1
2/*
3 * Copyright (c) 2000-2007 Apple Inc. All rights reserved.
4 *
5 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
6 *
7 * This file contains Original Code and/or Modifications of Original Code
8 * as defined in and that are subject to the Apple Public Source License
9 * Version 2.0 (the 'License'). You may not use this file except in
10 * compliance with the License. The rights granted to you under the License
11 * may not be used to create, or enable the creation or redistribution of,
12 * unlawful or unlicensed copies of an Apple operating system, or to
13 * circumvent, violate, or enable the circumvention or violation of, any
14 * terms of an Apple operating system software license agreement.
15 *
16 * Please obtain a copy of the License at
17 * http://www.opensource.apple.com/apsl/ and read it before using this file.
18 *
19 * The Original Code and all software distributed under the License are
20 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
21 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
22 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
23 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
24 * Please see the License for the specific language governing rights and
25 * limitations under the License.
26 *
27 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
28 */
29/*
30 * This header contains the structures and function prototypes
31 * for the vfs journaling code.  The data types are not meant
32 * to be modified by user code.  Just use the functions and do
33 * not mess around with the structs.
34 */
35#ifndef _SYS_VFS_JOURNAL_H_
36#define _SYS_VFS_JOURNAL_H_
37
38#include <sys/appleapiopts.h>
39#include <sys/cdefs.h>
40
41#ifdef __APPLE_API_UNSTABLE
42
43#include <sys/types.h>
44#include <kern/locks.h>
45
46typedef struct block_info {
47    off_t       bnum;                // block # on the file system device
48    size_t      bsize;               // in bytes
49    union {
50	int32_t     cksum;
51	uint32_t    sequence_num;    // only used in block_list_header->binfo[0]
52	struct buf *bp;
53    } b;
54} block_info;
55
56typedef struct block_list_header {
57    u_int16_t   max_blocks;          // max number of blocks in this chunk
58    u_int16_t   num_blocks;          // number of valid block numbers in block_nums
59    int32_t     bytes_used;          // how many bytes of this tbuffer are used
60    int32_t     checksum;            // on-disk: checksum of this header and binfo[0]
61    int32_t     flags;               // check-checksums, initial blhdr, etc
62    block_info  binfo[1];            // so we can reference them by name
63} block_list_header;
64
65#define BLHDR_CHECK_CHECKSUMS   0x0001
66#define BLHDR_FIRST_HEADER      0x0002
67
68
69struct journal;
70
71typedef struct transaction {
72    int                 tbuffer_size;  // in bytes
73    char               *tbuffer;       // memory copy of the transaction
74    block_list_header  *blhdr;         // points to the first byte of tbuffer
75    int                 num_blhdrs;    // how many buffers we've allocated
76    int                 total_bytes;   // total # of bytes in transaction
77    int                 num_flushed;   // how many bytes have been flushed
78    int                 num_killed;    // how many bytes were "killed"
79    off_t               journal_start; // where in the journal this transaction starts
80    off_t               journal_end;   // where in the journal this transaction ends
81    struct journal     *jnl;           // ptr back to the journal structure
82    struct transaction *next;          // list of tr's (either completed or to be free'd)
83    uint32_t            sequence_num;
84} transaction;
85
86
87/*
88 * This is written to block zero of the journal and it
89 * maintains overall state about the journal.
90 */
91typedef struct journal_header {
92    int32_t        magic;
93    int32_t        endian;
94    volatile off_t start;         // zero-based byte offset of the start of the first transaction
95    volatile off_t end;           // zero-based byte offset of where free space begins
96    off_t          size;          // size in bytes of the entire journal
97    int32_t        blhdr_size;    // size in bytes of each block_list_header in the journal
98    int32_t        checksum;
99    int32_t        jhdr_size;     // block size (in bytes) of the journal header
100    uint32_t       sequence_num;  // NEW FIELD: a monotonically increasing value assigned to all txn's
101} journal_header;
102
103#define JOURNAL_HEADER_MAGIC  0x4a4e4c78   // 'JNLx'
104#define ENDIAN_MAGIC          0x12345678
105
106//
107// we only checksum the original size of the journal_header to remain
108// backwards compatible.  the size of the original journal_heade is
109// everything up to the the sequence_num field, hence we use the
110// offsetof macro to calculate the size.
