Print this page
5056 ZFS deadlock on db_mtx and dn_holds
Reviewed by: Will Andrews <willa@spectralogic.com>
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Approved by: Dan McDonald <danmcd@omniti.com>
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/sys/dnode.h
+++ new/usr/src/uts/common/fs/zfs/sys/dnode.h
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
|
↓ open down ↓ |
13 lines elided |
↑ open up ↑ |
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 23 * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
24 + * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
24 25 */
25 26
26 27 #ifndef _SYS_DNODE_H
27 28 #define _SYS_DNODE_H
28 29
29 30 #include <sys/zfs_context.h>
30 31 #include <sys/avl.h>
31 32 #include <sys/spa.h>
32 33 #include <sys/txg.h>
33 34 #include <sys/zio.h>
34 35 #include <sys/refcount.h>
35 36 #include <sys/dmu_zfetch.h>
36 37 #include <sys/zrlock.h>
37 38
38 39 #ifdef __cplusplus
39 40 extern "C" {
40 41 #endif
41 42
42 43 /*
43 44 * dnode_hold() flags.
44 45 */
45 46 #define DNODE_MUST_BE_ALLOCATED 1
46 47 #define DNODE_MUST_BE_FREE 2
47 48
48 49 /*
49 50 * dnode_next_offset() flags.
50 51 */
51 52 #define DNODE_FIND_HOLE 1
52 53 #define DNODE_FIND_BACKWARDS 2
53 54 #define DNODE_FIND_HAVELOCK 4
54 55
55 56 /*
56 57 * Fixed constants.
57 58 */
58 59 #define DNODE_SHIFT 9 /* 512 bytes */
59 60 #define DN_MIN_INDBLKSHIFT 12 /* 4k */
60 61 #define DN_MAX_INDBLKSHIFT 14 /* 16k */
61 62 #define DNODE_BLOCK_SHIFT 14 /* 16k */
62 63 #define DNODE_CORE_SIZE 64 /* 64 bytes for dnode sans blkptrs */
63 64 #define DN_MAX_OBJECT_SHIFT 48 /* 256 trillion (zfs_fid_t limit) */
64 65 #define DN_MAX_OFFSET_SHIFT 64 /* 2^64 bytes in a dnode */
65 66
66 67 /*
67 68 * dnode id flags
68 69 *
69 70 * Note: a file will never ever have its
70 71 * ids moved from bonus->spill
71 72 * and only in a crypto environment would it be on spill
72 73 */
73 74 #define DN_ID_CHKED_BONUS 0x1
74 75 #define DN_ID_CHKED_SPILL 0x2
75 76 #define DN_ID_OLD_EXIST 0x4
76 77 #define DN_ID_NEW_EXIST 0x8
77 78
78 79 /*
79 80 * Derived constants.
80 81 */
81 82 #define DNODE_SIZE (1 << DNODE_SHIFT)
82 83 #define DN_MAX_NBLKPTR ((DNODE_SIZE - DNODE_CORE_SIZE) >> SPA_BLKPTRSHIFT)
83 84 #define DN_MAX_BONUSLEN (DNODE_SIZE - DNODE_CORE_SIZE - (1 << SPA_BLKPTRSHIFT))
84 85 #define DN_MAX_OBJECT (1ULL << DN_MAX_OBJECT_SHIFT)
85 86 #define DN_ZERO_BONUSLEN (DN_MAX_BONUSLEN + 1)
86 87 #define DN_KILL_SPILLBLK (1)
87 88
88 89 #define DNODES_PER_BLOCK_SHIFT (DNODE_BLOCK_SHIFT - DNODE_SHIFT)
89 90 #define DNODES_PER_BLOCK (1ULL << DNODES_PER_BLOCK_SHIFT)
90 91 #define DNODES_PER_LEVEL_SHIFT (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT)
91 92 #define DNODES_PER_LEVEL (1ULL << DNODES_PER_LEVEL_SHIFT)
92 93
93 94 /* The +2 here is a cheesy way to round up */
