sync(wl, msg) | | 2188 | * wapbl_cache_sync(wl, msg) |
2144 | * | | 2189 | * |
2145 | * Issue DIOCCACHESYNC to wl->wl_devvp. | | 2190 | * Issue DIOCCACHESYNC to wl->wl_devvp. |
2146 | * | | 2191 | * |
2147 | * If sysctl(vfs.wapbl.verbose_commit) >= 2, print a message | | 2192 | * If sysctl(vfs.wapbl.verbose_commit) >= 2, print a message |
2148 | * including msg about the duration of the cache sync. | | 2193 | * including msg about the duration of the cache sync. |
2149 | */ | | 2194 | */ |
2150 | static int | | 2195 | static int |
2151 | wapbl_cache_sync(struct wapbl *wl, const char *msg) | | 2196 | wapbl_cache_sync(struct wapbl *wl, const char *msg) |
2152 | { | | 2197 | { |
2153 | const bool verbose = wapbl_verbose_commit >= 2; | | 2198 | const bool verbose = wapbl_verbose_commit >= 2; |
2154 | struct bintime start_time; | | 2199 | struct bintime start_time; |
2155 | int force = 1; | | 2200 | int force = 1; |
2156 | int error; | | 2201 | int error; |
2157 | | | 2202 | |
2158 | if (!wapbl_flush_disk_cache) { | | 2203 | if (!wapbl_flush_disk_cache) { |
2159 | return 0; | | 2204 | return 0; |
2160 | } | | 2205 | } |
2161 | if (verbose) { | | 2206 | if (verbose) { |
2162 | bintime(&start_time); | | 2207 | bintime(&start_time); |
2163 | } | | 2208 | } |
2164 | error = VOP_IOCTL(wl->wl_devvp, DIOCCACHESYNC, &force, | | 2209 | error = VOP_IOCTL(wl->wl_devvp, DIOCCACHESYNC, &force, |
2165 | FWRITE, FSCRED); | | 2210 | FWRITE, FSCRED); |
2166 | if (error) { | | 2211 | if (error) { |
2167 | WAPBL_PRINTF(WAPBL_PRINT_ERROR, | | 2212 | WAPBL_PRINTF(WAPBL_PRINT_ERROR, |
2168 | ("wapbl_cache_sync: DIOCCACHESYNC on dev 0x%jx " | | 2213 | ("wapbl_cache_sync: DIOCCACHESYNC on dev 0x%jx " |
2169 | "returned %d\n", (uintmax_t)wl->wl_devvp->v_rdev, error)); | | 2214 | "returned %d\n", (uintmax_t)wl->wl_devvp->v_rdev, error)); |
2170 | } | | 2215 | } |
2171 | if (verbose) { | | 2216 | if (verbose) { |
2172 | struct bintime d; | | 2217 | struct bintime d; |
2173 | struct timespec ts; | | 2218 | struct timespec ts; |
2174 | | | 2219 | |
2175 | bintime(&d); | | 2220 | bintime(&d); |
2176 | bintime_sub(&d, &start_time); | | 2221 | bintime_sub(&d, &start_time); |
2177 | bintime2timespec(&d, &ts); | | 2222 | bintime2timespec(&d, &ts); |
2178 | printf("wapbl_cache_sync: %s: dev 0x%jx %ju.%09lu\n", | | 2223 | printf("wapbl_cache_sync: %s: dev 0x%jx %ju.%09lu\n", |
2179 | msg, (uintmax_t)wl->wl_devvp->v_rdev, | | 2224 | msg, (uintmax_t)wl->wl_devvp->v_rdev, |
2180 | (uintmax_t)ts.tv_sec, ts.tv_nsec); | | 2225 | (uintmax_t)ts.tv_sec, ts.tv_nsec); |
2181 | } | | 2226 | } |
| | | 2227 | |
| | | 2228 | wl->wl_ev_cacheflush.ev_count++; |
| | | 2229 | |
2182 | return error; | | 2230 | return error; |
2183 | } | | 2231 | } |
2184 | | | 2232 | |
2185 | /* | | 2233 | /* |
2186 | * wapbl_write_commit(wl, head, tail) | | 2234 | * wapbl_write_commit(wl, head, tail) |
2187 | * | | 2235 | * |
2188 | * Issue a disk cache sync to wait for all pending writes to the | | 2236 | * Issue a disk cache sync to wait for all pending writes to the |
2189 | * log to complete, and then synchronously commit the current | | 2237 | * log to complete, and then synchronously commit the current |
2190 | * circular queue head and tail to the log, in the next of two | | 2238 | * circular queue head and tail to the log, in the next of two |
2191 | * locations for commit headers on disk. | | 2239 | * locations for commit headers on disk. |
2192 | * | | 2240 | * |
2193 | * Increment the generation number. If the generation number | | 2241 | * Increment the generation number. If the generation number |
2194 | * rolls over to zero, then a subsequent commit would appear to | | 2242 | * rolls over to zero, then a subsequent commit would appear to |
2195 | * have an older generation than this one -- in that case, issue a | | 2243 | * have an older generation than this one -- in that case, issue a |
2196 | * duplicate commit to avoid this. | | 2244 | * duplicate commit to avoid this. |
2197 | * | | 2245 | * |
2198 | * => Caller must have exclusive access to wl, either by holding | | 2246 | * => Caller must have exclusive access to wl, either by holding |
2199 | * wl->wl_rwlock for writer or by being wapbl_start before anyone | | 2247 | * wl->wl_rwlock for writer or by being wapbl_start before anyone |
2200 | * else has seen wl. | | 2248 | * else has seen wl. |
2201 | */ | | 2249 | */ |
2202 | static int | | 2250 | static int |
2203 | wapbl_write_commit(struct wapbl *wl, off_t head, off_t tail) | | 2251 | wapbl_write_commit(struct wapbl *wl, off_t head, off_t tail) |
2204 | { | | 2252 | { |
2205 | struct wapbl_wc_header *wc = wl->wl_wc_header; | | 2253 | struct wapbl_wc_header *wc = wl->wl_wc_header; |
2206 | struct timespec ts; | | 2254 | struct timespec ts; |
2207 | int error; | | 2255 | int error; |
2208 | daddr_t pbn; | | 2256 | daddr_t pbn; |
2209 | | | 2257 | |
2210 | error = wapbl_buffered_flush(wl); | | 2258 | error = wapbl_buffered_flush(wl); |
2211 | if (error) | | 2259 | if (error) |
2212 | return error; | | 2260 | return error; |
2213 | /* | | 2261 | /* |
2214 | * flush disk cache to ensure that blocks we've written are actually | | 2262 | * flush disk cache to ensure that blocks we've written are actually |
2215 | * written to the stable storage before the commit header. | | 2263 | * written to the stable storage before the commit header. |
2216 | * | | 2264 | * |
2217 | * XXX Calc checksum here, instead we do this for now | | 2265 | * XXX Calc checksum here, instead we do this for now |
2218 | */ | | 2266 | */ |
2219 | wapbl_cache_sync(wl, "1"); | | 2267 | wapbl_cache_sync(wl, "1"); |
2220 | | | 2268 | |
2221 | wc->wc_head = head; | | 2269 | wc->wc_head = head; |
2222 | wc->wc_tail = tail; | | 2270 | wc->wc_tail = tail; |
2223 | wc->wc_checksum = 0; | | 2271 | wc->wc_checksum = 0; |
2224 | wc->wc_version = 1; | | 2272 | wc->wc_version = 1; |
2225 | getnanotime(&ts); | | 2273 | getnanotime(&ts); |
2226 | wc->wc_time = ts.tv_sec; | | 2274 | wc->wc_time = ts.tv_sec; |
2227 | wc->wc_timensec = ts.tv_nsec; | | 2275 | wc->wc_timensec = ts.tv_nsec; |
2228 | | | 2276 | |
2229 | WAPBL_PRINTF(WAPBL_PRINT_WRITE, | | 2277 | WAPBL_PRINTF(WAPBL_PRINT_WRITE, |
2230 | ("wapbl_write_commit: head = %"PRIdMAX "tail = %"PRIdMAX"\n", | | 2278 | ("wapbl_write_commit: head = %"PRIdMAX "tail = %"PRIdMAX"\n", |
2231 | (intmax_t)head, (intmax_t)tail)); | | 2279 | (intmax_t)head, (intmax_t)tail)); |
2232 | | | 2280 | |
2233 | /* | | 2281 | /* |
2234 | * write the commit header. | | 2282 | * write the commit header. |
2235 | * | | 2283 | * |
2236 | * XXX if generation will rollover, then first zero | | 2284 | * XXX if generation will rollover, then first zero |
2237 | * over second commit header before trying to write both headers. | | 2285 | * over second commit header before trying to write both headers. |
2238 | */ | | 2286 | */ |
2239 | | | 2287 | |
2240 | pbn = wl->wl_logpbn + (wc->wc_generation % 2); | | 2288 | pbn = wl->wl_logpbn + (wc->wc_generation % 2); |
2241 | #ifdef _KERNEL | | 2289 | #ifdef _KERNEL |
2242 | pbn = btodb(pbn << wc->wc_log_dev_bshift); | | 2290 | pbn = btodb(pbn << wc->wc_log_dev_bshift); |
2243 | #endif | | 2291 | #endif |
2244 | error = wapbl_buffered_write(wc, wc->wc_len, wl, pbn); | | 2292 | error = wapbl_buffered_write(wc, wc->wc_len, wl, pbn); |
2245 | if (error) | | 2293 | if (error) |
2246 | return error; | | 2294 | return error; |
2247 | error = wapbl_buffered_flush(wl); | | 2295 | error = wapbl_buffered_flush(wl); |
2248 | if (error) | | 2296 | if (error) |
2249 | return error; | | 2297 | return error; |
2250 | | | 2298 | |
2251 | /* | | 2299 | /* |
2252 | * flush disk cache to ensure that the commit header is actually | | 2300 | * flush disk cache to ensure that the commit header is actually |
2253 | * written before meta data blocks. | | 2301 | * written before meta data blocks. |
2254 | */ | | 2302 | */ |
2255 | wapbl_cache_sync(wl, "2"); | | 2303 | wapbl_cache_sync(wl, "2"); |
2256 | | | 2304 | |
2257 | /* | | 2305 | /* |
2258 | * If the generation number was zero, write it out a second time. | | 2306 | * If the generation number was zero, write it out a second time. |
2259 | * This handles initialization and generation number rollover | | 2307 | * This handles initialization and generation number rollover |
2260 | */ | | 2308 | */ |
2261 | if (wc->wc_generation++ == 0) { | | 2309 | if (wc->wc_generation++ == 0) { |