111//
112#define JOURNAL_HEADER_CKSUM_SIZE  (offsetof(struct journal_header, sequence_num))
113
114#define OLD_JOURNAL_HEADER_MAGIC  0x4a484452   // 'JHDR'
115
116
117/*
118 * In memory structure about the journal.
119 */
120typedef struct journal {
121    lck_mtx_t           jlock;             // protects the struct journal data
122
123    struct vnode       *jdev;              // vnode of the device where the journal lives
124    off_t               jdev_offset;       // byte offset to the start of the journal
125    const char         *jdev_name;
126
127    struct vnode       *fsdev;             // vnode of the file system device
128
129    void              (*flush)(void *arg); // fs callback to flush meta data blocks
130    void               *flush_arg;         // arg that's passed to flush()
131
132    int32_t             flags;
133    int32_t             tbuffer_size;      // default transaction buffer size
134
135    char               *header_buf;        // in-memory copy of the journal header
136    journal_header     *jhdr;              // points to the first byte of header_buf
137
138    off_t               max_read_size;
139    off_t               max_write_size;
140
141    transaction        *cur_tr;            // for group-commit
142    transaction        *completed_trs;     // out-of-order transactions that completed
143    transaction        *active_tr;         // for nested transactions
144    int32_t             nested_count;      // for nested transactions
145    void               *owner;             // a ptr that's unique to the calling process
146
147    transaction        *tr_freeme;         // transaction structs that need to be free'd
148
149    volatile off_t      active_start;      // the active start that we only keep in memory
150    lck_mtx_t           old_start_lock;    // protects the old_start
151    volatile off_t      old_start[16];     // this is how we do lazy start update
152
153    int                 last_flush_err;    // last error from flushing the cache
154} journal;
155
156/* internal-only journal flags (top 16 bits) */
157#define JOURNAL_CLOSE_PENDING     0x00010000
158#define JOURNAL_INVALID           0x00020000
159#define JOURNAL_FLUSHCACHE_ERR    0x00040000   // means we already printed this err
160#define JOURNAL_NEED_SWAP         0x00080000   // swap any data read from disk
161#define JOURNAL_DO_FUA_WRITES     0x00100000   // do force-unit-access writes
162
163/* journal_open/create options are always in the low-16 bits */
164#define JOURNAL_OPTION_FLAGS_MASK 0x0000ffff
165
166__BEGIN_DECLS
167/*
168 * Prototypes.
169 */
170
171/*
172 * Call journal_init() to initialize the journaling code (sets up lock attributes)
173 */
174void      journal_init(void) __attribute__((section("__TEXT, initcode")));
175
176/*
177 * Call journal_create() to create a new journal.  You only
178 * call this once, typically at file system creation time.
179 *
180 * The "jvp" argument is the vnode where the journal is written.
181 * The journal starts at "offset" and is "journal_size" bytes long.
182 *
183 * The "fsvp" argument is the vnode of your file system.  It may be
184 * the same as "jvp".
185 *
186 * The "min_fs_block_size" argument is the minimum block size
187 * (in bytes) that the file system will ever write.  Typically
188 * this is the block size of the file system (1k, 4k, etc) but
189 * on HFS+ it is the minimum block size of the underlying device.
190 *
191 * The flags argument lets you disable group commit if you
192 * want tighter guarantees on transactions (in exchange for
193 * lower performance).
194 *
195 * The tbuffer_size is the size of the transaction buffer
196 * used by the journal. If you specify zero, the journal code
197 * will use a reasonable defaults.  The tbuffer_size should
198 * be an integer multiple of the min_fs_block_size.
199 *
200 * Returns a valid journal pointer or NULL if one could not
201 * be created.
202 */
203journal *journal_create(struct vnode *jvp,
204						off_t         offset,
205						off_t         journal_size,
206						struct vnode *fsvp,
207						size_t        min_fs_block_size,
208						int32_t       flags,
209						int32_t       tbuffer_size,
210						void        (*flush)(void *arg),
211						void         *arg);
212
213/*
214 * Call journal_open() when mounting an existing file system
215 * that has a previously created journal.  It will take care
216 * of validating the journal and replaying it if necessary.
217 *
218 * See journal_create() for a description of the arguments.
219 *
220 * Returns a valid journal pointer of NULL if it runs into
221 * trouble reading/playing back the journal.
222 */
223journal  *journal_open(struct vnode *jvp,
224					   off_t         offset,
225					   off_t         journal_size,
226					   struct vnode *fsvp,
227					   size_t        min_fs_block_size,
228					   int32_t       flags,
229					   int32_t       tbuffer_size,
230					   void        (*flush)(void *arg),
231					   void         *arg);
232
233/*
234 * Test whether the journal is clean or not.  This is intended
235 * to be used when you're mounting read-only.  If the journal
236 * is not clean for some reason then you should not mount the
237 * volume as your data structures may be in an unknown state.
238 */
239int journal_is_clean(struct vnode *jvp,
240		     off_t         offset,
241		     off_t         journal_size,
242		     struct vnode *fsvp,
243                     size_t        min_fs_block_size);
244
245
246/*
247 * Call journal_close() just before your file system is unmounted.
248 * It flushes any outstanding transactions and makes sure the
249 * journal is in a consistent state.
250 */
251void      journal_close(journal *journalp);
252
253/*
254 * flags for journal_create/open.  only can use
255 * the low 16 bits for flags because internal
256 * bits go in the high 16.
257 */
258#define JOURNAL_NO_GROUP_COMMIT   0x00000001
259#define JOURNAL_RESET             0x00000002
260
261/*
262 * Transaction related functions.
263 *
264 * Before you start modifying file system meta data, you
265 * should call journal_start_transaction().  Then before
266 * you modify each block, call journal_modify_block_start()
267 * and when you're done, journal_modify_block_end().  When
268 * you've modified the last block as part of a transaction,
269 * call journal_end_transaction() to commit the changes.
270 *
271 * If you decide to abort the modifications to a block you
272 * should call journal_modify_block_abort().
273 *
274 * If as part of a transaction you need want to throw out
275 * any previous copies of a block (because it got deleted)
276 * then call journal_kill_block().  This will mark it so
277 * that the journal does not play it back (effectively
278 * dropping it).
279 */
280int   journal_start_transaction(journal *jnl);
281int   journal_modify_block_start(journal *jnl, struct buf *bp);
282int   journal_modify_block_abort(journal *jnl, struct buf *bp);
283int   journal_modify_block_end(journal *jnl, struct buf *bp, void (*func)(struct buf *bp, void *arg), void *arg);
284int   journal_kill_block(journal *jnl, struct buf *bp);
285int   journal_end_transaction(journal *jnl);
286
287int   journal_active(journal *jnl);
288int   journal_flush(journal *jnl);
289void *journal_owner(journal *jnl);    // compare against current_thread()
290int   journal_uses_fua(journal *jnl);
291
292
293/*
294 * Relocate the journal.
295 *
296 * You provide the new starting offset and size for the journal. You may
297 * optionally provide a new tbuffer_size; passing zero defaults to not
298 * changing the tbuffer size except as needed to fit within the new journal
299 * size.
300 *
301 * You must have already started a transaction. The transaction may contain
302 * modified blocks (such as those needed to deallocate the old journal,
303 * allocate the new journal, and update the location and size of the journal
304 * in filesystem-private structures). Any transactions prior to the active
305 * transaction will be flushed to the old journal. The new journal will be
306 * initialized, and the blocks from the active transaction will be written to
307 * the new journal. The caller will need to update the structures that
308 * identify the location and size of the journal from the callback routine.
309 */
310int journal_relocate(journal *jnl, off_t offset, off_t journal_size, int32_t tbuffer_size,
311	errno_t (*callback)(void *), void *callback_arg);
312
313__END_DECLS
314
315#endif /* __APPLE_API_UNSTABLE */
316#endif /* !_SYS_VFS_JOURNAL_H_ */
317