94 95 #define DN_MAX_LEVELS (2 + ((DN_MAX_OFFSET_SHIFT - SPA_MINBLOCKSHIFT) / \
95 96 (DN_MIN_INDBLKSHIFT - SPA_BLKPTRSHIFT)))
96 97
97 98 #define DN_BONUS(dnp) ((void*)((dnp)->dn_bonus + \
98 99 (((dnp)->dn_nblkptr - 1) * sizeof (blkptr_t))))
99 100
100 101 #define DN_USED_BYTES(dnp) (((dnp)->dn_flags & DNODE_FLAG_USED_BYTES) ? \
101 102 (dnp)->dn_used : (dnp)->dn_used << SPA_MINBLOCKSHIFT)
102 103
103 104 #define EPB(blkshift, typeshift) (1 << (blkshift - typeshift))
104 105
105 106 struct dmu_buf_impl;
106 107 struct objset;
107 108 struct zio;
108 109
109 110 enum dnode_dirtycontext {
110 111 DN_UNDIRTIED,
111 112 DN_DIRTY_OPEN,
112 113 DN_DIRTY_SYNC
113 114 };
114 115
115 116 /* Is dn_used in bytes? if not, it's in multiples of SPA_MINBLOCKSIZE */
116 117 #define DNODE_FLAG_USED_BYTES (1<<0)
117 118 #define DNODE_FLAG_USERUSED_ACCOUNTED (1<<1)
118 119
119 120 /* Does dnode have a SA spill blkptr in bonus? */
120 121 #define DNODE_FLAG_SPILL_BLKPTR (1<<2)
121 122
122 123 typedef struct dnode_phys {
123 124 uint8_t dn_type; /* dmu_object_type_t */
124 125 uint8_t dn_indblkshift; /* ln2(indirect block size) */
125 126 uint8_t dn_nlevels; /* 1=dn_blkptr->data blocks */
126 127 uint8_t dn_nblkptr; /* length of dn_blkptr */
127 128 uint8_t dn_bonustype; /* type of data in bonus buffer */
128 129 uint8_t dn_checksum; /* ZIO_CHECKSUM type */
129 130 uint8_t dn_compress; /* ZIO_COMPRESS type */
130 131 uint8_t dn_flags; /* DNODE_FLAG_* */
131 132 uint16_t dn_datablkszsec; /* data block size in 512b sectors */
132 133 uint16_t dn_bonuslen; /* length of dn_bonus */
133 134 uint8_t dn_pad2[4];
134 135
135 136 /* accounting is protected by dn_dirty_mtx */
136 137 uint64_t dn_maxblkid; /* largest allocated block ID */
137 138 uint64_t dn_used; /* bytes (or sectors) of disk space */
138 139
139 140 uint64_t dn_pad3[4];
140 141
141 142 blkptr_t dn_blkptr[1];
142 143 uint8_t dn_bonus[DN_MAX_BONUSLEN - sizeof (blkptr_t)];
143 144 blkptr_t dn_spill;
144 145 } dnode_phys_t;
145 146
146 147 typedef struct dnode {
147 148 /*
148 149 * Protects the structure of the dnode, including the number of levels
149 150 * of indirection (dn_nlevels), dn_maxblkid, and dn_next_*
150 151 */
151 152 krwlock_t dn_struct_rwlock;
152 153
153 154 /* Our link on dn_objset->os_dnodes list; protected by os_lock. */
154 155 list_node_t dn_link;
155 156
156 157 /* immutable: */
157 158 struct objset *dn_objset;
158 159 uint64_t dn_object;
159 160 struct dmu_buf_impl *dn_dbuf;
160 161 struct dnode_handle *dn_handle;
161 162 dnode_phys_t *dn_phys; /* pointer into dn->dn_dbuf->db.db_data */
162 163
163 164 /*
164 165 * Copies of stuff in dn_phys. They're valid in the open
165 166 * context (eg. even before the dnode is first synced).
166 167 * Where necessary, these are protected by dn_struct_rwlock.