2262 | error = wapbl_write_commit(wl, head, tail); | | 2310 | error = wapbl_write_commit(wl, head, tail); |
2263 | /* | | 2311 | /* |
2264 | * This panic should be able to be removed if we do the | | 2312 | * This panic should be able to be removed if we do the |
2265 | * zero'ing mentioned above, and we are certain to roll | | 2313 | * zero'ing mentioned above, and we are certain to roll |
2266 | * back generation number on failure. | | 2314 | * back generation number on failure. |
2267 | */ | | 2315 | */ |
2268 | if (error) | | 2316 | if (error) |
2269 | panic("wapbl_write_commit: error writing duplicate " | | 2317 | panic("wapbl_write_commit: error writing duplicate " |
2270 | "log header: %d", error); | | 2318 | "log header: %d", error); |
2271 | } | | 2319 | } |
| | | 2320 | |
| | | 2321 | wl->wl_ev_commit.ev_count++; |
| | | 2322 | |
2272 | return 0; | | 2323 | return 0; |
2273 | } | | 2324 | } |
2274 | | | 2325 | |
2275 | /* | | 2326 | /* |
2276 | * wapbl_write_blocks(wl, offp) | | 2327 | * wapbl_write_blocks(wl, offp) |
2277 | * | | 2328 | * |
2278 | * Write all pending physical blocks in the current transaction | | 2329 | * Write all pending physical blocks in the current transaction |
2279 | * from wapbl_add_buf to the log on disk, adding to the circular | | 2330 | * from wapbl_add_buf to the log on disk, adding to the circular |
2280 | * queue head at byte offset *offp, and returning the new head's | | 2331 | * queue head at byte offset *offp, and returning the new head's |
2281 | * byte offset in *offp. | | 2332 | * byte offset in *offp. |
2282 | */ | | 2333 | */ |
2283 | static int | | 2334 | static int |
2284 | wapbl_write_blocks(struct wapbl *wl, off_t *offp) | | 2335 | wapbl_write_blocks(struct wapbl *wl, off_t *offp) |
2285 | { | | 2336 | { |
2286 | struct wapbl_wc_blocklist *wc = | | 2337 | struct wapbl_wc_blocklist *wc = |
2287 | (struct wapbl_wc_blocklist *)wl->wl_wc_scratch; | | 2338 | (struct wapbl_wc_blocklist *)wl->wl_wc_scratch; |
2288 | int blocklen = 1<<wl->wl_log_dev_bshift; | | 2339 | int blocklen = 1<<wl->wl_log_dev_bshift; |
2289 | struct buf *bp; | | 2340 | struct buf *bp; |
2290 | off_t off = *offp; | | 2341 | off_t off = *offp; |
2291 | int error; | | 2342 | int error; |
2292 | size_t padding; | | 2343 | size_t padding; |
2293 | | | 2344 | |
2294 | KASSERT(rw_write_held(&wl->wl_rwlock)); | | 2345 | KASSERT(rw_write_held(&wl->wl_rwlock)); |
2295 | | | 2346 | |
2296 | bp = LIST_FIRST(&wl->wl_bufs); | | 2347 | bp = LIST_FIRST(&wl->wl_bufs); |
2297 | | | 2348 | |
2298 | while (bp) { | | 2349 | while (bp) { |
2299 | int cnt; | | 2350 | int cnt; |
2300 | struct buf *obp = bp; | | 2351 | struct buf *obp = bp; |
2301 | | | 2352 | |
2302 | KASSERT(bp->b_flags & B_LOCKED); | | 2353 | KASSERT(bp->b_flags & B_LOCKED); |
2303 | | | 2354 | |
2304 | wc->wc_type = WAPBL_WC_BLOCKS; | | 2355 | wc->wc_type = WAPBL_WC_BLOCKS; |
2305 | wc->wc_len = blocklen; | | 2356 | wc->wc_len = blocklen; |
2306 | wc->wc_blkcount = 0; | | 2357 | wc->wc_blkcount = 0; |
2307 | while (bp && (wc->wc_blkcount < wl->wl_brperjblock)) { | | 2358 | while (bp && (wc->wc_blkcount < wl->wl_brperjblock)) { |
2308 | /* | | 2359 | /* |
2309 | * Make sure all the physical block numbers are up to | | 2360 | * Make sure all the physical block numbers are up to |
2310 | * date. If this is not always true on a given | | 2361 | * date. If this is not always true on a given |
2311 | * filesystem, then VOP_BMAP must be called. We | | 2362 | * filesystem, then VOP_BMAP must be called. We |
2312 | * could call VOP_BMAP here, or else in the filesystem | | 2363 | * could call VOP_BMAP here, or else in the filesystem |
2313 | * specific flush callback, although neither of those | | 2364 | * specific flush callback, although neither of those |
2314 | * solutions allow us to take the vnode lock. If a | | 2365 | * solutions allow us to take the vnode lock. If a |
2315 | * filesystem requires that we must take the vnode lock | | 2366 | * filesystem requires that we must take the vnode lock |
2316 | * to call VOP_BMAP, then we can probably do it in | | 2367 | * to call VOP_BMAP, then we can probably do it in |
2317 | * bwrite when the vnode lock should already be held | | 2368 | * bwrite when the vnode lock should already be held |
2318 | * by the invoking code. | | 2369 | * by the invoking code. |
2319 | */ | | 2370 | */ |
2320 | KASSERT((bp->b_vp->v_type == VBLK) || | | 2371 | KASSERT((bp->b_vp->v_type == VBLK) || |
2321 | (bp->b_blkno != bp->b_lblkno)); | | 2372 | (bp->b_blkno != bp->b_lblkno)); |
2322 | KASSERT(bp->b_blkno > 0); | | 2373 | KASSERT(bp->b_blkno > 0); |
2323 | | | 2374 | |
2324 | wc->wc_blocks[wc->wc_blkcount].wc_daddr = bp->b_blkno; | | 2375 | wc->wc_blocks[wc->wc_blkcount].wc_daddr = bp->b_blkno; |
2325 | wc->wc_blocks[wc->wc_blkcount].wc_dlen = bp->b_bcount; | | 2376 | wc->wc_blocks[wc->wc_blkcount].wc_dlen = bp->b_bcount; |
2326 | wc->wc_len += bp->b_bcount; | | 2377 | wc->wc_len += bp->b_bcount; |
2327 | wc->wc_blkcount++; | | 2378 | wc->wc_blkcount++; |
2328 | bp = LIST_NEXT(bp, b_wapbllist); | | 2379 | bp = LIST_NEXT(bp, b_wapbllist); |
2329 | } | | 2380 | } |
2330 | if (wc->wc_len % blocklen != 0) { | | 2381 | if (wc->wc_len % blocklen != 0) { |
2331 | padding = blocklen - wc->wc_len % blocklen; | | 2382 | padding = blocklen - wc->wc_len % blocklen; |
2332 | wc->wc_len += padding; | | 2383 | wc->wc_len += padding; |
2333 | } else { | | 2384 | } else { |
2334 | padding = 0; | | 2385 | padding = 0; |
2335 | } | | 2386 | } |
2336 | | | 2387 | |
2337 | WAPBL_PRINTF(WAPBL_PRINT_WRITE, | | 2388 | WAPBL_PRINTF(WAPBL_PRINT_WRITE, |
2338 | ("wapbl_write_blocks: len = %u (padding %zu) off = %"PRIdMAX"\n", | | 2389 | ("wapbl_write_blocks: len = %u (padding %zu) off = %"PRIdMAX"\n", |
2339 | wc->wc_len, padding, (intmax_t)off)); | | 2390 | wc->wc_len, padding, (intmax_t)off)); |
2340 | | | 2391 | |
2341 | error = wapbl_circ_write(wl, wc, blocklen, &off); | | 2392 | error = wapbl_circ_write(wl, wc, blocklen, &off); |
2342 | if (error) | | 2393 | if (error) |
2343 | return error; | | 2394 | return error; |
2344 | bp = obp; | | 2395 | bp = obp; |
2345 | cnt = 0; | | 2396 | cnt = 0; |
2346 | while (bp && (cnt++ < wl->wl_brperjblock)) { | | 2397 | while (bp && (cnt++ < wl->wl_brperjblock)) { |
2347 | error = wapbl_circ_write(wl, bp->b_data, | | 2398 | error = wapbl_circ_write(wl, bp->b_data, |
2348 | bp->b_bcount, &off); | | 2399 | bp->b_bcount, &off); |
2349 | if (error) | | 2400 | if (error) |
2350 | return error; | | 2401 | return error; |
2351 | bp = LIST_NEXT(bp, b_wapbllist); | | 2402 | bp = LIST_NEXT(bp, b_wapbllist); |
2352 | } | | 2403 | } |
2353 | if (padding) { | | 2404 | if (padding) { |
2354 | void *zero; | | 2405 | void *zero; |
2355 | | | 2406 | |
2356 | zero = wapbl_alloc(padding); | | 2407 | zero = wapbl_alloc(padding); |
2357 | memset(zero, 0, padding); | | 2408 | memset(zero, 0, padding); |
2358 | error = wapbl_circ_write(wl, zero, padding, &off); | | 2409 | error = wapbl_circ_write(wl, zero, padding, &off); |
2359 | wapbl_free(zero, padding); | | 2410 | wapbl_free(zero, padding); |
2360 | if (error) | | 2411 | if (error) |
2361 | return error; | | 2412 | return error; |
2362 | } | | 2413 | } |
2363 | } | | 2414 | } |
2364 | *offp = off; | | 2415 | *offp = off; |
2365 | return 0; | | 2416 | return 0; |
2366 | } | | 2417 | } |
2367 | | | 2418 | |
2368 | /* | | 2419 | /* |
2369 | * wapbl_write_revocations(wl, offp) | | 2420 | * wapbl_write_revocations(wl, offp) |
2370 | * | | 2421 | * |
2371 | * Write all pending deallocations in the current transaction from | | 2422 | * Write all pending deallocations in the current transaction from |
2372 | * wapbl_register_deallocation to the log on disk, adding to the | | 2423 | * wapbl_register_deallocation to the log on disk, adding to the |
2373 | * circular queue's head at byte offset *offp, and returning the | | 2424 | * circular queue's head at byte offset *offp, and returning the |
2374 | * new head's byte offset in *offp. | | 2425 | * new head's byte offset in *offp. |
2375 | */ | | 2426 | */ |
2376 | static int | | 2427 | static int |
2377 | wapbl_write_revocations(struct wapbl *wl, off_t *offp) | | 2428 | wapbl_write_revocations(struct wapbl *wl, off_t *offp) |
2378 | { | | 2429 | { |
2379 | struct wapbl_wc_blocklist *wc = | | 2430 | struct wapbl_wc_blocklist *wc = |
2380 | (struct wapbl_wc_blocklist *)wl->wl_wc_scratch; | | 2431 | (struct wapbl_wc_blocklist *)wl->wl_wc_scratch; |
2381 | struct wapbl_dealloc *wd, *lwd; | | 2432 | struct wapbl_dealloc *wd, *lwd; |
2382 | int blocklen = 1<<wl->wl_log_dev_bshift; | | 2433 | int blocklen = 1<<wl->wl_log_dev_bshift; |
2383 | off_t off = *offp; | | 2434 | off_t off = *offp; |
2384 | int error; | | 2435 | int error; |
2385 | | | 2436 | |
2386 | if (wl->wl_dealloccnt == 0) | | 2437 | if (wl->wl_dealloccnt == 0) |
2387 | return 0; | | 2438 | return 0; |
2388 | | | 2439 | |
2389 | while ((wd = TAILQ_FIRST(&wl->wl_dealloclist)) != NULL) { | | 2440 | while ((wd = TAILQ_FIRST(&wl->wl_dealloclist)) != NULL) { |
2390 | wc->wc_type = WAPBL_WC_REVOCATIONS; | | 2441 | wc->wc_type = WAPBL_WC_REVOCATIONS; |
2391 | wc->wc_len = blocklen; | | 2442 | wc->wc_len = blocklen; |
2392 | wc->wc_blkcount = 0; | | 2443 | wc->wc_blkcount = 0; |
2393 | while (wd && (wc->wc_blkcount < wl->wl_brperjblock)) { | | 2444 | while (wd && (wc->wc_blkcount < wl->wl_brperjblock)) { |
2394 | wc->wc_blocks[wc->wc_blkcount].wc_daddr = | | 2445 | wc->wc_blocks[wc->wc_blkcount].wc_daddr = |
2395 | wd->wd_blkno; | | 2446 | wd->wd_blkno; |
2396 | wc->wc_blocks[wc->wc_blkcount].wc_dlen = | | 2447 | wc->wc_blocks[wc->wc_blkcount].wc_dlen = |
2397 | wd->wd_len; | | 2448 | wd->wd_len; |
2398 | wc->wc_blkcount++; | | 2449 | wc->wc_blkcount++; |
2399 | | | 2450 | |
2400 | wd = TAILQ_NEXT(wd, wd_entries); | | 2451 | wd = TAILQ_NEXT(wd, wd_entries); |
2401 | } | | 2452 | } |
2402 | WAPBL_PRINTF(WAPBL_PRINT_WRITE, | | 2453 | WAPBL_PRINTF(WAPBL_PRINT_WRITE, |
2403 | ("wapbl_write_revocations: len = %u off = %"PRIdMAX"\n", | | 2454 | ("wapbl_write_revocations: len = %u off = %"PRIdMAX"\n", |
2404 | wc->wc_len, (intmax_t)off)); | | 2455 | wc->wc_len, (intmax_t)off)); |
2405 | error = wapbl_circ_write(wl, wc, blocklen, &off); | | 2456 | error = wapbl_circ_write(wl, wc, blocklen, &off); |
2406 | if (error) | | 2457 | if (error) |
2407 | return error; | | 2458 | return error; |
2408 | | | 2459 | |
2409 | /* free all successfully written deallocs */ | | 2460 | /* free all successfully written deallocs */ |
2410 | lwd = wd; | | 2461 | lwd = wd; |
2411 | while ((wd = TAILQ_FIRST(&wl->wl_dealloclist)) != NULL) { | | 2462 | while ((wd = TAILQ_FIRST(&wl->wl_dealloclist)) != NULL) { |
2412 | if (wd == lwd) | | 2463 | if (wd == lwd) |
2413 | break; | | 2464 | break; |
2414 | wapbl_deallocation_free(wl, wd, true); | | 2465 | wapbl_deallocation_free(wl, wd, true); |
2415 | } | | 2466 | } |
2416 | } | | 2467 | } |
2417 | *offp = off; | | 2468 | *offp = off; |
2418 | return 0; | | 2469 | return 0; |
2419 | } | | 2470 | } |
2420 | | | 2471 | |
2421 | /* | | 2472 | /* |
2422 | * wapbl_write_inodes(wl, offp) | | 2473 | * wapbl_write_inodes(wl, offp) |
2423 | * | | 2474 | * |
2424 | * Write all pending inode allocations in the current transaction | | 2475 | * Write all pending inode allocations in the current transaction |
2425 | * from wapbl_register_inode to the log on disk, adding to the | | 2476 | * from wapbl_register_inode to the log on disk, adding to the |
2426 | * circular queue's head at byte offset *offp and returning the | | 2477 | * circular queue's head at byte offset *offp and returning the |
2427 | * new head's byte offset in *offp. | | 2478 | * new head's byte offset in *offp. |
2428 | */ | | 2479 | */ |
2429 | static int | | 2480 | static int |
2430 | wapbl_write_inodes(struct wapbl *wl, off_t *offp) | | 2481 | wapbl_write_inodes(struct wapbl *wl, off_t *offp) |
2431 | { | | 2482 | { |
2432 | struct wapbl_wc_inodelist *wc = | | 2483 | struct wapbl_wc_inodelist *wc = |
2433 | (struct wapbl_wc_inodelist *)wl->wl_wc_scratch; | | 2484 | (struct wapbl_wc_inodelist *)wl->wl_wc_scratch; |
2434 | int i; | | 2485 | int i; |
2435 | int blocklen = 1 << wl->wl_log_dev_bshift; | | 2486 | int blocklen = 1 << wl->wl_log_dev_bshift; |
2436 | off_t off = *offp; | | 2487 | off_t off = *offp; |
2437 | int error; | | 2488 | int error; |
2438 | | | 2489 | |
2439 | struct wapbl_ino_head *wih; | | 2490 | struct wapbl_ino_head *wih; |
2440 | struct wapbl_ino *wi; | | 2491 | struct wapbl_ino *wi; |
2441 | int iph; | | 2492 | int iph; |
2442 | | | 2493 | |
2443 | iph = (blocklen - offsetof(struct wapbl_wc_inodelist, wc_inodes)) / | | 2494 | iph = (blocklen - offsetof(struct wapbl_wc_inodelist, wc_inodes)) / |
2444 | sizeof(((struct wapbl_wc_inodelist *)0)->wc_inodes[0]); | | 2495 | sizeof(((struct wapbl_wc_inodelist *)0)->wc_inodes[0]); |
2445 | | | 2496 | |
2446 | i = 0; | | 2497 | i = 0; |
2447 | wih = &wl->wl_inohash[0]; | | 2498 | wih = &wl->wl_inohash[0]; |
2448 | wi = 0; | | 2499 | wi = 0; |
2449 | do { | | 2500 | do { |
2450 | wc->wc_type = WAPBL_WC_INODES; | | 2501 | wc->wc_type = WAPBL_WC_INODES; |
2451 | wc->wc_len = blocklen; | | 2502 | wc->wc_len = blocklen; |
2452 | wc->wc_inocnt = 0; | | 2503 | wc->wc_inocnt = 0; |
2453 | wc->wc_clear = (i == 0); | | 2504 | wc->wc_clear = (i == 0); |
2454 | while ((i < wl->wl_inohashcnt) && (wc->wc_inocnt < iph)) { | | 2505 | while ((i < wl->wl_inohashcnt) && (wc->wc_inocnt < iph)) { |
2455 | while (!wi) { | | 2506 | while (!wi) { |
2456 | KASSERT((wih - &wl->wl_inohash[0]) | | 2507 | KASSERT((wih - &wl->wl_inohash[0]) |
2457 | <= wl->wl_inohashmask); | | 2508 | <= wl->wl_inohashmask); |
2458 | wi = LIST_FIRST(wih++); | | 2509 | wi = LIST_FIRST(wih++); |
2459 | } | | 2510 | } |
2460 | wc->wc_inodes[wc->wc_inocnt].wc_inumber = wi->wi_ino; | | 2511 | wc->wc_inodes[wc->wc_inocnt].wc_inumber = wi->wi_ino; |
2461 | wc->wc_inodes[wc->wc_inocnt].wc_imode = wi->wi_mode; | | 2512 | wc->wc_inodes[wc->wc_inocnt].wc_imode = wi->wi_mode; |
2462 | wc->wc_inocnt++; | | 2513 | wc->wc_inocnt++; |
2463 | i++; | | 2514 | i++; |
2464 | wi = LIST_NEXT(wi, wi_hash); | | 2515 | wi = LIST_NEXT(wi, wi_hash); |
2465 | } | | 2516 | } |
2466 | WAPBL_PRINTF(WAPBL_PRINT_WRITE, | | 2517 | WAPBL_PRINTF(WAPBL_PRINT_WRITE, |
2467 | ("wapbl_write_inodes: len = %u off = %"PRIdMAX"\n", | | 2518 | ("wapbl_write_inodes: len = %u off = %"PRIdMAX"\n", |
2468 | wc->wc_len, (intmax_t)off)); | | 2519 | wc->wc_len, (intmax_t)off)); |
2469 | error = wapbl_circ_write(wl, wc, blocklen, &off); | | 2520 | error = wapbl_circ_write(wl, wc, blocklen, &off); |
2470 | if (error) | | 2521 | if (error) |
2471 | return error; | | 2522 | return error; |
2472 | } while (i < wl->wl_inohashcnt); | | 2523 | } while (i < wl->wl_inohashcnt); |
2473 | | | 2524 | |
2474 | *offp = off; | | 2525 | *offp = off; |
2475 | return 0; | | 2526 | return 0; |
2476 | } | | 2527 | } |
2477 | | | 2528 | |
2478 | #endif /* _KERNEL */ | | 2529 | #endif /* _KERNEL */ |
2479 | | | 2530 | |
2480 | /****************************************************************/ | | 2531 | /****************************************************************/ |
2481 | | | 2532 | |
2482 | struct wapbl_blk { | | 2533 | struct wapbl_blk { |
2483 | LIST_ENTRY(wapbl_blk) wb_hash; | | 2534 | LIST_ENTRY(wapbl_blk) wb_hash; |
2484 | daddr_t wb_blk; | | 2535 | daddr_t wb_blk; |
2485 | off_t wb_off; /* Offset of this block in the log */ | | 2536 | off_t wb_off; /* Offset of this block in the log */ |
2486 | }; | | 2537 | }; |
2487 | #define WAPBL_BLKPOOL_MIN 83 | | 2538 | #define WAPBL_BLKPOOL_MIN 83 |
2488 | | | 2539 | |
2489 | static void | | 2540 | static void |
2490 | wapbl_blkhash_init(struct wapbl_replay *wr, u_int size) | | 2541 | wapbl_blkhash_init(struct wapbl_replay *wr, u_int size) |
2491 | { | | 2542 | { |
2492 | if (size < WAPBL_BLKPOOL_MIN) | | 2543 | if (size < WAPBL_BLKPOOL_MIN) |
2493 | size = WAPBL_BLKPOOL_MIN; | | 2544 | size = WAPBL_BLKPOOL_MIN; |
2494 | KASSERT(wr->wr_blkhash == 0); | | 2545 | KASSERT(wr->wr_blkhash == 0); |
2495 | #ifdef _KERNEL | | 2546 | #ifdef _KERNEL |
2496 | wr->wr_blkhash = hashinit(size, HASH_LIST, true, &wr->wr_blkhashmask); | | 2547 | wr->wr_blkhash = hashinit(size, HASH_LIST, true, &wr->wr_blkhashmask); |
2497 | #else /* ! _KERNEL */ | | 2548 | #else /* ! _KERNEL */ |
2498 | /* Manually implement hashinit */ | | 2549 | /* Manually implement hashinit */ |
2499 | { | | 2550 | { |
2500 | unsigned long i, hashsize; | | 2551 | unsigned long i, hashsize; |
2501 | for (hashsize = 1; hashsize < size; hashsize <<= 1) | | 2552 | for (hashsize = 1; hashsize < size; hashsize <<= 1) |
2502 | continue; | | 2553 | continue; |
2503 | wr->wr_blkhash = wapbl_alloc(hashsize * sizeof(*wr->wr_blkhash)); | | 2554 | wr->wr_blkhash = wapbl_alloc(hashsize * sizeof(*wr->wr_blkhash)); |
2504 | for (i = 0; i < hashsize; i++) | | 2555 | for (i = 0; i < hashsize; i++) |
2505 | LIST_INIT(&wr->wr_blkhash[i]); | | 2556 | LIST_INIT(&wr->wr_blkhash[i]); |
2506 | wr->wr_blkhashmask = hashsize - 1; | | 2557 | wr->wr_blkhashmask = hashsize - 1; |
2507 | } | | 2558 | } |
2508 | #endif /* ! _KERNEL */ | | 2559 | #endif /* ! _KERNEL */ |
2509 | } | | 2560 | } |
2510 | | | 2561 | |
2511 | static void | | 2562 | static void |
2512 | wapbl_blkhash_free(struct wapbl_replay *wr) | | 2563 | wapbl_blkhash_free(struct wapbl_replay *wr) |
2513 | { | | 2564 | { |
2514 | KASSERT(wr->wr_blkhashcnt == 0); | | 2565 | KASSERT(wr->wr_blkhashcnt == 0); |
2515 | #ifdef _KERNEL | | 2566 | #ifdef _KERNEL |
2516 | hashdone(wr->wr_blkhash, HASH_LIST, wr->wr_blkhashmask); | | 2567 | hashdone(wr->wr_blkhash, HASH_LIST, wr->wr_blkhashmask); |
2517 | #else /* ! _KERNEL */ | | 2568 | #else /* ! _KERNEL */ |
2518 | wapbl_free(wr->wr_blkhash, | | 2569 | wapbl_free(wr->wr_blkhash, |
2519 | (wr->wr_blkhashmask + 1) * sizeof(*wr->wr_blkhash)); | | 2570 | (wr->wr_blkhashmask + 1) * sizeof(*wr->wr_blkhash)); |
2520 | #endif /* ! _KERNEL */ | | 2571 | #endif /* ! _KERNEL */ |
2521 | } | | 2572 | } |
2522 | | | 2573 | |
2523 | static struct wapbl_blk * | | 2574 | static struct wapbl_blk * |
2524 | wapbl_blkhash_get(struct wapbl_replay *wr, daddr_t blk) | | 2575 | wapbl_blkhash_get(struct wapbl_replay *wr, daddr_t blk) |
2525 | { | | 2576 | { |
2526 | struct wapbl_blk_head *wbh; | | 2577 | struct wapbl_blk_head *wbh; |
2527 | struct wapbl_blk *wb; | | 2578 | struct wapbl_blk *wb; |
2528 | wbh = &wr->wr_blkhash[blk & wr->wr_blkhashmask]; | | 2579 | wbh = &wr->wr_blkhash[blk & wr->wr_blkhashmask]; |
2529 | LIST_FOREACH(wb, wbh, wb_hash) { | | 2580 | LIST_FOREACH(wb, wbh, wb_hash) { |
2530 | if (blk == wb->wb_blk) | | 2581 | if (blk == wb->wb_blk) |
2531 | return wb; | | 2582 | return wb; |
2532 | } | | 2583 | } |
2533 | return 0; | | 2584 | return 0; |
2534 | } | | 2585 | } |
2535 | | | 2586 | |
2536 | static void | | 2587 | static void |
2537 | wapbl_blkhash_ins(struct wapbl_replay *wr, daddr_t blk, off_t off) | | 2588 | wapbl_blkhash_ins(struct wapbl_replay *wr, daddr_t blk, off_t off) |
2538 | { | | 2589 | { |
2539 | struct wapbl_blk_head *wbh; | | 2590 | struct wapbl_blk_head *wbh; |
2540 | struct wapbl_blk *wb; | | 2591 | struct wapbl_blk *wb; |
2541 | wb = wapbl_blkhash_get(wr, blk); | | 2592 | wb = wapbl_blkhash_get(wr, blk); |
2542 | if (wb) { | | 2593 | if (wb) { |
2543 | KASSERT(wb->wb_blk == blk); | | 2594 | KASSERT(wb->wb_blk == blk); |
2544 | wb->wb_off = off; | | 2595 | wb->wb_off = off; |
2545 | } else { | | 2596 | } else { |
2546 | wb = wapbl_alloc(sizeof(*wb)); | | 2597 | wb = wapbl_alloc(sizeof(*wb)); |
2547 | wb->wb_blk = blk; | | 2598 | wb->wb_blk = blk; |
2548 | wb->wb_off = off; | | 2599 | wb->wb_off = off; |
2549 | wbh = &wr->wr_blkhash[blk & wr->wr_blkhashmask]; | | 2600 | wbh = &wr->wr_blkhash[blk & wr->wr_blkhashmask]; |
2550 | LIST_INSERT_HEAD(wbh, wb, wb_hash); | | 2601 | LIST_INSERT_HEAD(wbh, wb, wb_hash); |
2551 | wr->wr_blkhashcnt++; | | 2602 | wr->wr_blkhashcnt++; |
2552 | } | | 2603 | } |
2553 | } | | 2604 | } |
2554 | | | 2605 | |
2555 | static void | | 2606 | static void |
2556 | wapbl_blkhash_rem(struct wapbl_replay *wr, daddr_t blk) | | 2607 | wapbl_blkhash_rem(struct wapbl_replay *wr, daddr_t blk) |
2557 | { | | 2608 | { |
2558 | struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk); | | 2609 | struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk); |
2559 | if (wb) { | | 2610 | if (wb) { |
2560 | KASSERT(wr->wr_blkhashcnt > 0); | | 2611 | KASSERT(wr->wr_blkhashcnt > 0); |
2561 | wr->wr_blkhashcnt--; | | 2612 | wr->wr_blkhashcnt--; |
2562 | LIST_REMOVE(wb, wb_hash); | | 2613 | LIST_REMOVE(wb, wb_hash); |
2563 | wapbl_free(wb, sizeof(*wb)); | | 2614 | wapbl_free(wb, sizeof(*wb)); |
2564 | } | | 2615 | } |
2565 | } | | 2616 | } |
2566 | | | 2617 | |
2567 | static void | | 2618 | static void |
2568 | wapbl_blkhash_clear(struct wapbl_replay *wr) | | 2619 | wapbl_blkhash_clear(struct wapbl_replay *wr) |
2569 | { | | 2620 | { |
2570 | unsigned long i; | | 2621 | unsigned long i; |
2571 | for (i = 0; i <= wr->wr_blkhashmask; i++) { | | 2622 | for (i = 0; i <= wr->wr_blkhashmask; i++) { |
2572 | struct wapbl_blk *wb; | | 2623 | struct wapbl_blk *wb; |
2573 | | | 2624 | |
2574 | while ((wb = LIST_FIRST(&wr->wr_blkhash[i]))) { | | 2625 | while ((wb = LIST_FIRST(&wr->wr_blkhash[i]))) { |
2575 | KASSERT(wr->wr_blkhashcnt > 0); | | 2626 | KASSERT(wr->wr_blkhashcnt > 0); |
2576 | wr->wr_blkhashcnt--; | | 2627 | wr->wr_blkhashcnt--; |
2577 | LIST_REMOVE(wb, wb_hash); | | 2628 | LIST_REMOVE(wb, wb_hash); |
2578 | wapbl_free(wb, sizeof(*wb)); | | 2629 | wapbl_free(wb, sizeof(*wb)); |
2579 | } | | 2630 | } |
2580 | } | | 2631 | } |
2581 | KASSERT(wr->wr_blkhashcnt == 0); | | 2632 | KASSERT(wr->wr_blkhashcnt == 0); |
2582 | } | | 2633 | } |
2583 | | | 2634 | |
2584 | /****************************************************************/ | | 2635 | /****************************************************************/ |
2585 | | | 2636 | |
2586 | /* | | 2637 | /* |
2587 | * wapbl_circ_read(wr, data, len, offp) | | 2638 | * wapbl_circ_read(wr, data, len, offp) |
2588 | * | | 2639 | * |
2589 | * Read len bytes into data from the circular queue of wr, | | 2640 | * Read len bytes into data from the circular queue of wr, |
2590 | * starting at the linear byte offset *offp, and returning the new | | 2641 | * starting at the linear byte offset *offp, and returning the new |
2591 | * linear byte offset in *offp. | | 2642 | * linear byte offset in *offp. |
2592 | * | | 2643 | * |
2593 | * If the starting linear byte offset precedes wr->wr_circ_off, | | 2644 | * If the starting linear byte offset precedes wr->wr_circ_off, |
2594 | * the read instead begins at wr->wr_circ_off. XXX WTF? This | | 2645 | * the read instead begins at wr->wr_circ_off. XXX WTF? This |
2595 | * should be a KASSERT, not a conditional. | | 2646 | * should be a KASSERT, not a conditional. |
2596 | */ | | 2647 | */ |
2597 | static int | | 2648 | static int |
2598 | wapbl_circ_read(struct wapbl_replay *wr, void *data, size_t len, off_t *offp) | | 2649 | wapbl_circ_read(struct wapbl_replay *wr, void *data, size_t len, off_t *offp) |
2599 | { | | 2650 | { |
2600 | size_t slen; | | 2651 | size_t slen; |
2601 | off_t off = *offp; | | 2652 | off_t off = *offp; |
2602 | int error; | | 2653 | int error; |
2603 | daddr_t pbn; | | 2654 | daddr_t pbn; |
2604 | | | 2655 | |
2605 | KASSERT(((len >> wr->wr_log_dev_bshift) << | | 2656 | KASSERT(((len >> wr->wr_log_dev_bshift) << |
2606 | wr->wr_log_dev_bshift) == len); | | 2657 | wr->wr_log_dev_bshift) == len); |
2607 | | | 2658 | |
2608 | if (off < wr->wr_circ_off) | | 2659 | if (off < wr->wr_circ_off) |
2609 | off = wr->wr_circ_off; | | 2660 | off = wr->wr_circ_off; |
2610 | slen = wr->wr_circ_off + wr->wr_circ_size - off; | | 2661 | slen = wr->wr_circ_off + wr->wr_circ_size - off; |
2611 | if (slen < len) { | | 2662 | if (slen < len) { |
2612 | pbn = wr->wr_logpbn + (off >> wr->wr_log_dev_bshift); | | 2663 | pbn = wr->wr_logpbn + (off >> wr->wr_log_dev_bshift); |
2613 | #ifdef _KERNEL | | 2664 | #ifdef _KERNEL |
2614 | pbn = btodb(pbn << wr->wr_log_dev_bshift); | | 2665 | pbn = btodb(pbn << wr->wr_log_dev_bshift); |
2615 | #endif | | 2666 | #endif |
2616 | error = wapbl_read(data, slen, wr->wr_devvp, pbn); | | 2667 | error = wapbl_read(data, slen, wr->wr_devvp, pbn); |
2617 | if (error) | | 2668 | if (error) |
2618 | return error; | | 2669 | return error; |
2619 | data = (uint8_t *)data + slen; | | 2670 | data = (uint8_t *)data + slen; |
2620 | len -= slen; | | 2671 | len -= slen; |
2621 | off = wr->wr_circ_off; | | 2672 | off = wr->wr_circ_off; |
2622 | } | | 2673 | } |
2623 | pbn = wr->wr_logpbn + (off >> wr->wr_log_dev_bshift); | | 2674 | pbn = wr->wr_logpbn + (off >> wr->wr_log_dev_bshift); |
2624 | #ifdef _KERNEL | | 2675 | #ifdef _KERNEL |
2625 | pbn = btodb(pbn << wr->wr_log_dev_bshift); | | 2676 | pbn = btodb(pbn << wr->wr_log_dev_bshift); |
2626 | #endif | | 2677 | #endif |
2627 | error = wapbl_read(data, len, wr->wr_devvp, pbn); | | 2678 | error = wapbl_read(data, len, wr->wr_devvp, pbn); |
2628 | if (error) | | 2679 | if (error) |
2629 | return error; | | 2680 | return error; |
2630 | off += len; | | 2681 | off += len; |
2631 | if (off >= wr->wr_circ_off + wr->wr_circ_size) | | 2682 | if (off >= wr->wr_circ_off + wr->wr_circ_size) |
2632 | off = wr->wr_circ_off; | | 2683 | off = wr->wr_circ_off; |
2633 | *offp = off; | | 2684 | *offp = off; |
2634 | return 0; | | 2685 | return 0; |
2635 | } | | 2686 | } |
2636 | | | 2687 | |
2637 | /* | | 2688 | /* |
2638 | * wapbl_circ_advance(wr, len, offp) | | 2689 | * wapbl_circ_advance(wr, len, offp) |
2639 | * | | 2690 | * |
2640 | * Compute the linear byte offset of the circular queue of wr that | | 2691 | * Compute the linear byte offset of the circular queue of wr that |
2641 | * is len bytes past *offp, and store it in *offp. | | 2692 | * is len bytes past *offp, and store it in *offp. |
2642 | * | | 2693 | * |
2643 | * This is as if wapbl_circ_read, but without actually reading | | 2694 | * This is as if wapbl_circ_read, but without actually reading |
2644 | * anything. | | 2695 | * anything. |
2645 | * | | 2696 | * |
2646 | * If the starting linear byte offset precedes wr->wr_circ_off, it | | 2697 | * If the starting linear byte offset precedes wr->wr_circ_off, it |
2647 | * is taken to be wr->wr_circ_off instead. XXX WTF? This should | | 2698 | * is taken to be wr->wr_circ_off instead. XXX WTF? This should |
2648 | * be a KASSERT, not a conditional. | | 2699 | * be a KASSERT, not a conditional. |
2649 | */ | | 2700 | */ |
2650 | static void | | 2701 | static void |
2651 | wapbl_circ_advance(struct wapbl_replay *wr, size_t len, off_t *offp) | | 2702 | wapbl_circ_advance(struct wapbl_replay *wr, size_t len, off_t *offp) |
2652 | { | | 2703 | { |
2653 | size_t slen; | | 2704 | size_t slen; |
2654 | off_t off = *offp; | | 2705 | off_t off = *offp; |
2655 | | | 2706 | |
2656 | KASSERT(((len >> wr->wr_log_dev_bshift) << | | 2707 | KASSERT(((len >> wr->wr_log_dev_bshift) << |
2657 | wr->wr_log_dev_bshift) == len); | | 2708 | wr->wr_log_dev_bshift) == len); |
2658 | | | 2709 | |
2659 | if (off < wr->wr_circ_off) | | 2710 | if (off < wr->wr_circ_off) |
2660 | off = wr->wr_circ_off; | | 2711 | off = wr->wr_circ_off; |
2661 | slen = wr->wr_circ_off + wr->wr_circ_size - off; | | 2712 | slen = wr->wr_circ_off + wr->wr_circ_size - off; |
2662 | if (slen < len) { | | 2713 | if (slen < len) { |
2663 | len -= slen; | | 2714 | len -= slen; |
2664 | off = wr->wr_circ_off; | | 2715 | off = wr->wr_circ_off; |
2665 | } | | 2716 | } |
2666 | off += len; | | 2717 | off += len; |
2667 | if (off >= wr->wr_circ_off + wr->wr_circ_size) | | 2718 | if (off >= wr->wr_circ_off + wr->wr_circ_size) |
2668 | off = wr->wr_circ_off; | | 2719 | off = wr->wr_circ_off; |
2669 | *offp = off; | | 2720 | *offp = off; |
2670 | } | | 2721 | } |
2671 | | | 2722 | |
2672 | /****************************************************************/ | | 2723 | /****************************************************************/ |
2673 | | | 2724 | |
2674 | int | | 2725 | int |
2675 | wapbl_replay_start(struct wapbl_replay **wrp, struct vnode *vp, | | 2726 | wapbl_replay_start(struct wapbl_replay **wrp, struct vnode *vp, |
2676 | daddr_t off, size_t count, size_t blksize) | | 2727 | daddr_t off, size_t count, size_t blksize) |
2677 | { | | 2728 | { |
2678 | struct wapbl_replay *wr; | | 2729 | struct wapbl_replay *wr; |
2679 | int error; | | 2730 | int error; |
2680 | struct vnode *devvp; | | 2731 | struct vnode *devvp; |
2681 | daddr_t logpbn; | | 2732 | daddr_t logpbn; |
2682 | uint8_t *scratch; | | 2733 | uint8_t *scratch; |
2683 | struct wapbl_wc_header *wch; | | 2734 | struct wapbl_wc_header *wch; |
2684 | struct wapbl_wc_header *wch2; | | 2735 | struct wapbl_wc_header *wch2; |
2685 | /* Use this until we read the actual log header */ | | 2736 | /* Use this until we read the actual log header */ |
2686 | int log_dev_bshift = ilog2(blksize); | | 2737 | int log_dev_bshift = ilog2(blksize); |
2687 | size_t used; | | 2738 | size_t used; |
2688 | daddr_t pbn; | | 2739 | daddr_t pbn; |
2689 | | | 2740 | |
2690 | WAPBL_PRINTF(WAPBL_PRINT_REPLAY, | | 2741 | WAPBL_PRINTF(WAPBL_PRINT_REPLAY, |
2691 | ("wapbl_replay_start: vp=%p off=%"PRId64 " count=%zu blksize=%zu\n", | | 2742 | ("wapbl_replay_start: vp=%p off=%"PRId64 " count=%zu blksize=%zu\n", |
2692 | vp, off, count, blksize)); | | 2743 | vp, off, count, blksize)); |
2693 | | | 2744 | |
2694 | if (off < 0) | | 2745 | if (off < 0) |
2695 | return EINVAL; | | 2746 | return EINVAL; |
2696 | | | 2747 | |
2697 | if (blksize < DEV_BSIZE) | | 2748 | if (blksize < DEV_BSIZE) |
2698 | return EINVAL; | | 2749 | return EINVAL; |
2699 | if (blksize % DEV_BSIZE) | | 2750 | if (blksize % DEV_BSIZE) |
2700 | return EINVAL; | | 2751 | return EINVAL; |
2701 | | | 2752 | |
2702 | #ifdef _KERNEL | | 2753 | #ifdef _KERNEL |
2703 | #if 0 | | 2754 | #if 0 |
2704 | /* XXX vp->v_size isn't reliably set for VBLK devices, | | 2755 | /* XXX vp->v_size isn't reliably set for VBLK devices, |
2705 | * especially root. However, we might still want to verify | | 2756 | * especially root. However, we might still want to verify |
2706 | * that the full load is readable */ | | 2757 | * that the full load is readable */ |
2707 | if ((off + count) * blksize > vp->v_size) | | 2758 | if ((off + count) * blksize > vp->v_size) |
2708 | return EINVAL; | | 2759 | return EINVAL; |
2709 | #endif | | 2760 | #endif |
2710 | if ((error = VOP_BMAP(vp, off, &devvp, &logpbn, 0)) != 0) { | | 2761 | if ((error = VOP_BMAP(vp, off, &devvp, &logpbn, 0)) != 0) { |
2711 | return error; | | 2762 | return error; |
2712 | } | | 2763 | } |
2713 | #else /* ! _KERNEL */ | | 2764 | #else /* ! _KERNEL */ |
2714 | devvp = vp; | | 2765 | devvp = vp; |
2715 | logpbn = off; | | 2766 | logpbn = off; |
2716 | #endif /* ! _KERNEL */ | | 2767 | #endif /* ! _KERNEL */ |
2717 | | | 2768 | |
2718 | scratch = wapbl_alloc(MAXBSIZE); | | 2769 | scratch = wapbl_alloc(MAXBSIZE); |
2719 | | | 2770 | |
2720 | pbn = logpbn; | | 2771 | pbn = logpbn; |
2721 | #ifdef _KERNEL | | 2772 | #ifdef _KERNEL |
2722 | pbn = btodb(pbn << log_dev_bshift); | | 2773 | pbn = btodb(pbn << log_dev_bshift); |
2723 | #endif | | 2774 | #endif |
2724 | error = wapbl_read(scratch, 2<<log_dev_bshift, devvp, pbn); | | 2775 | error = wapbl_read(scratch, 2<<log_dev_bshift, devvp, pbn); |
2725 | if (error) | | 2776 | if (error) |
2726 | goto errout; | | 2777 | goto errout; |
2727 | | | 2778 | |
2728 | wch = (struct wapbl_wc_header *)scratch; | | 2779 | wch = (struct wapbl_wc_header *)scratch; |
2729 | wch2 = | | 2780 | wch2 = |
2730 | (struct wapbl_wc_header *)(scratch + (1<<log_dev_bshift)); | | 2781 | (struct wapbl_wc_header *)(scratch + (1<<log_dev_bshift)); |
2731 | /* XXX verify checksums and magic numbers */ | | 2782 | /* XXX verify checksums and magic numbers */ |
2732 | if (wch->wc_type != WAPBL_WC_HEADER) { | | 2783 | if (wch->wc_type != WAPBL_WC_HEADER) { |
2733 | printf("Unrecognized wapbl magic: 0x%08x\n", wch->wc_type); | | 2784 | printf("Unrecognized wapbl magic: 0x%08x\n", wch->wc_type); |
2734 | error = EFTYPE; | | 2785 | error = EFTYPE; |
2735 | goto errout; | | 2786 | goto errout; |
2736 | } | | 2787 | } |
2737 | | | 2788 | |
2738 | if (wch2->wc_generation > wch->wc_generation) | | 2789 | if (wch2->wc_generation > wch->wc_generation) |
2739 | wch = wch2; | | 2790 | wch = wch2; |
2740 | | | 2791 | |
2741 | wr = wapbl_calloc(1, sizeof(*wr)); | | 2792 | wr = wapbl_calloc(1, sizeof(*wr)); |
2742 | | | 2793 | |
2743 | wr->wr_logvp = vp; | | 2794 | wr->wr_logvp = vp; |
2744 | wr->wr_devvp = devvp; | | 2795 | wr->wr_devvp = devvp; |
2745 | wr->wr_logpbn = logpbn; | | 2796 | wr->wr_logpbn = logpbn; |
2746 | | | 2797 | |
2747 | wr->wr_scratch = scratch; | | 2798 | wr->wr_scratch = scratch; |
2748 | | | 2799 | |
2749 | wr->wr_log_dev_bshift = wch->wc_log_dev_bshift; | | 2800 | wr->wr_log_dev_bshift = wch->wc_log_dev_bshift; |
2750 | wr->wr_fs_dev_bshift = wch->wc_fs_dev_bshift; | | 2801 | wr->wr_fs_dev_bshift = wch->wc_fs_dev_bshift; |
2751 | wr->wr_circ_off = wch->wc_circ_off; | | 2802 | wr->wr_circ_off = wch->wc_circ_off; |
2752 | wr->wr_circ_size = wch->wc_circ_size; | | 2803 | wr->wr_circ_size = wch->wc_circ_size; |
2753 | wr->wr_generation = wch->wc_generation; | | 2804 | wr->wr_generation = wch->wc_generation; |
2754 | | | 2805 | |
2755 | used = wapbl_space_used(wch->wc_circ_size, wch->wc_head, wch->wc_tail); | | 2806 | used = wapbl_space_used(wch->wc_circ_size, wch->wc_head, wch->wc_tail); |
2756 | | | 2807 | |
2757 | WAPBL_PRINTF(WAPBL_PRINT_REPLAY, | | 2808 | WAPBL_PRINTF(WAPBL_PRINT_REPLAY, |
2758 | ("wapbl_replay: head=%"PRId64" tail=%"PRId64" off=%"PRId64 | | 2809 | ("wapbl_replay: head=%"PRId64" tail=%"PRId64" off=%"PRId64 |
2759 | " len=%"PRId64" used=%zu\n", | | 2810 | " len=%"PRId64" used=%zu\n", |
2760 | wch->wc_head, wch->wc_tail, wch->wc_circ_off, | | 2811 | wch->wc_head, wch->wc_tail, wch->wc_circ_off, |
2761 | wch->wc_circ_size, used)); | | 2812 | wch->wc_circ_size, used)); |
2762 | | | 2813 | |
2763 | wapbl_blkhash_init(wr, (used >> wch->wc_fs_dev_bshift)); | | 2814 | wapbl_blkhash_init(wr, (used >> wch->wc_fs_dev_bshift)); |
2764 | | | 2815 | |
2765 | error = wapbl_replay_process(wr, wch->wc_head, wch->wc_tail); | | 2816 | error = wapbl_replay_process(wr, wch->wc_head, wch->wc_tail); |
2766 | if (error) { | | 2817 | if (error) { |
2767 | wapbl_replay_stop(wr); | | 2818 | wapbl_replay_stop(wr); |
2768 | wapbl_replay_free(wr); | | 2819 | wapbl_replay_free(wr); |
2769 | return error; | | 2820 | return error; |
2770 | } | | 2821 | } |
2771 | | | 2822 | |
2772 | *wrp = wr; | | 2823 | *wrp = wr; |
2773 | return 0; | | 2824 | return 0; |
2774 | | | 2825 | |
2775 | errout: | | 2826 | errout: |
2776 | wapbl_free(scratch, MAXBSIZE); | | 2827 | wapbl_free(scratch, MAXBSIZE); |
2777 | return error; | | 2828 | return error; |
2778 | } | | 2829 | } |
2779 | | | 2830 | |
2780 | void | | 2831 | void |
2781 | wapbl_replay_stop(struct wapbl_replay *wr) | | 2832 | wapbl_replay_stop(struct wapbl_replay *wr) |
2782 | { | | 2833 | { |
2783 | | | 2834 | |
2784 | if (!wapbl_replay_isopen(wr)) | | 2835 | if (!wapbl_replay_isopen(wr)) |
2785 | return; | | 2836 | return; |
2786 | | | 2837 | |
2787 | WAPBL_PRINTF(WAPBL_PRINT_REPLAY, ("wapbl_replay_stop called\n")); | | 2838 | WAPBL_PRINTF(WAPBL_PRINT_REPLAY, ("wapbl_replay_stop called\n")); |
2788 | | | 2839 | |
2789 | wapbl_free(wr->wr_scratch, MAXBSIZE); | | 2840 | wapbl_free(wr->wr_scratch, MAXBSIZE); |
2790 | wr->wr_scratch = NULL; | | 2841 | wr->wr_scratch = NULL; |
2791 | | | 2842 | |
2792 | wr->wr_logvp = NULL; | | 2843 | wr->wr_logvp = NULL; |
2793 | | | 2844 | |
2794 | wapbl_blkhash_clear(wr); | | 2845 | wapbl_blkhash_clear(wr); |
2795 | wapbl_blkhash_free(wr); | | 2846 | wapbl_blkhash_free(wr); |
2796 | } | | 2847 | } |
2797 | | | 2848 | |
2798 | void | | 2849 | void |
2799 | wapbl_replay_free(struct wapbl_replay *wr) | | 2850 | wapbl_replay_free(struct wapbl_replay *wr) |
2800 | { | | 2851 | { |
2801 | | | 2852 | |
2802 | KDASSERT(!wapbl_replay_isopen(wr)); | | 2853 | KDASSERT(!wapbl_replay_isopen(wr)); |
2803 | | | 2854 | |
2804 | if (wr->wr_inodes) | | 2855 | if (wr->wr_inodes) |
2805 | wapbl_free(wr->wr_inodes, | | 2856 | wapbl_free(wr->wr_inodes, |
2806 | wr->wr_inodescnt * sizeof(wr->wr_inodes[0])); | | 2857 | wr->wr_inodescnt * sizeof(wr->wr_inodes[0])); |
2807 | wapbl_free(wr, sizeof(*wr)); | | 2858 | wapbl_free(wr, sizeof(*wr)); |
2808 | } | | 2859 | } |
2809 | | | 2860 | |
2810 | #ifdef _KERNEL | | 2861 | #ifdef _KERNEL |
2811 | int | | 2862 | int |
2812 | wapbl_replay_isopen1(struct wapbl_replay *wr) | | 2863 | wapbl_replay_isopen1(struct wapbl_replay *wr) |
2813 | { | | 2864 | { |
2814 | | | 2865 | |
2815 | return wapbl_replay_isopen(wr); | | 2866 | return wapbl_replay_isopen(wr); |
2816 | } | | 2867 | } |
2817 | #endif | | 2868 | #endif |
2818 | | | 2869 | |
2819 | /* | | 2870 | /* |
2820 | * calculate the disk address for the i'th block in the wc_blockblist | | 2871 | * calculate the disk address for the i'th block in the wc_blockblist |
2821 | * offset by j blocks of size blen. | | 2872 | * offset by j blocks of size blen. |
2822 | * | | 2873 | * |
2823 | * wc_daddr is always a kernel disk address in DEV_BSIZE units that | | 2874 | * wc_daddr is always a kernel disk address in DEV_BSIZE units that |
2824 | * was written to the journal. | | 2875 | * was written to the journal. |
2825 | * | | 2876 | * |
2826 | * The kernel needs that address plus the offset in DEV_BSIZE units. | | 2877 | * The kernel needs that address plus the offset in DEV_BSIZE units. |
2827 | * | | 2878 | * |
2828 | * Userland needs that address plus the offset in blen units. | | 2879 | * Userland needs that address plus the offset in blen units. |
2829 | * | | 2880 | * |
2830 | */ | | 2881 | */ |
2831 | static daddr_t | | 2882 | static daddr_t |
2832 | wapbl_block_daddr(struct wapbl_wc_blocklist *wc, int i, int j, int blen) | | 2883 | wapbl_block_daddr(struct wapbl_wc_blocklist *wc, int i, int j, int blen) |
2833 | { | | 2884 | { |
2834 | daddr_t pbn; | | 2885 | daddr_t pbn; |
2835 | | | 2886 | |
2836 | #ifdef _KERNEL | | 2887 | #ifdef _KERNEL |
2837 | pbn = wc->wc_blocks[i].wc_daddr + btodb(j * blen); | | 2888 | pbn = wc->wc_blocks[i].wc_daddr + btodb(j * blen); |
2838 | #else | | 2889 | #else |
2839 | pbn = dbtob(wc->wc_blocks[i].wc_daddr) / blen + j; | | 2890 | pbn = dbtob(wc->wc_blocks[i].wc_daddr) / blen + j; |
2840 | #endif | | 2891 | #endif |
2841 | | | 2892 | |
2842 | return pbn; | | 2893 | return pbn; |
2843 | } | | 2894 | } |
2844 | | | 2895 | |
2845 | static void | | 2896 | static void |
2846 | wapbl_replay_process_blocks(struct wapbl_replay *wr, off_t *offp) | | 2897 | wapbl_replay_process_blocks(struct wapbl_replay *wr, off_t *offp) |
2847 | { | | 2898 | { |
2848 | struct wapbl_wc_blocklist *wc = | | 2899 | struct wapbl_wc_blocklist *wc = |
2849 | (struct wapbl_wc_blocklist *)wr->wr_scratch; | | 2900 | (struct wapbl_wc_blocklist *)wr->wr_scratch; |
2850 | int fsblklen = 1 << wr->wr_fs_dev_bshift; | | 2901 | int fsblklen = 1 << wr->wr_fs_dev_bshift; |
2851 | int i, j, n; | | 2902 | int i, j, n; |
2852 | | | 2903 | |
2853 | for (i = 0; i < wc->wc_blkcount; i++) { | | 2904 | for (i = 0; i < wc->wc_blkcount; i++) { |
2854 | /* | | 2905 | /* |
2855 | * Enter each physical block into the hashtable independently. | | 2906 | * Enter each physical block into the hashtable independently. |
2856 | */ | | 2907 | */ |
2857 | n = wc->wc_blocks[i].wc_dlen >> wr->wr_fs_dev_bshift; | | 2908 | n = wc->wc_blocks[i].wc_dlen >> wr->wr_fs_dev_bshift; |
2858 | for (j = 0; j < n; j++) { | | 2909 | for (j = 0; j < n; j++) { |
2859 | wapbl_blkhash_ins(wr, wapbl_block_daddr(wc, i, j, fsblklen), | | 2910 | wapbl_blkhash_ins(wr, wapbl_block_daddr(wc, i, j, fsblklen), |
2860 | *offp); | | 2911 | *offp); |
2861 | wapbl_circ_advance(wr, fsblklen, offp); | | 2912 | wapbl_circ_advance(wr, fsblklen, offp); |
2862 | } | | 2913 | } |
2863 | } | | 2914 | } |
2864 | } | | 2915 | } |
2865 | | | 2916 | |
2866 | static void | | 2917 | static void |
2867 | wapbl_replay_process_revocations(struct wapbl_replay *wr) | | 2918 | wapbl_replay_process_revocations(struct wapbl_replay *wr) |
2868 | { | | 2919 | { |
2869 | struct wapbl_wc_blocklist *wc = | | 2920 | struct wapbl_wc_blocklist *wc = |
2870 | (struct wapbl_wc_blocklist *)wr->wr_scratch; | | 2921 | (struct wapbl_wc_blocklist *)wr->wr_scratch; |
2871 | int fsblklen = 1 << wr->wr_fs_dev_bshift; | | 2922 | int fsblklen = 1 << wr->wr_fs_dev_bshift; |
2872 | int i, j, n; | | 2923 | int i, j, n; |
2873 | | | 2924 | |
2874 | for (i = 0; i < wc->wc_blkcount; i++) { | | 2925 | for (i = 0; i < wc->wc_blkcount; i++) { |
2875 | /* | | 2926 | /* |
2876 | * Remove any blocks found from the hashtable. | | 2927 | * Remove any blocks found from the hashtable. |
2877 | */ | | 2928 | */ |
2878 | n = wc->wc_blocks[i].wc_dlen >> wr->wr_fs_dev_bshift; | | 2929 | n = wc->wc_blocks[i].wc_dlen >> wr->wr_fs_dev_bshift; |
2879 | for (j = 0; j < n; j++) | | 2930 | for (j = 0; j < n; j++) |
2880 | wapbl_blkhash_rem(wr, wapbl_block_daddr(wc, i, j, fsblklen)); | | 2931 | wapbl_blkhash_rem(wr, wapbl_block_daddr(wc, i, j, fsblklen)); |
2881 | } | | 2932 | } |
2882 | } | | 2933 | } |
2883 | | | 2934 | |
2884 | static void | | 2935 | static void |
2885 | wapbl_replay_process_inodes(struct wapbl_replay *wr, off_t oldoff, off_t newoff) | | 2936 | wapbl_replay_process_inodes(struct wapbl_replay *wr, off_t oldoff, off_t newoff) |
2886 | { | | 2937 | { |
2887 | struct wapbl_wc_inodelist *wc = | | 2938 | struct wapbl_wc_inodelist *wc = |
2888 | (struct wapbl_wc_inodelist *)wr->wr_scratch; | | 2939 | (struct wapbl_wc_inodelist *)wr->wr_scratch; |
2889 | void *new_inodes; | | 2940 | void *new_inodes; |
2890 | const size_t oldsize = wr->wr_inodescnt * sizeof(wr->wr_inodes[0]); | | 2941 | const size_t oldsize = wr->wr_inodescnt * sizeof(wr->wr_inodes[0]); |
2891 | | | 2942 | |
2892 | KASSERT(sizeof(wr->wr_inodes[0]) == sizeof(wc->wc_inodes[0])); | | 2943 | KASSERT(sizeof(wr->wr_inodes[0]) == sizeof(wc->wc_inodes[0])); |
2893 | | | 2944 | |
2894 | /* | | 2945 | /* |
2895 | * Keep track of where we found this so location won't be | | 2946 | * Keep track of where we found this so location won't be |
2896 | * overwritten. | | 2947 | * overwritten. |
2897 | */ | | 2948 | */ |
2898 | if (wc->wc_clear) { | | 2949 | if (wc->wc_clear) { |
2899 | wr->wr_inodestail = oldoff; | | 2950 | wr->wr_inodestail = oldoff; |
2900 | wr->wr_inodescnt = 0; | | 2951 | wr->wr_inodescnt = 0; |
2901 | if (wr->wr_inodes != NULL) { | | 2952 | if (wr->wr_inodes != NULL) { |
2902 | wapbl_free(wr->wr_inodes, oldsize); | | 2953 | wapbl_free(wr->wr_inodes, oldsize); |
2903 | wr->wr_inodes = NULL; | | 2954 | wr->wr_inodes = NULL; |
2904 | } | | 2955 | } |
2905 | } | | 2956 | } |
2906 | wr->wr_inodeshead = newoff; | | 2957 | wr->wr_inodeshead = newoff; |
2907 | if (wc->wc_inocnt == 0) | | 2958 | if (wc->wc_inocnt == 0) |
2908 | return; | | 2959 | return; |
2909 | | | 2960 | |
2910 | new_inodes = wapbl_alloc((wr->wr_inodescnt + wc->wc_inocnt) * | | 2961 | new_inodes = wapbl_alloc((wr->wr_inodescnt + wc->wc_inocnt) * |
2911 | sizeof(wr->wr_inodes[0])); | | 2962 | sizeof(wr->wr_inodes[0])); |
2912 | if (wr->wr_inodes != NULL) { | | 2963 | if (wr->wr_inodes != NULL) { |
2913 | memcpy(new_inodes, wr->wr_inodes, oldsize); | | 2964 | memcpy(new_inodes, wr->wr_inodes, oldsize); |
2914 | wapbl_free(wr->wr_inodes, oldsize); | | 2965 | wapbl_free(wr->wr_inodes, oldsize); |
2915 | } | | 2966 | } |
2916 | wr->wr_inodes = new_inodes; | | 2967 | wr->wr_inodes = new_inodes; |
2917 | memcpy(&wr->wr_inodes[wr->wr_inodescnt], wc->wc_inodes, | | 2968 | memcpy(&wr->wr_inodes[wr->wr_inodescnt], wc->wc_inodes, |
2918 | wc->wc_inocnt * sizeof(wr->wr_inodes[0])); | | 2969 | wc->wc_inocnt * sizeof(wr->wr_inodes[0])); |
2919 | wr->wr_inodescnt += wc->wc_inocnt; | | 2970 | wr->wr_inodescnt += wc->wc_inocnt; |
2920 | } | | 2971 | } |
2921 | | | 2972 | |
2922 | static int | | 2973 | static int |
2923 | wapbl_replay_process(struct wapbl_replay *wr, off_t head, off_t tail) | | 2974 | wapbl_replay_process(struct wapbl_replay *wr, off_t head, off_t tail) |
2924 | { | | 2975 | { |
2925 | off_t off; | | 2976 | off_t off; |
2926 | int error; | | 2977 | int error; |
2927 | | | 2978 | |
2928 | int logblklen = 1 << wr->wr_log_dev_bshift; | | 2979 | int logblklen = 1 << wr->wr_log_dev_bshift; |
2929 | | | 2980 | |
2930 | wapbl_blkhash_clear(wr); | | 2981 | wapbl_blkhash_clear(wr); |
2931 | | | 2982 | |
2932 | off = tail; | | 2983 | off = tail; |
2933 | while (off != head) { | | 2984 | while (off != head) { |
2934 | struct wapbl_wc_null *wcn; | | 2985 | struct wapbl_wc_null *wcn; |
2935 | off_t saveoff = off; | | 2986 | off_t saveoff = off; |
2936 | error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off); | | 2987 | error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off); |
2937 | if (error) | | 2988 | if (error) |
2938 | goto errout; | | 2989 | goto errout; |
2939 | wcn = (struct wapbl_wc_null *)wr->wr_scratch; | | 2990 | wcn = (struct wapbl_wc_null *)wr->wr_scratch; |
2940 | switch (wcn->wc_type) { | | 2991 | switch (wcn->wc_type) { |
2941 | case WAPBL_WC_BLOCKS: | | 2992 | case WAPBL_WC_BLOCKS: |
2942 | wapbl_replay_process_blocks(wr, &off); | | 2993 | wapbl_replay_process_blocks(wr, &off); |
2943 | break; | | 2994 | break; |
2944 | | | 2995 | |
2945 | case WAPBL_WC_REVOCATIONS: | | 2996 | case WAPBL_WC_REVOCATIONS: |
2946 | wapbl_replay_process_revocations(wr); | | 2997 | wapbl_replay_process_revocations(wr); |
2947 | break; | | 2998 | break; |
2948 | | | 2999 | |
2949 | case WAPBL_WC_INODES: | | 3000 | case WAPBL_WC_INODES: |
2950 | wapbl_replay_process_inodes(wr, saveoff, off); | | 3001 | wapbl_replay_process_inodes(wr, saveoff, off); |
2951 | break; | | 3002 | break; |
2952 | | | 3003 | |
2953 | default: | | 3004 | default: |
2954 | printf("Unrecognized wapbl type: 0x%08x\n", | | 3005 | printf("Unrecognized wapbl type: 0x%08x\n", |
2955 | wcn->wc_type); | | 3006 | wcn->wc_type); |
2956 | error = EFTYPE; | | 3007 | error = EFTYPE; |
2957 | goto errout; | | 3008 | goto errout; |
2958 | } | | 3009 | } |
2959 | wapbl_circ_advance(wr, wcn->wc_len, &saveoff); | | 3010 | wapbl_circ_advance(wr, wcn->wc_len, &saveoff); |
2960 | if (off != saveoff) { | | 3011 | if (off != saveoff) { |
2961 | printf("wapbl_replay: corrupted records\n"); | | 3012 | printf("wapbl_replay: corrupted records\n"); |
2962 | error = EFTYPE; | | 3013 | error = EFTYPE; |
2963 | goto errout; | | 3014 | goto errout; |
2964 | } | | 3015 | } |
2965 | } | | 3016 | } |
2966 | return 0; | | 3017 | return 0; |
2967 | | | 3018 | |
2968 | errout: | | 3019 | errout: |
2969 | wapbl_blkhash_clear(wr); | | 3020 | wapbl_blkhash_clear(wr); |
2970 | return error; | | 3021 | return error; |
2971 | } | | 3022 | } |
2972 | | | 3023 | |
2973 | #if 0 | | 3024 | #if 0 |
2974 | int | | 3025 | int |
2975 | wapbl_replay_verify(struct wapbl_replay *wr, struct vnode *fsdevvp) | | 3026 | wapbl_replay_verify(struct wapbl_replay *wr, struct vnode *fsdevvp) |
2976 | { | | 3027 | { |
2977 | off_t off; | | 3028 | off_t off; |
2978 | int mismatchcnt = 0; | | 3029 | int mismatchcnt = 0; |
2979 | int logblklen = 1 << wr->wr_log_dev_bshift; | | 3030 | int logblklen = 1 << wr->wr_log_dev_bshift; |
2980 | int fsblklen = 1 << wr->wr_fs_dev_bshift; | | 3031 | int fsblklen = 1 << wr->wr_fs_dev_bshift; |
2981 | void *scratch1 = wapbl_alloc(MAXBSIZE); | | 3032 | void *scratch1 = wapbl_alloc(MAXBSIZE); |
2982 | void *scratch2 = wapbl_alloc(MAXBSIZE); | | 3033 | void *scratch2 = wapbl_alloc(MAXBSIZE); |
2983 | int error = 0; | | 3034 | int error = 0; |
2984 | | | 3035 | |
2985 | KDASSERT(wapbl_replay_isopen(wr)); | | 3036 | KDASSERT(wapbl_replay_isopen(wr)); |
2986 | | | 3037 | |
2987 | off = wch->wc_tail; | | 3038 | off = wch->wc_tail; |
2988 | while (off != wch->wc_head) { | | 3039 | while (off != wch->wc_head) { |
2989 | struct wapbl_wc_null *wcn; | | 3040 | struct wapbl_wc_null *wcn; |
2990 | #ifdef DEBUG | | 3041 | #ifdef DEBUG |
2991 | off_t saveoff = off; | | 3042 | off_t saveoff = off; |
2992 | #endif | | 3043 | #endif |
2993 | error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off); | | 3044 | error = wapbl_circ_read(wr, wr->wr_scratch, logblklen, &off); |
2994 | if (error) | | 3045 | if (error) |
2995 | goto out; | | 3046 | goto out; |
2996 | wcn = (struct wapbl_wc_null *)wr->wr_scratch; | | 3047 | wcn = (struct wapbl_wc_null *)wr->wr_scratch; |
2997 | switch (wcn->wc_type) { | | 3048 | switch (wcn->wc_type) { |
2998 | case WAPBL_WC_BLOCKS: | | 3049 | case WAPBL_WC_BLOCKS: |
2999 | { | | 3050 | { |
3000 | struct wapbl_wc_blocklist *wc = | | 3051 | struct wapbl_wc_blocklist *wc = |
3001 | (struct wapbl_wc_blocklist *)wr->wr_scratch; | | 3052 | (struct wapbl_wc_blocklist *)wr->wr_scratch; |
3002 | int i; | | 3053 | int i; |
3003 | for (i = 0; i < wc->wc_blkcount; i++) { | | 3054 | for (i = 0; i < wc->wc_blkcount; i++) { |
3004 | int foundcnt = 0; | | 3055 | int foundcnt = 0; |
3005 | int dirtycnt = 0; | | 3056 | int dirtycnt = 0; |
3006 | int j, n; | | 3057 | int j, n; |
3007 | /* | | 3058 | /* |
3008 | * Check each physical block into the | | 3059 | * Check each physical block into the |
3009 | * hashtable independently | | 3060 | * hashtable independently |
3010 | */ | | 3061 | */ |
3011 | n = wc->wc_blocks[i].wc_dlen >> | | 3062 | n = wc->wc_blocks[i].wc_dlen >> |
3012 | wch->wc_fs_dev_bshift; | | 3063 | wch->wc_fs_dev_bshift; |
3013 | for (j = 0; j < n; j++) { | | 3064 | for (j = 0; j < n; j++) { |
3014 | struct wapbl_blk *wb = | | 3065 | struct wapbl_blk *wb = |
3015 | wapbl_blkhash_get(wr, | | 3066 | wapbl_blkhash_get(wr, |
3016 | wapbl_block_daddr(wc, i, j, fsblklen)); | | 3067 | wapbl_block_daddr(wc, i, j, fsblklen)); |
3017 | if (wb && (wb->wb_off == off)) { | | 3068 | if (wb && (wb->wb_off == off)) { |
3018 | foundcnt++; | | 3069 | foundcnt++; |
3019 | error = | | 3070 | error = |
3020 | wapbl_circ_read(wr, | | 3071 | wapbl_circ_read(wr, |
3021 | scratch1, fsblklen, | | 3072 | scratch1, fsblklen, |
3022 | &off); | | 3073 | &off); |
3023 | if (error) | | 3074 | if (error) |
3024 | goto out; | | 3075 | goto out; |
3025 | error = | | 3076 | error = |
3026 | wapbl_read(scratch2, | | 3077 | wapbl_read(scratch2, |
3027 | fsblklen, fsdevvp, | | 3078 | fsblklen, fsdevvp, |
3028 | wb->wb_blk); | | 3079 | wb->wb_blk); |
3029 | if (error) | | 3080 | if (error) |
3030 | goto out; | | 3081 | goto out; |
3031 | if (memcmp(scratch1, | | 3082 | if (memcmp(scratch1, |
3032 | scratch2, | | 3083 | scratch2, |
3033 | fsblklen)) { | | 3084 | fsblklen)) { |
3034 | printf( | | 3085 | printf( |
3035 | "wapbl_verify: mismatch block %"PRId64" at off %"PRIdMAX"\n", | | 3086 | "wapbl_verify: mismatch block %"PRId64" at off %"PRIdMAX"\n", |
3036 | wb->wb_blk, (intmax_t)off); | | 3087 | wb->wb_blk, (intmax_t)off); |
3037 | dirtycnt++; | | 3088 | dirtycnt++; |
3038 | mismatchcnt++; | | 3089 | mismatchcnt++; |
3039 | } | | 3090 | } |
3040 | } else { | | 3091 | } else { |
3041 | wapbl_circ_advance(wr, | | 3092 | wapbl_circ_advance(wr, |
3042 | fsblklen, &off); | | 3093 | fsblklen, &off); |
3043 | } | | 3094 | } |
3044 | } | | 3095 | } |
3045 | #if 0 | | 3096 | #if 0 |
3046 | /* | | 3097 | /* |
3047 | * If all of the blocks in an entry | | 3098 | * If all of the blocks in an entry |
3048 | * are clean, then remove all of its | | 3099 | * are clean, then remove all of its |
3049 | * blocks from the hashtable since they | | 3100 | * blocks from the hashtable since they |
3050 | * never will need replay. | | 3101 | * never will need replay. |
3051 | */ | | 3102 | */ |
3052 | if ((foundcnt != 0) && | | 3103 | if ((foundcnt != 0) && |
3053 | (dirtycnt == 0)) { | | 3104 | (dirtycnt == 0)) { |
3054 | off = saveoff; | | 3105 | off = saveoff; |
3055 | wapbl_circ_advance(wr, | | 3106 | wapbl_circ_advance(wr, |
3056 | logblklen, &off); | | 3107 | logblklen, &off); |
3057 | for (j = 0; j < n; j++) { | | 3108 | for (j = 0; j < n; j++) { |
3058 | struct wapbl_blk *wb = | | 3109 | struct wapbl_blk *wb = |
3059 | wapbl_blkhash_get(wr, | | 3110 | wapbl_blkhash_get(wr, |
3060 | wapbl_block_daddr(wc, i, j, fsblklen)); | | 3111 | wapbl_block_daddr(wc, i, j, fsblklen)); |
3061 | if (wb && | | 3112 | if (wb && |
3062 | (wb->wb_off == off)) { | | 3113 | (wb->wb_off == off)) { |
3063 | wapbl_blkhash_rem(wr, wb->wb_blk); | | 3114 | wapbl_blkhash_rem(wr, wb->wb_blk); |
3064 | } | | 3115 | } |
3065 | wapbl_circ_advance(wr, | | 3116 | wapbl_circ_advance(wr, |
3066 | fsblklen, &off); | | 3117 | fsblklen, &off); |
3067 | } | | 3118 | } |
3068 | } | | 3119 | } |
3069 | #endif | | 3120 | #endif |
3070 | } | | 3121 | } |
3071 | } | | 3122 | } |
3072 | break; | | 3123 | break; |
3073 | case WAPBL_WC_REVOCATIONS: | | 3124 | case WAPBL_WC_REVOCATIONS: |
3074 | case WAPBL_WC_INODES: | | 3125 | case WAPBL_WC_INODES: |
3075 | break; | | 3126 | break; |
3076 | default: | | 3127 | default: |
3077 | KASSERT(0); | | 3128 | KASSERT(0); |
3078 | } | | 3129 | } |
3079 | #ifdef DEBUG | | 3130 | #ifdef DEBUG |
3080 | wapbl_circ_advance(wr, wcn->wc_len, &saveoff); | | 3131 | wapbl_circ_advance(wr, wcn->wc_len, &saveoff); |
3081 | KASSERT(off == saveoff); | | 3132 | KASSERT(off == saveoff); |
3082 | #endif | | 3133 | #endif |
3083 | } | | 3134 | } |
3084 | out: | | 3135 | out: |
3085 | wapbl_free(scratch1, MAXBSIZE); | | 3136 | wapbl_free(scratch1, MAXBSIZE); |
3086 | wapbl_free(scratch2, MAXBSIZE); | | 3137 | wapbl_free(scratch2, MAXBSIZE); |
3087 | if (!error && mismatchcnt) | | 3138 | if (!error && mismatchcnt) |
3088 | error = EFTYPE; | | 3139 | error = EFTYPE; |
3089 | return error; | | 3140 | return error; |
3090 | } | | 3141 | } |
3091 | #endif | | 3142 | #endif |
3092 | | | 3143 | |
3093 | int | | 3144 | int |
3094 | wapbl_replay_write(struct wapbl_replay *wr, struct vnode *fsdevvp) | | 3145 | wapbl_replay_write(struct wapbl_replay *wr, struct vnode *fsdevvp) |
3095 | { | | 3146 | { |
3096 | struct wapbl_blk *wb; | | 3147 | struct wapbl_blk *wb; |
3097 | size_t i; | | 3148 | size_t i; |
3098 | off_t off; | | 3149 | off_t off; |
3099 | void *scratch; | | 3150 | void *scratch; |
3100 | int error = 0; | | 3151 | int error = 0; |
3101 | int fsblklen = 1 << wr->wr_fs_dev_bshift; | | 3152 | int fsblklen = 1 << wr->wr_fs_dev_bshift; |
3102 | | | 3153 | |
3103 | KDASSERT(wapbl_replay_isopen(wr)); | | 3154 | KDASSERT(wapbl_replay_isopen(wr)); |
3104 | | | 3155 | |
3105 | scratch = wapbl_alloc(MAXBSIZE); | | 3156 | scratch = wapbl_alloc(MAXBSIZE); |
3106 | | | 3157 | |
3107 | for (i = 0; i <= wr->wr_blkhashmask; ++i) { | | 3158 | for (i = 0; i <= wr->wr_blkhashmask; ++i) { |
3108 | LIST_FOREACH(wb, &wr->wr_blkhash[i], wb_hash) { | | 3159 | LIST_FOREACH(wb, &wr->wr_blkhash[i], wb_hash) { |
3109 | off = wb->wb_off; | | 3160 | off = wb->wb_off; |
3110 | error = wapbl_circ_read(wr, scratch, fsblklen, &off); | | 3161 | error = wapbl_circ_read(wr, scratch, fsblklen, &off); |
3111 | if (error) | | 3162 | if (error) |
3112 | break; | | 3163 | break; |
3113 | error = wapbl_write(scratch, fsblklen, fsdevvp, | | 3164 | error = wapbl_write(scratch, fsblklen, fsdevvp, |
3114 | wb->wb_blk); | | 3165 | wb->wb_blk); |
3115 | if (error) | | 3166 | if (error) |
3116 | break; | | 3167 | break; |
3117 | } | | 3168 | } |
3118 | } | | 3169 | } |
3119 | | | 3170 | |
3120 | wapbl_free(scratch, MAXBSIZE); | | 3171 | wapbl_free(scratch, MAXBSIZE); |
3121 | return error; | | 3172 | return error; |
3122 | } | | 3173 | } |
3123 | | | 3174 | |
3124 | int | | 3175 | int |
3125 | wapbl_replay_can_read(struct wapbl_replay *wr, daddr_t blk, long len) | | 3176 | wapbl_replay_can_read(struct wapbl_replay *wr, daddr_t blk, long len) |
3126 | { | | 3177 | { |
3127 | int fsblklen = 1 << wr->wr_fs_dev_bshift; | | 3178 | int fsblklen = 1 << wr->wr_fs_dev_bshift; |
3128 | | | 3179 | |
3129 | KDASSERT(wapbl_replay_isopen(wr)); | | 3180 | KDASSERT(wapbl_replay_isopen(wr)); |
3130 | KASSERT((len % fsblklen) == 0); | | 3181 | KASSERT((len % fsblklen) == 0); |
3131 | | | 3182 | |
3132 | while (len != 0) { | | 3183 | while (len != 0) { |
3133 | struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk); | | 3184 | struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk); |
3134 | if (wb) | | 3185 | if (wb) |
3135 | return 1; | | 3186 | return 1; |
3136 | len -= fsblklen; | | 3187 | len -= fsblklen; |
3137 | } | | 3188 | } |
3138 | return 0; | | 3189 | return 0; |
3139 | } | | 3190 | } |
3140 | | | 3191 | |
3141 | int | | 3192 | int |
3142 | wapbl_replay_read(struct wapbl_replay *wr, void *data, daddr_t blk, long len) | | 3193 | wapbl_replay_read(struct wapbl_replay *wr, void *data, daddr_t blk, long len) |
3143 | { | | 3194 | { |
3144 | int fsblklen = 1 << wr->wr_fs_dev_bshift; | | 3195 | int fsblklen = 1 << wr->wr_fs_dev_bshift; |
3145 | | | 3196 | |
3146 | KDASSERT(wapbl_replay_isopen(wr)); | | 3197 | KDASSERT(wapbl_replay_isopen(wr)); |
3147 | | | 3198 | |
3148 | KASSERT((len % fsblklen) == 0); | | 3199 | KASSERT((len % fsblklen) == 0); |
3149 | | | 3200 | |
3150 | while (len != 0) { | | 3201 | while (len != 0) { |
3151 | struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk); | | 3202 | struct wapbl_blk *wb = wapbl_blkhash_get(wr, blk); |
3152 | if (wb) { | | 3203 | if (wb) { |
3153 | off_t off = wb->wb_off; | | 3204 | off_t off = wb->wb_off; |
3154 | int error; | | 3205 | int error; |
3155 | error = wapbl_circ_read(wr, data, fsblklen, &off); | | 3206 | error = wapbl_circ_read(wr, data, fsblklen, &off); |
3156 | if (error) | | 3207 | if (error) |
3157 | return error; | | 3208 | return error; |
3158 | } | | 3209 | } |
3159 | data = (uint8_t *)data + fsblklen; | | 3210 | data = (uint8_t *)data + fsblklen; |
3160 | len -= fsblklen; | | 3211 | len -= fsblklen; |
3161 | blk++; | | 3212 | blk++; |
3162 | } | | 3213 | } |
3163 | return 0; | | 3214 | return 0; |
3164 | } | | 3215 | } |
3165 | | | 3216 | |
3166 | #ifdef _KERNEL | | 3217 | #ifdef _KERNEL |
3167 | | | 3218 | |
3168 | MODULE(MODULE_CLASS_VFS, wapbl, NULL); | | 3219 | MODULE(MODULE_CLASS_VFS, wapbl, NULL); |
3169 | | | 3220 | |
3170 | static int | | 3221 | static int |
3171 | wapbl_modcmd(modcmd_t cmd, void *arg) | | 3222 | wapbl_modcmd(modcmd_t cmd, void *arg) |
3172 | { | | 3223 | { |
3173 | | | 3224 | |
3174 | switch (cmd) { | | 3225 | switch (cmd) { |
3175 | case MODULE_CMD_INIT: | | 3226 | case MODULE_CMD_INIT: |
3176 | wapbl_init(); | | 3227 | wapbl_init(); |
3177 | return 0; | | 3228 | return 0; |
3178 | case MODULE_CMD_FINI: | | 3229 | case MODULE_CMD_FINI: |
3179 | return wapbl_fini(); | | 3230 | return wapbl_fini(); |
3180 | default: | | 3231 | default: |
3181 | return ENOTTY; | | 3232 | return ENOTTY; |
3182 | } | | 3233 | } |
3183 | } | | 3234 | } |
3184 | #endif /* _KERNEL */ | | 3235 | #endif /* _KERNEL */ |