167 168 */
168 169 dmu_object_type_t dn_type; /* object type */
169 170 uint16_t dn_bonuslen; /* bonus length */
170 171 uint8_t dn_bonustype; /* bonus type */
171 172 uint8_t dn_nblkptr; /* number of blkptrs (immutable) */
172 173 uint8_t dn_checksum; /* ZIO_CHECKSUM type */
173 174 uint8_t dn_compress; /* ZIO_COMPRESS type */
174 175 uint8_t dn_nlevels;
175 176 uint8_t dn_indblkshift;
176 177 uint8_t dn_datablkshift; /* zero if blksz not power of 2! */
177 178 uint8_t dn_moved; /* Has this dnode been moved? */
178 179 uint16_t dn_datablkszsec; /* in 512b sectors */
179 180 uint32_t dn_datablksz; /* in bytes */
180 181 uint64_t dn_maxblkid;
181 182 uint8_t dn_next_type[TXG_SIZE];
182 183 uint8_t dn_next_nblkptr[TXG_SIZE];
183 184 uint8_t dn_next_nlevels[TXG_SIZE];
184 185 uint8_t dn_next_indblkshift[TXG_SIZE];
185 186 uint8_t dn_next_bonustype[TXG_SIZE];
186 187 uint8_t dn_rm_spillblk[TXG_SIZE]; /* for removing spill blk */
187 188 uint16_t dn_next_bonuslen[TXG_SIZE];
188 189 uint32_t dn_next_blksz[TXG_SIZE]; /* next block size in bytes */
189 190
190 191 /* protected by dn_dbufs_mtx; declared here to fill 32-bit hole */
191 192 uint32_t dn_dbufs_count; /* count of dn_dbufs */
192 193 /* There are no level-0 blocks of this blkid or higher in dn_dbufs */
193 194 uint64_t dn_unlisted_l0_blkid;
194 195
195 196 /* protected by os_lock: */
196 197 list_node_t dn_dirty_link[TXG_SIZE]; /* next on dataset's dirty */
197 198
198 199 /* protected by dn_mtx: */
199 200 kmutex_t dn_mtx;
200 201 list_t dn_dirty_records[TXG_SIZE];
201 202 struct range_tree *dn_free_ranges[TXG_SIZE];
202 203 uint64_t dn_allocated_txg;
203 204 uint64_t dn_free_txg;
204 205 uint64_t dn_assigned_txg;
205 206 kcondvar_t dn_notxholds;
206 207 enum dnode_dirtycontext dn_dirtyctx;
207 208 uint8_t *dn_dirtyctx_firstset; /* dbg: contents meaningless */
208 209
209 210 /* protected by own devices */
210 211 refcount_t dn_tx_holds;
211 212 refcount_t dn_holds;
212 213
213 214 kmutex_t dn_dbufs_mtx;
214 215 /*
215 216 * Descendent dbufs, ordered by dbuf_compare. Note that dn_dbufs
216 217 * can contain multiple dbufs of the same (level, blkid) when a
217 218 * dbuf is marked DB_EVICTING without being removed from
218 219 * dn_dbufs. To maintain the avl invariant that there cannot be
219 220 * duplicate entries, we order the dbufs by an arbitrary value -
220 221 * their address in memory. This means that dn_dbufs cannot be used to
221 222 * directly look up a dbuf. Instead, callers must use avl_walk, have
222 223 * a reference to the dbuf, or look up a non-existant node with
223 224 * db_state = DB_SEARCH (see dbuf_free_range for an example).
224 225 */
225 226 avl_tree_t dn_dbufs;
226 227
227 228 /* protected by dn_struct_rwlock */
228 229 struct dmu_buf_impl *dn_bonus; /* bonus buffer dbuf */
229 230
230 231 boolean_t dn_have_spill; /* have spill or are spilling */
231 232
232 233 /* parent IO for current sync write */
233 234 zio_t *dn_zio;
234 235
235 236 /* used in syncing context */
236 237 uint64_t dn_oldused; /* old phys used bytes */
237 238 uint64_t dn_oldflags; /* old phys dn_flags */
238 239 uint64_t dn_olduid, dn_oldgid;
239 240 uint64_t dn_newuid, dn_newgid;
240 241 int dn_id_flags;
241 242
242 243 /* holds prefetch structure */
243 244 struct zfetch dn_zfetch;
244 245 } dnode_t;
245 246
246 247 /*
247 248 * Adds a level of indirection between the dbuf and the dnode to avoid
|
↓ open down ↓ |
214 lines elided |
↑ open up ↑ |
248 249 * iterating descendent dbufs in dnode_move(). Handles are not allocated
249 250 * individually, but as an array of child dnodes in dnode_hold_impl().
250 251 */
251 252 typedef struct dnode_handle {
252 253 /* Protects dnh_dnode from modification by dnode_move(). */
253 254 zrlock_t dnh_zrlock;
254 255 dnode_t *dnh_dnode;
255 256 } dnode_handle_t;
256 257
257 258 typedef struct dnode_children {
259 + dmu_buf_user_t dnc_dbu; /* User evict data */
258 260 size_t dnc_count; /* number of children */
259 261 dnode_handle_t dnc_children[]; /* sized dynamically */
260 262 } dnode_children_t;
261 263
262 264 typedef struct free_range {
263 265 avl_node_t fr_node;
264 266 uint64_t fr_blkid;
265 267 uint64_t fr_nblks;
266 268 } free_range_t;
267 269
268 -dnode_t *dnode_special_open(struct objset *dd, dnode_phys_t *dnp,
270 +void dnode_special_open(struct objset *dd, dnode_phys_t *dnp,
269 271 uint64_t object, dnode_handle_t *dnh);
270 272 void dnode_special_close(dnode_handle_t *dnh);
271 273
272 274 void dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx);
273 275 void dnode_setbonus_type(dnode_t *dn, dmu_object_type_t, dmu_tx_t *tx);
274 276 void dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx);
275 277
276 278 int dnode_hold(struct objset *dd, uint64_t object,
277 279 void *ref, dnode_t **dnp);
278 280 int dnode_hold_impl(struct objset *dd, uint64_t object, int flag,
279 281 void *ref, dnode_t **dnp);
280 282 boolean_t dnode_add_ref(dnode_t *dn, void *ref);
281 283 void dnode_rele(dnode_t *dn, void *ref);
282 284 void dnode_setdirty(dnode_t *dn, dmu_tx_t *tx);
283 285 void dnode_sync(dnode_t *dn, dmu_tx_t *tx);
284 286 void dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs,
285 287 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
286 288 void dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize,
287 289 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
288 290 void dnode_free(dnode_t *dn, dmu_tx_t *tx);
289 291 void dnode_byteswap(dnode_phys_t *dnp);
290 292 void dnode_buf_byteswap(void *buf, size_t size);
291 293 void dnode_verify(dnode_t *dn);
292 294 int dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx);
293 295 void dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx);
294 296 void dnode_diduse_space(dnode_t *dn, int64_t space);
295 297 void dnode_willuse_space(dnode_t *dn, int64_t space, dmu_tx_t *tx);
296 298 void dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, boolean_t);
297 299 uint64_t dnode_block_freed(dnode_t *dn, uint64_t blkid);
298 300 void dnode_init(void);
299 301 void dnode_fini(void);
300 302 int dnode_next_offset(dnode_t *dn, int flags, uint64_t *off,
301 303 int minlvl, uint64_t blkfill, uint64_t txg);
302 304 void dnode_evict_dbufs(dnode_t *dn);
303 305
304 306 #ifdef ZFS_DEBUG
305 307
306 308 /*
307 309 * There should be a ## between the string literal and fmt, to make it
308 310 * clear that we're joining two strings together, but that piece of shit
309 311 * gcc doesn't support that preprocessor token.
310 312 */
311 313 #define dprintf_dnode(dn, fmt, ...) do { \
312 314 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
313 315 char __db_buf[32]; \
314 316 uint64_t __db_obj = (dn)->dn_object; \
315 317 if (__db_obj == DMU_META_DNODE_OBJECT) \
316 318 (void) strcpy(__db_buf, "mdn"); \
317 319 else \
318 320 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
319 321 (u_longlong_t)__db_obj);\
320 322 dprintf_ds((dn)->dn_objset->os_dsl_dataset, "obj=%s " fmt, \
321 323 __db_buf, __VA_ARGS__); \
322 324 } \
323 325 _NOTE(CONSTCOND) } while (0)
324 326
325 327 #define DNODE_VERIFY(dn) dnode_verify(dn)
326 328 #define FREE_VERIFY(db, start, end, tx) free_verify(db, start, end, tx)
327 329
328 330 #else
329 331
330 332 #define dprintf_dnode(db, fmt, ...)
331 333 #define DNODE_VERIFY(dn)
332 334 #define FREE_VERIFY(db, start, end, tx)
333 335
334 336 #endif
335 337
336 338 #ifdef __cplusplus
337 339 }
338 340 #endif
339 341
340 342 #endif /* _SYS_DNODE_H */
|
↓ open down ↓ |
62 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX