1 // SPDX-License-Identifier: GPL-2.0 1 // SPDX-License-Identifier: GPL-2.0
2 /* 2 /*
3 * Copyright (c) 2000-2003,2005 Silicon Graphi 3 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
4 * Copyright (C) 2010 Red Hat, Inc. 4 * Copyright (C) 2010 Red Hat, Inc.
5 * All Rights Reserved. 5 * All Rights Reserved.
6 */ 6 */
7 #include "xfs.h" 7 #include "xfs.h"
8 #include "xfs_fs.h" 8 #include "xfs_fs.h"
9 #include "xfs_shared.h" 9 #include "xfs_shared.h"
10 #include "xfs_format.h" 10 #include "xfs_format.h"
11 #include "xfs_log_format.h" 11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h" 12 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h" 13 #include "xfs_mount.h"
14 #include "xfs_extent_busy.h" 14 #include "xfs_extent_busy.h"
15 #include "xfs_quota.h" 15 #include "xfs_quota.h"
16 #include "xfs_trans.h" 16 #include "xfs_trans.h"
17 #include "xfs_trans_priv.h" 17 #include "xfs_trans_priv.h"
18 #include "xfs_log.h" 18 #include "xfs_log.h"
19 #include "xfs_log_priv.h" 19 #include "xfs_log_priv.h"
20 #include "xfs_trace.h" 20 #include "xfs_trace.h"
21 #include "xfs_error.h" 21 #include "xfs_error.h"
22 #include "xfs_defer.h" 22 #include "xfs_defer.h"
23 #include "xfs_inode.h" 23 #include "xfs_inode.h"
24 #include "xfs_dquot_item.h" 24 #include "xfs_dquot_item.h"
25 #include "xfs_dquot.h" 25 #include "xfs_dquot.h"
26 #include "xfs_icache.h" 26 #include "xfs_icache.h"
27 #include "xfs_rtbitmap.h" <<
28 27
29 struct kmem_cache *xfs_trans_cache; 28 struct kmem_cache *xfs_trans_cache;
30 29
31 #if defined(CONFIG_TRACEPOINTS) 30 #if defined(CONFIG_TRACEPOINTS)
32 static void 31 static void
33 xfs_trans_trace_reservations( 32 xfs_trans_trace_reservations(
34 struct xfs_mount *mp) 33 struct xfs_mount *mp)
35 { 34 {
>> 35 struct xfs_trans_res resv;
36 struct xfs_trans_res *res; 36 struct xfs_trans_res *res;
37 struct xfs_trans_res *end_res; 37 struct xfs_trans_res *end_res;
38 int i; 38 int i;
39 39
40 res = (struct xfs_trans_res *)M_RES(mp 40 res = (struct xfs_trans_res *)M_RES(mp);
41 end_res = (struct xfs_trans_res *)(M_R 41 end_res = (struct xfs_trans_res *)(M_RES(mp) + 1);
42 for (i = 0; res < end_res; i++, res++) 42 for (i = 0; res < end_res; i++, res++)
43 trace_xfs_trans_resv_calc(mp, 43 trace_xfs_trans_resv_calc(mp, i, res);
>> 44 xfs_log_get_max_trans_res(mp, &resv);
>> 45 trace_xfs_trans_resv_calc(mp, -1, &resv);
44 } 46 }
45 #else 47 #else
46 # define xfs_trans_trace_reservations(mp) 48 # define xfs_trans_trace_reservations(mp)
47 #endif 49 #endif
48 50
49 /* 51 /*
50 * Initialize the precomputed transaction rese 52 * Initialize the precomputed transaction reservation values
51 * in the mount structure. 53 * in the mount structure.
52 */ 54 */
53 void 55 void
54 xfs_trans_init( 56 xfs_trans_init(
55 struct xfs_mount *mp) 57 struct xfs_mount *mp)
56 { 58 {
57 xfs_trans_resv_calc(mp, M_RES(mp)); 59 xfs_trans_resv_calc(mp, M_RES(mp));
58 xfs_trans_trace_reservations(mp); 60 xfs_trans_trace_reservations(mp);
59 } 61 }
60 62
61 /* 63 /*
62 * Free the transaction structure. If there i 64 * Free the transaction structure. If there is more clean up
63 * to do when the structure is freed, add it h 65 * to do when the structure is freed, add it here.
64 */ 66 */
65 STATIC void 67 STATIC void
66 xfs_trans_free( 68 xfs_trans_free(
67 struct xfs_trans *tp) 69 struct xfs_trans *tp)
68 { 70 {
69 xfs_extent_busy_sort(&tp->t_busy); 71 xfs_extent_busy_sort(&tp->t_busy);
70 xfs_extent_busy_clear(tp->t_mountp, &t 72 xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false);
71 73
72 trace_xfs_trans_free(tp, _RET_IP_); 74 trace_xfs_trans_free(tp, _RET_IP_);
73 xfs_trans_clear_context(tp); 75 xfs_trans_clear_context(tp);
74 if (!(tp->t_flags & XFS_TRANS_NO_WRITE 76 if (!(tp->t_flags & XFS_TRANS_NO_WRITECOUNT))
75 sb_end_intwrite(tp->t_mountp-> 77 sb_end_intwrite(tp->t_mountp->m_super);
76 xfs_trans_free_dqinfo(tp); 78 xfs_trans_free_dqinfo(tp);
77 kmem_cache_free(xfs_trans_cache, tp); 79 kmem_cache_free(xfs_trans_cache, tp);
78 } 80 }
79 81
80 /* 82 /*
81 * This is called to create a new transaction 83 * This is called to create a new transaction which will share the
82 * permanent log reservation of the given tran 84 * permanent log reservation of the given transaction. The remaining
83 * unused block and rt extent reservations are 85 * unused block and rt extent reservations are also inherited. This
84 * implies that the original transaction is no 86 * implies that the original transaction is no longer allowed to allocate
85 * blocks. Locks and log items, however, are 87 * blocks. Locks and log items, however, are no inherited. They must
86 * be added to the new transaction explicitly. 88 * be added to the new transaction explicitly.
87 */ 89 */
88 STATIC struct xfs_trans * 90 STATIC struct xfs_trans *
89 xfs_trans_dup( 91 xfs_trans_dup(
90 struct xfs_trans *tp) 92 struct xfs_trans *tp)
91 { 93 {
92 struct xfs_trans *ntp; 94 struct xfs_trans *ntp;
93 95
94 trace_xfs_trans_dup(tp, _RET_IP_); 96 trace_xfs_trans_dup(tp, _RET_IP_);
95 97
96 ntp = kmem_cache_zalloc(xfs_trans_cach 98 ntp = kmem_cache_zalloc(xfs_trans_cache, GFP_KERNEL | __GFP_NOFAIL);
97 99
98 /* 100 /*
99 * Initialize the new transaction stru 101 * Initialize the new transaction structure.
100 */ 102 */
101 ntp->t_magic = XFS_TRANS_HEADER_MAGIC; 103 ntp->t_magic = XFS_TRANS_HEADER_MAGIC;
102 ntp->t_mountp = tp->t_mountp; 104 ntp->t_mountp = tp->t_mountp;
103 INIT_LIST_HEAD(&ntp->t_items); 105 INIT_LIST_HEAD(&ntp->t_items);
104 INIT_LIST_HEAD(&ntp->t_busy); 106 INIT_LIST_HEAD(&ntp->t_busy);
105 INIT_LIST_HEAD(&ntp->t_dfops); 107 INIT_LIST_HEAD(&ntp->t_dfops);
106 ntp->t_highest_agno = NULLAGNUMBER; !! 108 ntp->t_firstblock = NULLFSBLOCK;
107 109
108 ASSERT(tp->t_flags & XFS_TRANS_PERM_LO 110 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
109 ASSERT(tp->t_ticket != NULL); 111 ASSERT(tp->t_ticket != NULL);
110 112
111 ntp->t_flags = XFS_TRANS_PERM_LOG_RES 113 ntp->t_flags = XFS_TRANS_PERM_LOG_RES |
112 (tp->t_flags & XFS_TRAN 114 (tp->t_flags & XFS_TRANS_RESERVE) |
113 (tp->t_flags & XFS_TRAN 115 (tp->t_flags & XFS_TRANS_NO_WRITECOUNT) |
114 (tp->t_flags & XFS_TRAN 116 (tp->t_flags & XFS_TRANS_RES_FDBLKS);
115 /* We gave our writer reference to the 117 /* We gave our writer reference to the new transaction */
116 tp->t_flags |= XFS_TRANS_NO_WRITECOUNT 118 tp->t_flags |= XFS_TRANS_NO_WRITECOUNT;
117 ntp->t_ticket = xfs_log_ticket_get(tp- 119 ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
118 120
119 ASSERT(tp->t_blk_res >= tp->t_blk_res_ 121 ASSERT(tp->t_blk_res >= tp->t_blk_res_used);
120 ntp->t_blk_res = tp->t_blk_res - tp->t 122 ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
121 tp->t_blk_res = tp->t_blk_res_used; 123 tp->t_blk_res = tp->t_blk_res_used;
122 124
123 ntp->t_rtx_res = tp->t_rtx_res - tp->t 125 ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
124 tp->t_rtx_res = tp->t_rtx_res_used; 126 tp->t_rtx_res = tp->t_rtx_res_used;
125 127
126 xfs_trans_switch_context(tp, ntp); 128 xfs_trans_switch_context(tp, ntp);
127 129
128 /* move deferred ops over to the new t 130 /* move deferred ops over to the new tp */
129 xfs_defer_move(ntp, tp); 131 xfs_defer_move(ntp, tp);
130 132
131 xfs_trans_dup_dqinfo(tp, ntp); 133 xfs_trans_dup_dqinfo(tp, ntp);
132 return ntp; 134 return ntp;
133 } 135 }
134 136
135 /* 137 /*
136 * This is called to reserve free disk blocks 138 * This is called to reserve free disk blocks and log space for the
137 * given transaction. This must be done befor 139 * given transaction. This must be done before allocating any resources
138 * within the transaction. 140 * within the transaction.
139 * 141 *
140 * This will return ENOSPC if there are not en 142 * This will return ENOSPC if there are not enough blocks available.
141 * It will sleep waiting for available log spa 143 * It will sleep waiting for available log space.
142 * The only valid value for the flags paramete 144 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
143 * is used by long running transactions. If a 145 * is used by long running transactions. If any one of the reservations
144 * fails then they will all be backed out. 146 * fails then they will all be backed out.
145 * 147 *
146 * This does not do quota reservations. That t 148 * This does not do quota reservations. That typically is done by the
147 * caller afterwards. 149 * caller afterwards.
148 */ 150 */
149 static int 151 static int
150 xfs_trans_reserve( 152 xfs_trans_reserve(
151 struct xfs_trans *tp, 153 struct xfs_trans *tp,
152 struct xfs_trans_res *resp, 154 struct xfs_trans_res *resp,
153 uint blocks, 155 uint blocks,
154 uint rtextents) 156 uint rtextents)
155 { 157 {
156 struct xfs_mount *mp = tp->t_mo 158 struct xfs_mount *mp = tp->t_mountp;
157 int error = 0; 159 int error = 0;
158 bool rsvd = (tp->t_ 160 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
159 161
160 /* 162 /*
161 * Attempt to reserve the needed disk 163 * Attempt to reserve the needed disk blocks by decrementing
162 * the number needed from the number a 164 * the number needed from the number available. This will
163 * fail if the count would go below ze 165 * fail if the count would go below zero.
164 */ 166 */
165 if (blocks > 0) { 167 if (blocks > 0) {
166 error = xfs_dec_fdblocks(mp, b !! 168 error = xfs_mod_fdblocks(mp, -((int64_t)blocks), rsvd);
167 if (error != 0) 169 if (error != 0)
168 return -ENOSPC; 170 return -ENOSPC;
169 tp->t_blk_res += blocks; 171 tp->t_blk_res += blocks;
170 } 172 }
171 173
172 /* 174 /*
173 * Reserve the log space needed for th 175 * Reserve the log space needed for this transaction.
174 */ 176 */
175 if (resp->tr_logres > 0) { 177 if (resp->tr_logres > 0) {
176 bool permanent = false; 178 bool permanent = false;
177 179
178 ASSERT(tp->t_log_res == 0 || 180 ASSERT(tp->t_log_res == 0 ||
179 tp->t_log_res == resp-> 181 tp->t_log_res == resp->tr_logres);
180 ASSERT(tp->t_log_count == 0 || 182 ASSERT(tp->t_log_count == 0 ||
181 tp->t_log_count == resp 183 tp->t_log_count == resp->tr_logcount);
182 184
183 if (resp->tr_logflags & XFS_TR 185 if (resp->tr_logflags & XFS_TRANS_PERM_LOG_RES) {
184 tp->t_flags |= XFS_TRA 186 tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
185 permanent = true; 187 permanent = true;
186 } else { 188 } else {
187 ASSERT(tp->t_ticket == 189 ASSERT(tp->t_ticket == NULL);
188 ASSERT(!(tp->t_flags & 190 ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
189 } 191 }
190 192
191 if (tp->t_ticket != NULL) { 193 if (tp->t_ticket != NULL) {
192 ASSERT(resp->tr_logfla 194 ASSERT(resp->tr_logflags & XFS_TRANS_PERM_LOG_RES);
193 error = xfs_log_regran 195 error = xfs_log_regrant(mp, tp->t_ticket);
194 } else { 196 } else {
195 error = xfs_log_reserv !! 197 error = xfs_log_reserve(mp,
>> 198 resp->tr_logres,
196 199 resp->tr_logcount,
197 !! 200 &tp->t_ticket, XFS_TRANSACTION,
>> 201 permanent);
198 } 202 }
199 203
200 if (error) 204 if (error)
201 goto undo_blocks; 205 goto undo_blocks;
202 206
203 tp->t_log_res = resp->tr_logre 207 tp->t_log_res = resp->tr_logres;
204 tp->t_log_count = resp->tr_log 208 tp->t_log_count = resp->tr_logcount;
205 } 209 }
206 210
207 /* 211 /*
208 * Attempt to reserve the needed realt 212 * Attempt to reserve the needed realtime extents by decrementing
209 * the number needed from the number a 213 * the number needed from the number available. This will
210 * fail if the count would go below ze 214 * fail if the count would go below zero.
211 */ 215 */
212 if (rtextents > 0) { 216 if (rtextents > 0) {
213 error = xfs_dec_frextents(mp, !! 217 error = xfs_mod_frextents(mp, -((int64_t)rtextents));
214 if (error) { 218 if (error) {
215 error = -ENOSPC; 219 error = -ENOSPC;
216 goto undo_log; 220 goto undo_log;
217 } 221 }
218 tp->t_rtx_res += rtextents; 222 tp->t_rtx_res += rtextents;
219 } 223 }
220 224
221 return 0; 225 return 0;
222 226
223 /* 227 /*
224 * Error cases jump to one of these la 228 * Error cases jump to one of these labels to undo any
225 * reservations which have already bee 229 * reservations which have already been performed.
226 */ 230 */
227 undo_log: 231 undo_log:
228 if (resp->tr_logres > 0) { 232 if (resp->tr_logres > 0) {
229 xfs_log_ticket_ungrant(mp->m_l 233 xfs_log_ticket_ungrant(mp->m_log, tp->t_ticket);
230 tp->t_ticket = NULL; 234 tp->t_ticket = NULL;
231 tp->t_log_res = 0; 235 tp->t_log_res = 0;
232 tp->t_flags &= ~XFS_TRANS_PERM 236 tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
233 } 237 }
234 238
235 undo_blocks: 239 undo_blocks:
236 if (blocks > 0) { 240 if (blocks > 0) {
237 xfs_add_fdblocks(mp, blocks); !! 241 xfs_mod_fdblocks(mp, (int64_t)blocks, rsvd);
238 tp->t_blk_res = 0; 242 tp->t_blk_res = 0;
239 } 243 }
240 return error; 244 return error;
241 } 245 }
242 246
243 int 247 int
244 xfs_trans_alloc( 248 xfs_trans_alloc(
245 struct xfs_mount *mp, 249 struct xfs_mount *mp,
246 struct xfs_trans_res *resp, 250 struct xfs_trans_res *resp,
247 uint blocks, 251 uint blocks,
248 uint rtextents, 252 uint rtextents,
249 uint flags, 253 uint flags,
250 struct xfs_trans **tpp) 254 struct xfs_trans **tpp)
251 { 255 {
252 struct xfs_trans *tp; 256 struct xfs_trans *tp;
253 bool want_retry = t 257 bool want_retry = true;
254 int error; 258 int error;
255 259
256 /* 260 /*
257 * Allocate the handle before we do ou 261 * Allocate the handle before we do our freeze accounting and setting up
258 * GFP_NOFS allocation context so that 262 * GFP_NOFS allocation context so that we avoid lockdep false positives
259 * by doing GFP_KERNEL allocations ins 263 * by doing GFP_KERNEL allocations inside sb_start_intwrite().
260 */ 264 */
261 retry: 265 retry:
262 tp = kmem_cache_zalloc(xfs_trans_cache 266 tp = kmem_cache_zalloc(xfs_trans_cache, GFP_KERNEL | __GFP_NOFAIL);
263 if (!(flags & XFS_TRANS_NO_WRITECOUNT) 267 if (!(flags & XFS_TRANS_NO_WRITECOUNT))
264 sb_start_intwrite(mp->m_super) 268 sb_start_intwrite(mp->m_super);
265 xfs_trans_set_context(tp); 269 xfs_trans_set_context(tp);
266 270
267 /* 271 /*
268 * Zero-reservation ("empty") transact 272 * Zero-reservation ("empty") transactions can't modify anything, so
269 * they're allowed to run while we're 273 * they're allowed to run while we're frozen.
270 */ 274 */
271 WARN_ON(resp->tr_logres > 0 && 275 WARN_ON(resp->tr_logres > 0 &&
272 mp->m_super->s_writers.frozen 276 mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE);
273 ASSERT(!(flags & XFS_TRANS_RES_FDBLKS) 277 ASSERT(!(flags & XFS_TRANS_RES_FDBLKS) ||
274 xfs_has_lazysbcount(mp)); 278 xfs_has_lazysbcount(mp));
275 279
276 tp->t_magic = XFS_TRANS_HEADER_MAGIC; 280 tp->t_magic = XFS_TRANS_HEADER_MAGIC;
277 tp->t_flags = flags; 281 tp->t_flags = flags;
278 tp->t_mountp = mp; 282 tp->t_mountp = mp;
279 INIT_LIST_HEAD(&tp->t_items); 283 INIT_LIST_HEAD(&tp->t_items);
280 INIT_LIST_HEAD(&tp->t_busy); 284 INIT_LIST_HEAD(&tp->t_busy);
281 INIT_LIST_HEAD(&tp->t_dfops); 285 INIT_LIST_HEAD(&tp->t_dfops);
282 tp->t_highest_agno = NULLAGNUMBER; !! 286 tp->t_firstblock = NULLFSBLOCK;
283 287
284 error = xfs_trans_reserve(tp, resp, bl 288 error = xfs_trans_reserve(tp, resp, blocks, rtextents);
285 if (error == -ENOSPC && want_retry) { 289 if (error == -ENOSPC && want_retry) {
286 xfs_trans_cancel(tp); 290 xfs_trans_cancel(tp);
287 291
288 /* 292 /*
289 * We weren't able to reserve 293 * We weren't able to reserve enough space for the transaction.
290 * Flush the other speculative 294 * Flush the other speculative space allocations to free space.
291 * Do not perform a synchronou 295 * Do not perform a synchronous scan because callers can hold
292 * other locks. 296 * other locks.
293 */ 297 */
294 error = xfs_blockgc_flush_all( !! 298 xfs_blockgc_flush_all(mp);
295 if (error) <<
296 return error; <<
297 want_retry = false; 299 want_retry = false;
298 goto retry; 300 goto retry;
299 } 301 }
300 if (error) { 302 if (error) {
301 xfs_trans_cancel(tp); 303 xfs_trans_cancel(tp);
302 return error; 304 return error;
303 } 305 }
304 306
305 trace_xfs_trans_alloc(tp, _RET_IP_); 307 trace_xfs_trans_alloc(tp, _RET_IP_);
306 308
307 *tpp = tp; 309 *tpp = tp;
308 return 0; 310 return 0;
309 } 311 }
310 312
311 /* 313 /*
312 * Create an empty transaction with no reserva 314 * Create an empty transaction with no reservation. This is a defensive
313 * mechanism for routines that query metadata 315 * mechanism for routines that query metadata without actually modifying them --
314 * if the metadata being queried is somehow cr 316 * if the metadata being queried is somehow cross-linked (think a btree block
315 * pointer that points higher in the tree), we 317 * pointer that points higher in the tree), we risk deadlock. However, blocks
316 * grabbed as part of a transaction can be re- 318 * grabbed as part of a transaction can be re-grabbed. The verifiers will
317 * notice the corrupt block and the operation 319 * notice the corrupt block and the operation will fail back to userspace
318 * without deadlocking. 320 * without deadlocking.
319 * 321 *
320 * Note the zero-length reservation; this tran 322 * Note the zero-length reservation; this transaction MUST be cancelled without
321 * any dirty data. 323 * any dirty data.
322 * 324 *
323 * Callers should obtain freeze protection to 325 * Callers should obtain freeze protection to avoid a conflict with fs freezing
324 * where we can be grabbing buffers at the sam 326 * where we can be grabbing buffers at the same time that freeze is trying to
325 * drain the buffer LRU list. 327 * drain the buffer LRU list.
326 */ 328 */
327 int 329 int
328 xfs_trans_alloc_empty( 330 xfs_trans_alloc_empty(
329 struct xfs_mount *mp, 331 struct xfs_mount *mp,
330 struct xfs_trans **tpp) 332 struct xfs_trans **tpp)
331 { 333 {
332 struct xfs_trans_res resv = 334 struct xfs_trans_res resv = {0};
333 335
334 return xfs_trans_alloc(mp, &resv, 0, 0 336 return xfs_trans_alloc(mp, &resv, 0, 0, XFS_TRANS_NO_WRITECOUNT, tpp);
335 } 337 }
336 338
337 /* 339 /*
338 * Record the indicated change to the given fi 340 * Record the indicated change to the given field for application
339 * to the file system's superblock when the tr 341 * to the file system's superblock when the transaction commits.
340 * For now, just store the change in the trans 342 * For now, just store the change in the transaction structure.
341 * 343 *
342 * Mark the transaction structure to indicate 344 * Mark the transaction structure to indicate that the superblock
343 * needs to be updated before committing. 345 * needs to be updated before committing.
344 * 346 *
345 * Because we may not be keeping track of allo 347 * Because we may not be keeping track of allocated/free inodes and
346 * used filesystem blocks in the superblock, w 348 * used filesystem blocks in the superblock, we do not mark the
347 * superblock dirty in this transaction if we 349 * superblock dirty in this transaction if we modify these fields.
348 * We still need to update the transaction del 350 * We still need to update the transaction deltas so that they get
349 * applied to the incore superblock, but we do 351 * applied to the incore superblock, but we don't want them to
350 * cause the superblock to get locked and logg 352 * cause the superblock to get locked and logged if these are the
351 * only fields in the superblock that the tran 353 * only fields in the superblock that the transaction modifies.
352 */ 354 */
353 void 355 void
354 xfs_trans_mod_sb( 356 xfs_trans_mod_sb(
355 xfs_trans_t *tp, 357 xfs_trans_t *tp,
356 uint field, 358 uint field,
357 int64_t delta) 359 int64_t delta)
358 { 360 {
359 uint32_t flags = (XFS_TRANS_DIR 361 uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
360 xfs_mount_t *mp = tp->t_mountp; 362 xfs_mount_t *mp = tp->t_mountp;
361 363
362 switch (field) { 364 switch (field) {
363 case XFS_TRANS_SB_ICOUNT: 365 case XFS_TRANS_SB_ICOUNT:
364 tp->t_icount_delta += delta; 366 tp->t_icount_delta += delta;
365 if (xfs_has_lazysbcount(mp)) 367 if (xfs_has_lazysbcount(mp))
366 flags &= ~XFS_TRANS_SB 368 flags &= ~XFS_TRANS_SB_DIRTY;
367 break; 369 break;
368 case XFS_TRANS_SB_IFREE: 370 case XFS_TRANS_SB_IFREE:
369 tp->t_ifree_delta += delta; 371 tp->t_ifree_delta += delta;
370 if (xfs_has_lazysbcount(mp)) 372 if (xfs_has_lazysbcount(mp))
371 flags &= ~XFS_TRANS_SB 373 flags &= ~XFS_TRANS_SB_DIRTY;
372 break; 374 break;
373 case XFS_TRANS_SB_FDBLOCKS: 375 case XFS_TRANS_SB_FDBLOCKS:
374 /* 376 /*
375 * Track the number of blocks 377 * Track the number of blocks allocated in the transaction.
376 * Make sure it does not excee 378 * Make sure it does not exceed the number reserved. If so,
377 * shutdown as this can lead t 379 * shutdown as this can lead to accounting inconsistency.
378 */ 380 */
379 if (delta < 0) { 381 if (delta < 0) {
380 tp->t_blk_res_used += 382 tp->t_blk_res_used += (uint)-delta;
381 if (tp->t_blk_res_used 383 if (tp->t_blk_res_used > tp->t_blk_res)
382 xfs_force_shut 384 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
383 } else if (delta > 0 && (tp->t 385 } else if (delta > 0 && (tp->t_flags & XFS_TRANS_RES_FDBLKS)) {
384 int64_t blkres_delta; 386 int64_t blkres_delta;
385 387
386 /* 388 /*
387 * Return freed blocks 389 * Return freed blocks directly to the reservation
388 * instead of the glob 390 * instead of the global pool, being careful not to
389 * overflow the trans 391 * overflow the trans counter. This is used to preserve
390 * reservation across 392 * reservation across chains of transaction rolls that
391 * repeatedly free and 393 * repeatedly free and allocate blocks.
392 */ 394 */
393 blkres_delta = min_t(i 395 blkres_delta = min_t(int64_t, delta,
394 U 396 UINT_MAX - tp->t_blk_res);
395 tp->t_blk_res += blkre 397 tp->t_blk_res += blkres_delta;
396 delta -= blkres_delta; 398 delta -= blkres_delta;
397 } 399 }
398 tp->t_fdblocks_delta += delta; 400 tp->t_fdblocks_delta += delta;
399 if (xfs_has_lazysbcount(mp)) 401 if (xfs_has_lazysbcount(mp))
400 flags &= ~XFS_TRANS_SB 402 flags &= ~XFS_TRANS_SB_DIRTY;
401 break; 403 break;
402 case XFS_TRANS_SB_RES_FDBLOCKS: 404 case XFS_TRANS_SB_RES_FDBLOCKS:
403 /* 405 /*
404 * The allocation has already 406 * The allocation has already been applied to the
405 * in-core superblock's counte 407 * in-core superblock's counter. This should only
406 * be applied to the on-disk s 408 * be applied to the on-disk superblock.
407 */ 409 */
408 tp->t_res_fdblocks_delta += de 410 tp->t_res_fdblocks_delta += delta;
409 if (xfs_has_lazysbcount(mp)) 411 if (xfs_has_lazysbcount(mp))
410 flags &= ~XFS_TRANS_SB 412 flags &= ~XFS_TRANS_SB_DIRTY;
411 break; 413 break;
412 case XFS_TRANS_SB_FREXTENTS: 414 case XFS_TRANS_SB_FREXTENTS:
413 /* 415 /*
414 * Track the number of blocks 416 * Track the number of blocks allocated in the
415 * transaction. Make sure it 417 * transaction. Make sure it does not exceed the
416 * number reserved. 418 * number reserved.
417 */ 419 */
418 if (delta < 0) { 420 if (delta < 0) {
419 tp->t_rtx_res_used += 421 tp->t_rtx_res_used += (uint)-delta;
420 ASSERT(tp->t_rtx_res_u 422 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
421 } 423 }
422 tp->t_frextents_delta += delta 424 tp->t_frextents_delta += delta;
423 break; 425 break;
424 case XFS_TRANS_SB_RES_FREXTENTS: 426 case XFS_TRANS_SB_RES_FREXTENTS:
425 /* 427 /*
426 * The allocation has already 428 * The allocation has already been applied to the
427 * in-core superblock's counte 429 * in-core superblock's counter. This should only
428 * be applied to the on-disk s 430 * be applied to the on-disk superblock.
429 */ 431 */
430 ASSERT(delta < 0); 432 ASSERT(delta < 0);
431 tp->t_res_frextents_delta += d 433 tp->t_res_frextents_delta += delta;
432 break; 434 break;
433 case XFS_TRANS_SB_DBLOCKS: 435 case XFS_TRANS_SB_DBLOCKS:
434 tp->t_dblocks_delta += delta; 436 tp->t_dblocks_delta += delta;
435 break; 437 break;
436 case XFS_TRANS_SB_AGCOUNT: 438 case XFS_TRANS_SB_AGCOUNT:
437 ASSERT(delta > 0); 439 ASSERT(delta > 0);
438 tp->t_agcount_delta += delta; 440 tp->t_agcount_delta += delta;
439 break; 441 break;
440 case XFS_TRANS_SB_IMAXPCT: 442 case XFS_TRANS_SB_IMAXPCT:
441 tp->t_imaxpct_delta += delta; 443 tp->t_imaxpct_delta += delta;
442 break; 444 break;
443 case XFS_TRANS_SB_REXTSIZE: 445 case XFS_TRANS_SB_REXTSIZE:
444 tp->t_rextsize_delta += delta; 446 tp->t_rextsize_delta += delta;
445 break; 447 break;
446 case XFS_TRANS_SB_RBMBLOCKS: 448 case XFS_TRANS_SB_RBMBLOCKS:
447 tp->t_rbmblocks_delta += delta 449 tp->t_rbmblocks_delta += delta;
448 break; 450 break;
449 case XFS_TRANS_SB_RBLOCKS: 451 case XFS_TRANS_SB_RBLOCKS:
450 tp->t_rblocks_delta += delta; 452 tp->t_rblocks_delta += delta;
451 break; 453 break;
452 case XFS_TRANS_SB_REXTENTS: 454 case XFS_TRANS_SB_REXTENTS:
453 tp->t_rextents_delta += delta; 455 tp->t_rextents_delta += delta;
454 break; 456 break;
455 case XFS_TRANS_SB_REXTSLOG: 457 case XFS_TRANS_SB_REXTSLOG:
456 tp->t_rextslog_delta += delta; 458 tp->t_rextslog_delta += delta;
457 break; 459 break;
458 default: 460 default:
459 ASSERT(0); 461 ASSERT(0);
460 return; 462 return;
461 } 463 }
462 464
463 tp->t_flags |= flags; 465 tp->t_flags |= flags;
464 } 466 }
465 467
466 /* 468 /*
467 * xfs_trans_apply_sb_deltas() is called from 469 * xfs_trans_apply_sb_deltas() is called from the commit code
468 * to bring the superblock buffer into the cur 470 * to bring the superblock buffer into the current transaction
469 * and modify it as requested by earlier calls 471 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
470 * 472 *
471 * For now we just look at each field allowed 473 * For now we just look at each field allowed to change and change
472 * it if necessary. 474 * it if necessary.
473 */ 475 */
474 STATIC void 476 STATIC void
475 xfs_trans_apply_sb_deltas( 477 xfs_trans_apply_sb_deltas(
476 xfs_trans_t *tp) 478 xfs_trans_t *tp)
477 { 479 {
478 struct xfs_dsb *sbp; 480 struct xfs_dsb *sbp;
479 struct xfs_buf *bp; 481 struct xfs_buf *bp;
480 int whole = 0; 482 int whole = 0;
481 483
482 bp = xfs_trans_getsb(tp); 484 bp = xfs_trans_getsb(tp);
483 sbp = bp->b_addr; 485 sbp = bp->b_addr;
484 486
485 /* 487 /*
486 * Only update the superblock counters 488 * Only update the superblock counters if we are logging them
487 */ 489 */
488 if (!xfs_has_lazysbcount((tp->t_mountp 490 if (!xfs_has_lazysbcount((tp->t_mountp))) {
489 if (tp->t_icount_delta) 491 if (tp->t_icount_delta)
490 be64_add_cpu(&sbp->sb_ 492 be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
491 if (tp->t_ifree_delta) 493 if (tp->t_ifree_delta)
492 be64_add_cpu(&sbp->sb_ 494 be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
493 if (tp->t_fdblocks_delta) 495 if (tp->t_fdblocks_delta)
494 be64_add_cpu(&sbp->sb_ 496 be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
495 if (tp->t_res_fdblocks_delta) 497 if (tp->t_res_fdblocks_delta)
496 be64_add_cpu(&sbp->sb_ 498 be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
497 } 499 }
498 500
499 /* !! 501 if (tp->t_frextents_delta)
500 * Updating frextents requires careful !! 502 be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
501 * behave like the lazysb counters bec !! 503 if (tp->t_res_frextents_delta)
502 * recovery in older kenels to recompu !! 504 be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
503 * This means that the ondisk frextent <<
504 * rtbitmap. <<
505 * <<
506 * Therefore, log the frextents change <<
507 * update the incore superblock so tha <<
508 * write the correct value ondisk. <<
509 * <<
510 * Don't touch m_frextents because it <<
511 * and those are handled by the unrese <<
512 */ <<
513 if (tp->t_frextents_delta || tp->t_res <<
514 struct xfs_mount *mp = <<
515 int64_t rtxdel <<
516 <<
517 rtxdelta = tp->t_frextents_del <<
518 <<
519 spin_lock(&mp->m_sb_lock); <<
520 be64_add_cpu(&sbp->sb_frextent <<
521 mp->m_sb.sb_frextents += rtxde <<
522 spin_unlock(&mp->m_sb_lock); <<
523 } <<
524 505
525 if (tp->t_dblocks_delta) { 506 if (tp->t_dblocks_delta) {
526 be64_add_cpu(&sbp->sb_dblocks, 507 be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
527 whole = 1; 508 whole = 1;
528 } 509 }
529 if (tp->t_agcount_delta) { 510 if (tp->t_agcount_delta) {
530 be32_add_cpu(&sbp->sb_agcount, 511 be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
531 whole = 1; 512 whole = 1;
532 } 513 }
533 if (tp->t_imaxpct_delta) { 514 if (tp->t_imaxpct_delta) {
534 sbp->sb_imax_pct += tp->t_imax 515 sbp->sb_imax_pct += tp->t_imaxpct_delta;
535 whole = 1; 516 whole = 1;
536 } 517 }
537 if (tp->t_rextsize_delta) { 518 if (tp->t_rextsize_delta) {
538 be32_add_cpu(&sbp->sb_rextsize 519 be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
539 whole = 1; 520 whole = 1;
540 } 521 }
541 if (tp->t_rbmblocks_delta) { 522 if (tp->t_rbmblocks_delta) {
542 be32_add_cpu(&sbp->sb_rbmblock 523 be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
543 whole = 1; 524 whole = 1;
544 } 525 }
545 if (tp->t_rblocks_delta) { 526 if (tp->t_rblocks_delta) {
546 be64_add_cpu(&sbp->sb_rblocks, 527 be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
547 whole = 1; 528 whole = 1;
548 } 529 }
549 if (tp->t_rextents_delta) { 530 if (tp->t_rextents_delta) {
550 be64_add_cpu(&sbp->sb_rextents 531 be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
551 whole = 1; 532 whole = 1;
552 } 533 }
553 if (tp->t_rextslog_delta) { 534 if (tp->t_rextslog_delta) {
554 sbp->sb_rextslog += tp->t_rext 535 sbp->sb_rextslog += tp->t_rextslog_delta;
555 whole = 1; 536 whole = 1;
556 } 537 }
557 538
558 xfs_trans_buf_set_type(tp, bp, XFS_BLF 539 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
559 if (whole) 540 if (whole)
560 /* 541 /*
561 * Log the whole thing, the fi 542 * Log the whole thing, the fields are noncontiguous.
562 */ 543 */
563 xfs_trans_log_buf(tp, bp, 0, s 544 xfs_trans_log_buf(tp, bp, 0, sizeof(struct xfs_dsb) - 1);
564 else 545 else
565 /* 546 /*
566 * Since all the modifiable fi 547 * Since all the modifiable fields are contiguous, we
567 * can get away with this. 548 * can get away with this.
568 */ 549 */
569 xfs_trans_log_buf(tp, bp, offs 550 xfs_trans_log_buf(tp, bp, offsetof(struct xfs_dsb, sb_icount),
570 offsetof(str 551 offsetof(struct xfs_dsb, sb_frextents) +
571 sizeof(sbp-> 552 sizeof(sbp->sb_frextents) - 1);
572 } 553 }
573 554
574 /* 555 /*
575 * xfs_trans_unreserve_and_mod_sb() is called 556 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations and
576 * apply superblock counter changes to the in- 557 * apply superblock counter changes to the in-core superblock. The
577 * t_res_fdblocks_delta and t_res_frextents_de 558 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
578 * applied to the in-core superblock. The ide 559 * applied to the in-core superblock. The idea is that that has already been
579 * done. 560 * done.
580 * 561 *
581 * If we are not logging superblock counters, 562 * If we are not logging superblock counters, then the inode allocated/free and
582 * used block counts are not updated in the on 563 * used block counts are not updated in the on disk superblock. In this case,
583 * XFS_TRANS_SB_DIRTY will not be set when the 564 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
584 * still need to update the incore superblock 565 * still need to update the incore superblock with the changes.
585 * 566 *
586 * Deltas for the inode count are +/-64, hence 567 * Deltas for the inode count are +/-64, hence we use a large batch size of 128
587 * so we don't need to take the counter lock o 568 * so we don't need to take the counter lock on every update.
588 */ 569 */
589 #define XFS_ICOUNT_BATCH 128 570 #define XFS_ICOUNT_BATCH 128
590 571
591 void 572 void
592 xfs_trans_unreserve_and_mod_sb( 573 xfs_trans_unreserve_and_mod_sb(
593 struct xfs_trans *tp) 574 struct xfs_trans *tp)
594 { 575 {
595 struct xfs_mount *mp = tp->t_mo 576 struct xfs_mount *mp = tp->t_mountp;
596 int64_t blkdelta = tp- !! 577 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
597 int64_t rtxdelta = tp- !! 578 int64_t blkdelta = 0;
>> 579 int64_t rtxdelta = 0;
598 int64_t idelta = 0; 580 int64_t idelta = 0;
599 int64_t ifreedelta = 0 581 int64_t ifreedelta = 0;
>> 582 int error;
600 583
601 /* !! 584 /* calculate deltas */
602 * Calculate the deltas. !! 585 if (tp->t_blk_res > 0)
603 * !! 586 blkdelta = tp->t_blk_res;
604 * t_fdblocks_delta and t_frextents_de !! 587 if ((tp->t_fdblocks_delta != 0) &&
605 * !! 588 (xfs_has_lazysbcount(mp) ||
606 * - positive values indicate blocks !! 589 (tp->t_flags & XFS_TRANS_SB_DIRTY)))
607 * - negative values indicate blocks <<
608 * <<
609 * Negative values can only happen if <<
610 * reservation that covers the allocat <<
611 * that the calculated delta values mu <<
612 * can only put back previous allocate <<
613 */ <<
614 ASSERT(tp->t_blk_res || tp->t_fdblocks <<
615 if (xfs_has_lazysbcount(mp) || (tp->t_ <<
616 blkdelta += tp->t_fdblocks_del 590 blkdelta += tp->t_fdblocks_delta;
617 ASSERT(blkdelta >= 0); <<
618 } <<
619 591
620 ASSERT(tp->t_rtx_res || tp->t_frextent !! 592 if (tp->t_rtx_res > 0)
621 if (tp->t_flags & XFS_TRANS_SB_DIRTY) !! 593 rtxdelta = tp->t_rtx_res;
>> 594 if ((tp->t_frextents_delta != 0) &&
>> 595 (tp->t_flags & XFS_TRANS_SB_DIRTY))
622 rtxdelta += tp->t_frextents_de 596 rtxdelta += tp->t_frextents_delta;
623 ASSERT(rtxdelta >= 0); <<
624 } <<
625 597
626 if (xfs_has_lazysbcount(mp) || (tp->t_ !! 598 if (xfs_has_lazysbcount(mp) ||
>> 599 (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
627 idelta = tp->t_icount_delta; 600 idelta = tp->t_icount_delta;
628 ifreedelta = tp->t_ifree_delta 601 ifreedelta = tp->t_ifree_delta;
629 } 602 }
630 603
631 /* apply the per-cpu counters */ 604 /* apply the per-cpu counters */
632 if (blkdelta) !! 605 if (blkdelta) {
633 xfs_add_fdblocks(mp, blkdelta) !! 606 error = xfs_mod_fdblocks(mp, blkdelta, rsvd);
>> 607 ASSERT(!error);
>> 608 }
634 609
635 if (idelta) 610 if (idelta)
636 percpu_counter_add_batch(&mp-> 611 percpu_counter_add_batch(&mp->m_icount, idelta,
637 XFS_I 612 XFS_ICOUNT_BATCH);
638 613
639 if (ifreedelta) 614 if (ifreedelta)
640 percpu_counter_add(&mp->m_ifre 615 percpu_counter_add(&mp->m_ifree, ifreedelta);
641 616
642 if (rtxdelta) !! 617 if (rtxdelta == 0 && !(tp->t_flags & XFS_TRANS_SB_DIRTY))
643 xfs_add_frextents(mp, rtxdelta <<
644 <<
645 if (!(tp->t_flags & XFS_TRANS_SB_DIRTY <<
646 return; 618 return;
647 619
648 /* apply remaining deltas */ 620 /* apply remaining deltas */
649 spin_lock(&mp->m_sb_lock); 621 spin_lock(&mp->m_sb_lock);
650 mp->m_sb.sb_fdblocks += tp->t_fdblocks 622 mp->m_sb.sb_fdblocks += tp->t_fdblocks_delta + tp->t_res_fdblocks_delta;
651 mp->m_sb.sb_icount += idelta; 623 mp->m_sb.sb_icount += idelta;
652 mp->m_sb.sb_ifree += ifreedelta; 624 mp->m_sb.sb_ifree += ifreedelta;
653 /* !! 625 mp->m_sb.sb_frextents += rtxdelta;
654 * Do not touch sb_frextents here beca <<
655 * reservation. sb_frextents is not p <<
656 * must be consistent with the ondisk <<
657 * incore reservations. <<
658 */ <<
659 mp->m_sb.sb_dblocks += tp->t_dblocks_d 626 mp->m_sb.sb_dblocks += tp->t_dblocks_delta;
660 mp->m_sb.sb_agcount += tp->t_agcount_d 627 mp->m_sb.sb_agcount += tp->t_agcount_delta;
661 mp->m_sb.sb_imax_pct += tp->t_imaxpct_ 628 mp->m_sb.sb_imax_pct += tp->t_imaxpct_delta;
662 mp->m_sb.sb_rextsize += tp->t_rextsize 629 mp->m_sb.sb_rextsize += tp->t_rextsize_delta;
663 if (tp->t_rextsize_delta) { <<
664 mp->m_rtxblklog = log2_if_powe <<
665 mp->m_rtxblkmask = mask64_if_p <<
666 } <<
667 mp->m_sb.sb_rbmblocks += tp->t_rbmbloc 630 mp->m_sb.sb_rbmblocks += tp->t_rbmblocks_delta;
668 mp->m_sb.sb_rblocks += tp->t_rblocks_d 631 mp->m_sb.sb_rblocks += tp->t_rblocks_delta;
669 mp->m_sb.sb_rextents += tp->t_rextents 632 mp->m_sb.sb_rextents += tp->t_rextents_delta;
670 mp->m_sb.sb_rextslog += tp->t_rextslog 633 mp->m_sb.sb_rextslog += tp->t_rextslog_delta;
671 spin_unlock(&mp->m_sb_lock); 634 spin_unlock(&mp->m_sb_lock);
672 635
673 /* 636 /*
674 * Debug checks outside of the spinloc 637 * Debug checks outside of the spinlock so they don't lock up the
675 * machine if they fail. 638 * machine if they fail.
676 */ 639 */
677 ASSERT(mp->m_sb.sb_imax_pct >= 0); 640 ASSERT(mp->m_sb.sb_imax_pct >= 0);
678 ASSERT(mp->m_sb.sb_rextslog >= 0); 641 ASSERT(mp->m_sb.sb_rextslog >= 0);
>> 642 return;
679 } 643 }
680 644
681 /* Add the given log item to the transaction's 645 /* Add the given log item to the transaction's list of log items. */
682 void 646 void
683 xfs_trans_add_item( 647 xfs_trans_add_item(
684 struct xfs_trans *tp, 648 struct xfs_trans *tp,
685 struct xfs_log_item *lip) 649 struct xfs_log_item *lip)
686 { 650 {
687 ASSERT(lip->li_log == tp->t_mountp->m_ !! 651 ASSERT(lip->li_mountp == tp->t_mountp);
688 ASSERT(lip->li_ailp == tp->t_mountp->m 652 ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
689 ASSERT(list_empty(&lip->li_trans)); 653 ASSERT(list_empty(&lip->li_trans));
690 ASSERT(!test_bit(XFS_LI_DIRTY, &lip->l 654 ASSERT(!test_bit(XFS_LI_DIRTY, &lip->li_flags));
691 655
692 list_add_tail(&lip->li_trans, &tp->t_i 656 list_add_tail(&lip->li_trans, &tp->t_items);
693 trace_xfs_trans_add_item(tp, _RET_IP_) 657 trace_xfs_trans_add_item(tp, _RET_IP_);
694 } 658 }
695 659
696 /* 660 /*
697 * Unlink the log item from the transaction. t 661 * Unlink the log item from the transaction. the log item is no longer
698 * considered dirty in this transaction, as th 662 * considered dirty in this transaction, as the linked transaction has
699 * finished, either by abort or commit complet 663 * finished, either by abort or commit completion.
700 */ 664 */
701 void 665 void
702 xfs_trans_del_item( 666 xfs_trans_del_item(
703 struct xfs_log_item *lip) 667 struct xfs_log_item *lip)
704 { 668 {
705 clear_bit(XFS_LI_DIRTY, &lip->li_flags 669 clear_bit(XFS_LI_DIRTY, &lip->li_flags);
706 list_del_init(&lip->li_trans); 670 list_del_init(&lip->li_trans);
707 } 671 }
708 672
709 /* Detach and unlock all of the items in a tra 673 /* Detach and unlock all of the items in a transaction */
710 static void 674 static void
711 xfs_trans_free_items( 675 xfs_trans_free_items(
712 struct xfs_trans *tp, 676 struct xfs_trans *tp,
713 bool abort) 677 bool abort)
714 { 678 {
715 struct xfs_log_item *lip, *next; 679 struct xfs_log_item *lip, *next;
716 680
717 trace_xfs_trans_free_items(tp, _RET_IP 681 trace_xfs_trans_free_items(tp, _RET_IP_);
718 682
719 list_for_each_entry_safe(lip, next, &t 683 list_for_each_entry_safe(lip, next, &tp->t_items, li_trans) {
720 xfs_trans_del_item(lip); 684 xfs_trans_del_item(lip);
721 if (abort) 685 if (abort)
722 set_bit(XFS_LI_ABORTED 686 set_bit(XFS_LI_ABORTED, &lip->li_flags);
723 if (lip->li_ops->iop_release) 687 if (lip->li_ops->iop_release)
724 lip->li_ops->iop_relea 688 lip->li_ops->iop_release(lip);
725 } 689 }
726 } 690 }
727 691
728 /* !! 692 static inline void
729 * Sort transaction items prior to running pre !! 693 xfs_log_item_batch_insert(
730 * attempt to order the items such that they w !! 694 struct xfs_ail *ailp,
731 * order. Items that have no sort function are !! 695 struct xfs_ail_cursor *cur,
732 * and so are locked last. !! 696 struct xfs_log_item **log_items,
733 * !! 697 int nr_items,
734 * This may need refinement as different types !! 698 xfs_lsn_t commit_lsn)
735 * !! 699 {
736 * Function is more complex than it needs to b !! 700 int i;
737 * values and the function only returns 32 bit !! 701
738 */ !! 702 spin_lock(&ailp->ail_lock);
739 static int !! 703 /* xfs_trans_ail_update_bulk drops ailp->ail_lock */
740 xfs_trans_precommit_sort( !! 704 xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
741 void *unused_arg, <<
742 const struct list_head *a, <<
743 const struct list_head *b) <<
744 { <<
745 struct xfs_log_item *lia = contain <<
746 struct <<
747 struct xfs_log_item *lib = contain <<
748 struct <<
749 int64_t diff; <<
750 705
751 /* !! 706 for (i = 0; i < nr_items; i++) {
752 * If both items are non-sortable, lea !! 707 struct xfs_log_item *lip = log_items[i];
753 * sortable, move the non-sortable ite !! 708
754 */ !! 709 if (lip->li_ops->iop_unpin)
755 if (!lia->li_ops->iop_sort && !lib->li !! 710 lip->li_ops->iop_unpin(lip, 0);
756 return 0; !! 711 }
757 if (!lia->li_ops->iop_sort) <<
758 return 1; <<
759 if (!lib->li_ops->iop_sort) <<
760 return -1; <<
761 <<
762 diff = lia->li_ops->iop_sort(lia) - li <<
763 if (diff < 0) <<
764 return -1; <<
765 if (diff > 0) <<
766 return 1; <<
767 return 0; <<
768 } 712 }
769 713
770 /* 714 /*
771 * Run transaction precommit functions. !! 715 * Bulk operation version of xfs_trans_committed that takes a log vector of
>> 716 * items to insert into the AIL. This uses bulk AIL insertion techniques to
>> 717 * minimise lock traffic.
>> 718 *
>> 719 * If we are called with the aborted flag set, it is because a log write during
>> 720 * a CIL checkpoint commit has failed. In this case, all the items in the
>> 721 * checkpoint have already gone through iop_committed and iop_committing, which
>> 722 * means that checkpoint commit abort handling is treated exactly the same
>> 723 * as an iclog write error even though we haven't started any IO yet. Hence in
>> 724 * this case all we need to do is iop_committed processing, followed by an
>> 725 * iop_unpin(aborted) call.
772 * 726 *
773 * If there is an error in any of the callouts !! 727 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
774 * trigger a shutdown to abort the transaction !! 728 * at the end of the AIL, the insert cursor avoids the need to walk
775 * from errors at this point as the transactio !! 729 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
>> 730 * call. This saves a lot of needless list walking and is a net win, even
>> 731 * though it slightly increases that amount of AIL lock traffic to set it up
>> 732 * and tear it down.
776 */ 733 */
777 static int !! 734 void
778 xfs_trans_run_precommits( !! 735 xfs_trans_committed_bulk(
779 struct xfs_trans *tp) !! 736 struct xfs_ail *ailp,
780 { !! 737 struct xfs_log_vec *log_vector,
781 struct xfs_mount *mp = tp->t_mo !! 738 xfs_lsn_t commit_lsn,
782 struct xfs_log_item *lip, *n; !! 739 bool aborted)
783 int error = 0; !! 740 {
>> 741 #define LOG_ITEM_BATCH_SIZE 32
>> 742 struct xfs_log_item *log_items[LOG_ITEM_BATCH_SIZE];
>> 743 struct xfs_log_vec *lv;
>> 744 struct xfs_ail_cursor cur;
>> 745 int i = 0;
>> 746
>> 747 spin_lock(&ailp->ail_lock);
>> 748 xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
>> 749 spin_unlock(&ailp->ail_lock);
>> 750
>> 751 /* unpin all the log items */
>> 752 for (lv = log_vector; lv; lv = lv->lv_next ) {
>> 753 struct xfs_log_item *lip = lv->lv_item;
>> 754 xfs_lsn_t item_lsn;
784 755
785 /* !! 756 if (aborted)
786 * Sort the item list to avoid ABBA de !! 757 set_bit(XFS_LI_ABORTED, &lip->li_flags);
787 * running precommit operations that l <<
788 * inode cluster buffers. <<
789 */ <<
790 list_sort(NULL, &tp->t_items, xfs_tran <<
791 758
792 /* !! 759 if (lip->li_ops->flags & XFS_ITEM_RELEASE_WHEN_COMMITTED) {
793 * Precommit operations can remove the !! 760 lip->li_ops->iop_release(lip);
794 * if the log item exists purely to de <<
795 * can be ordered against other operat <<
796 * list_for_each_entry_safe() here. <<
797 */ <<
798 list_for_each_entry_safe(lip, n, &tp-> <<
799 if (!test_bit(XFS_LI_DIRTY, &l <<
800 continue; 761 continue;
801 if (lip->li_ops->iop_precommit !! 762 }
802 error = lip->li_ops->i !! 763
803 if (error) !! 764 if (lip->li_ops->iop_committed)
804 break; !! 765 item_lsn = lip->li_ops->iop_committed(lip, commit_lsn);
>> 766 else
>> 767 item_lsn = commit_lsn;
>> 768
>> 769 /* item_lsn of -1 means the item needs no further processing */
>> 770 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
>> 771 continue;
>> 772
>> 773 /*
>> 774 * if we are aborting the operation, no point in inserting the
>> 775 * object into the AIL as we are in a shutdown situation.
>> 776 */
>> 777 if (aborted) {
>> 778 ASSERT(xfs_is_shutdown(ailp->ail_mount));
>> 779 if (lip->li_ops->iop_unpin)
>> 780 lip->li_ops->iop_unpin(lip, 1);
>> 781 continue;
>> 782 }
>> 783
>> 784 if (item_lsn != commit_lsn) {
>> 785
>> 786 /*
>> 787 * Not a bulk update option due to unusual item_lsn.
>> 788 * Push into AIL immediately, rechecking the lsn once
>> 789 * we have the ail lock. Then unpin the item. This does
>> 790 * not affect the AIL cursor the bulk insert path is
>> 791 * using.
>> 792 */
>> 793 spin_lock(&ailp->ail_lock);
>> 794 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
>> 795 xfs_trans_ail_update(ailp, lip, item_lsn);
>> 796 else
>> 797 spin_unlock(&ailp->ail_lock);
>> 798 if (lip->li_ops->iop_unpin)
>> 799 lip->li_ops->iop_unpin(lip, 0);
>> 800 continue;
>> 801 }
>> 802
>> 803 /* Item is a candidate for bulk AIL insert. */
>> 804 log_items[i++] = lv->lv_item;
>> 805 if (i >= LOG_ITEM_BATCH_SIZE) {
>> 806 xfs_log_item_batch_insert(ailp, &cur, log_items,
>> 807 LOG_ITEM_BATCH_SIZE, commit_lsn);
>> 808 i = 0;
805 } 809 }
806 } 810 }
807 if (error) !! 811
808 xfs_force_shutdown(mp, SHUTDOW !! 812 /* make sure we insert the remainder! */
809 return error; !! 813 if (i)
>> 814 xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
>> 815
>> 816 spin_lock(&ailp->ail_lock);
>> 817 xfs_trans_ail_cursor_done(&cur);
>> 818 spin_unlock(&ailp->ail_lock);
810 } 819 }
811 820
812 /* 821 /*
813 * Commit the given transaction to the log. 822 * Commit the given transaction to the log.
814 * 823 *
815 * XFS disk error handling mechanism is not ba 824 * XFS disk error handling mechanism is not based on a typical
816 * transaction abort mechanism. Logically afte 825 * transaction abort mechanism. Logically after the filesystem
817 * gets marked 'SHUTDOWN', we can't let any ne 826 * gets marked 'SHUTDOWN', we can't let any new transactions
818 * be durable - ie. committed to disk - becaus 827 * be durable - ie. committed to disk - because some metadata might
819 * be inconsistent. In such cases, this return 828 * be inconsistent. In such cases, this returns an error, and the
820 * caller may assume that all locked objects j 829 * caller may assume that all locked objects joined to the transaction
821 * have already been unlocked as if the commit 830 * have already been unlocked as if the commit had succeeded.
822 * Do not reference the transaction structure 831 * Do not reference the transaction structure after this call.
823 */ 832 */
824 static int 833 static int
825 __xfs_trans_commit( 834 __xfs_trans_commit(
826 struct xfs_trans *tp, 835 struct xfs_trans *tp,
827 bool regrant) 836 bool regrant)
828 { 837 {
829 struct xfs_mount *mp = tp->t_mo 838 struct xfs_mount *mp = tp->t_mountp;
830 struct xlog *log = mp->m_l <<
831 xfs_csn_t commit_seq = 0 839 xfs_csn_t commit_seq = 0;
832 int error = 0; 840 int error = 0;
833 int sync = tp->t_f 841 int sync = tp->t_flags & XFS_TRANS_SYNC;
834 842
835 trace_xfs_trans_commit(tp, _RET_IP_); 843 trace_xfs_trans_commit(tp, _RET_IP_);
836 844
837 error = xfs_trans_run_precommits(tp); <<
838 if (error) { <<
839 if (tp->t_flags & XFS_TRANS_PE <<
840 xfs_defer_cancel(tp); <<
841 goto out_unreserve; <<
842 } <<
843 <<
844 /* 845 /*
845 * Finish deferred items on final comm 846 * Finish deferred items on final commit. Only permanent transactions
846 * should ever have deferred ops. 847 * should ever have deferred ops.
847 */ 848 */
848 WARN_ON_ONCE(!list_empty(&tp->t_dfops) 849 WARN_ON_ONCE(!list_empty(&tp->t_dfops) &&
849 !(tp->t_flags & XFS_TRANS 850 !(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
850 if (!regrant && (tp->t_flags & XFS_TRA 851 if (!regrant && (tp->t_flags & XFS_TRANS_PERM_LOG_RES)) {
851 error = xfs_defer_finish_norol 852 error = xfs_defer_finish_noroll(&tp);
852 if (error) 853 if (error)
853 goto out_unreserve; 854 goto out_unreserve;
854 <<
855 /* Run precommits from final t <<
856 error = xfs_trans_run_precommi <<
857 if (error) <<
858 goto out_unreserve; <<
859 } 855 }
860 856
861 /* 857 /*
862 * If there is nothing to be logged by 858 * If there is nothing to be logged by the transaction,
863 * then unlock all of the items associ 859 * then unlock all of the items associated with the
864 * transaction and free the transactio 860 * transaction and free the transaction structure.
865 * Also make sure to return any reserv 861 * Also make sure to return any reserved blocks to
866 * the free pool. 862 * the free pool.
867 */ 863 */
868 if (!(tp->t_flags & XFS_TRANS_DIRTY)) 864 if (!(tp->t_flags & XFS_TRANS_DIRTY))
869 goto out_unreserve; 865 goto out_unreserve;
870 866
871 /* !! 867 if (xfs_is_shutdown(mp)) {
872 * We must check against log shutdown <<
873 * items and leave them dirty, inconsi <<
874 * the log is active. This leaves them <<
875 * disk, and that will lead to on-disk <<
876 */ <<
877 if (xlog_is_shutdown(log)) { <<
878 error = -EIO; 868 error = -EIO;
879 goto out_unreserve; 869 goto out_unreserve;
880 } 870 }
881 871
882 ASSERT(tp->t_ticket != NULL); 872 ASSERT(tp->t_ticket != NULL);
883 873
884 /* 874 /*
885 * If we need to update the superblock 875 * If we need to update the superblock, then do it now.
886 */ 876 */
887 if (tp->t_flags & XFS_TRANS_SB_DIRTY) 877 if (tp->t_flags & XFS_TRANS_SB_DIRTY)
888 xfs_trans_apply_sb_deltas(tp); 878 xfs_trans_apply_sb_deltas(tp);
889 xfs_trans_apply_dquot_deltas(tp); 879 xfs_trans_apply_dquot_deltas(tp);
890 880
891 xlog_cil_commit(log, tp, &commit_seq, !! 881 xlog_cil_commit(mp->m_log, tp, &commit_seq, regrant);
892 882
893 xfs_trans_free(tp); 883 xfs_trans_free(tp);
894 884
895 /* 885 /*
896 * If the transaction needs to be sync 886 * If the transaction needs to be synchronous, then force the
897 * log out now and wait for it. 887 * log out now and wait for it.
898 */ 888 */
899 if (sync) { 889 if (sync) {
900 error = xfs_log_force_seq(mp, 890 error = xfs_log_force_seq(mp, commit_seq, XFS_LOG_SYNC, NULL);
901 XFS_STATS_INC(mp, xs_trans_syn 891 XFS_STATS_INC(mp, xs_trans_sync);
902 } else { 892 } else {
903 XFS_STATS_INC(mp, xs_trans_asy 893 XFS_STATS_INC(mp, xs_trans_async);
904 } 894 }
905 895
906 return error; 896 return error;
907 897
908 out_unreserve: 898 out_unreserve:
909 xfs_trans_unreserve_and_mod_sb(tp); 899 xfs_trans_unreserve_and_mod_sb(tp);
910 900
911 /* 901 /*
912 * It is indeed possible for the trans 902 * It is indeed possible for the transaction to be not dirty but
913 * the dqinfo portion to be. All that 903 * the dqinfo portion to be. All that means is that we have some
914 * (non-persistent) quota reservations 904 * (non-persistent) quota reservations that need to be unreserved.
915 */ 905 */
916 xfs_trans_unreserve_and_mod_dquots(tp) 906 xfs_trans_unreserve_and_mod_dquots(tp);
917 if (tp->t_ticket) { 907 if (tp->t_ticket) {
918 if (regrant && !xlog_is_shutdo !! 908 if (regrant && !xlog_is_shutdown(mp->m_log))
919 xfs_log_ticket_regrant !! 909 xfs_log_ticket_regrant(mp->m_log, tp->t_ticket);
920 else 910 else
921 xfs_log_ticket_ungrant !! 911 xfs_log_ticket_ungrant(mp->m_log, tp->t_ticket);
922 tp->t_ticket = NULL; 912 tp->t_ticket = NULL;
923 } 913 }
924 xfs_trans_free_items(tp, !!error); 914 xfs_trans_free_items(tp, !!error);
925 xfs_trans_free(tp); 915 xfs_trans_free(tp);
926 916
927 XFS_STATS_INC(mp, xs_trans_empty); 917 XFS_STATS_INC(mp, xs_trans_empty);
928 return error; 918 return error;
929 } 919 }
930 920
931 int 921 int
932 xfs_trans_commit( 922 xfs_trans_commit(
933 struct xfs_trans *tp) 923 struct xfs_trans *tp)
934 { 924 {
935 return __xfs_trans_commit(tp, false); 925 return __xfs_trans_commit(tp, false);
936 } 926 }
937 927
938 /* 928 /*
939 * Unlock all of the transaction's items and f !! 929 * Unlock all of the transaction's items and free the transaction.
940 * transaction is dirty, we must shut down the !! 930 * The transaction must not have modified any of its items, because
941 * way to restore them to their previous state !! 931 * there is no way to restore them to their previous state.
942 * <<
943 * If the transaction has made a log reservati <<
944 * well. <<
945 * 932 *
946 * This is a high level function (equivalent t !! 933 * If the transaction has made a log reservation, make sure to release
947 * be called after the transaction has effecti !! 934 * it as well.
948 * being shut down. However, if the mount has <<
949 * transaction is dirty we will shut the mount <<
950 * guarantees that the log is shut down, too. <<
951 * careful with shutdown state and dirty items <<
952 * xfs_trans_commit(). <<
953 */ 935 */
954 void 936 void
955 xfs_trans_cancel( 937 xfs_trans_cancel(
956 struct xfs_trans *tp) 938 struct xfs_trans *tp)
957 { 939 {
958 struct xfs_mount *mp = tp->t_mo 940 struct xfs_mount *mp = tp->t_mountp;
959 struct xlog *log = mp->m_l <<
960 bool dirty = (tp->t 941 bool dirty = (tp->t_flags & XFS_TRANS_DIRTY);
961 942
962 trace_xfs_trans_cancel(tp, _RET_IP_); 943 trace_xfs_trans_cancel(tp, _RET_IP_);
963 944
964 /* 945 /*
965 * It's never valid to cancel a transa 946 * It's never valid to cancel a transaction with deferred ops attached,
966 * because the transaction is effectiv 947 * because the transaction is effectively dirty. Complain about this
967 * loudly before freeing the in-memory !! 948 * loudly before freeing the in-memory defer items.
968 * filesystem. <<
969 */ 949 */
970 if (!list_empty(&tp->t_dfops)) { 950 if (!list_empty(&tp->t_dfops)) {
>> 951 ASSERT(xfs_is_shutdown(mp) || list_empty(&tp->t_dfops));
971 ASSERT(tp->t_flags & XFS_TRANS 952 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
972 dirty = true; 953 dirty = true;
973 xfs_defer_cancel(tp); 954 xfs_defer_cancel(tp);
974 } 955 }
975 956
976 /* 957 /*
977 * See if the caller is relying on us !! 958 * See if the caller is relying on us to shut down the
978 * only want an error report if there !! 959 * filesystem. This happens in paths where we detect
979 * progress, so we only need to check !! 960 * corruption and decide to give up.
980 * here. <<
981 */ 961 */
982 if (dirty && !xfs_is_shutdown(mp)) { 962 if (dirty && !xfs_is_shutdown(mp)) {
983 XFS_ERROR_REPORT("xfs_trans_ca 963 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
984 xfs_force_shutdown(mp, SHUTDOW 964 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
985 } 965 }
986 #ifdef DEBUG 966 #ifdef DEBUG
987 /* Log items need to be consistent unt !! 967 if (!dirty && !xfs_is_shutdown(mp)) {
988 if (!dirty && !xlog_is_shutdown(log)) <<
989 struct xfs_log_item *lip; 968 struct xfs_log_item *lip;
990 969
991 list_for_each_entry(lip, &tp-> 970 list_for_each_entry(lip, &tp->t_items, li_trans)
992 ASSERT(!xlog_item_is_i 971 ASSERT(!xlog_item_is_intent_done(lip));
993 } 972 }
994 #endif 973 #endif
995 xfs_trans_unreserve_and_mod_sb(tp); 974 xfs_trans_unreserve_and_mod_sb(tp);
996 xfs_trans_unreserve_and_mod_dquots(tp) 975 xfs_trans_unreserve_and_mod_dquots(tp);
997 976
998 if (tp->t_ticket) { 977 if (tp->t_ticket) {
999 xfs_log_ticket_ungrant(log, tp !! 978 xfs_log_ticket_ungrant(mp->m_log, tp->t_ticket);
1000 tp->t_ticket = NULL; 979 tp->t_ticket = NULL;
1001 } 980 }
1002 981
1003 xfs_trans_free_items(tp, dirty); 982 xfs_trans_free_items(tp, dirty);
1004 xfs_trans_free(tp); 983 xfs_trans_free(tp);
1005 } 984 }
1006 985
1007 /* 986 /*
1008 * Roll from one trans in the sequence of PER 987 * Roll from one trans in the sequence of PERMANENT transactions to
1009 * the next: permanent transactions are only 988 * the next: permanent transactions are only flushed out when
1010 * committed with xfs_trans_commit(), but we 989 * committed with xfs_trans_commit(), but we still want as soon
1011 * as possible to let chunks of it go to the 990 * as possible to let chunks of it go to the log. So we commit the
1012 * chunk we've been working on and get a new 991 * chunk we've been working on and get a new transaction to continue.
1013 */ 992 */
1014 int 993 int
1015 xfs_trans_roll( 994 xfs_trans_roll(
1016 struct xfs_trans **tpp) 995 struct xfs_trans **tpp)
1017 { 996 {
1018 struct xfs_trans *trans = *tpp 997 struct xfs_trans *trans = *tpp;
1019 struct xfs_trans_res tres; 998 struct xfs_trans_res tres;
1020 int error; 999 int error;
1021 1000
1022 trace_xfs_trans_roll(trans, _RET_IP_) 1001 trace_xfs_trans_roll(trans, _RET_IP_);
1023 1002
1024 /* 1003 /*
1025 * Copy the critical parameters from 1004 * Copy the critical parameters from one trans to the next.
1026 */ 1005 */
1027 tres.tr_logres = trans->t_log_res; 1006 tres.tr_logres = trans->t_log_res;
1028 tres.tr_logcount = trans->t_log_count 1007 tres.tr_logcount = trans->t_log_count;
1029 1008
1030 *tpp = xfs_trans_dup(trans); 1009 *tpp = xfs_trans_dup(trans);
1031 1010
1032 /* 1011 /*
1033 * Commit the current transaction. 1012 * Commit the current transaction.
1034 * If this commit failed, then it'd j 1013 * If this commit failed, then it'd just unlock those items that
1035 * are not marked ihold. That also me 1014 * are not marked ihold. That also means that a filesystem shutdown
1036 * is in progress. The caller takes t 1015 * is in progress. The caller takes the responsibility to cancel
1037 * the duplicate transaction that get 1016 * the duplicate transaction that gets returned.
1038 */ 1017 */
1039 error = __xfs_trans_commit(trans, tru 1018 error = __xfs_trans_commit(trans, true);
1040 if (error) 1019 if (error)
1041 return error; 1020 return error;
1042 1021
1043 /* 1022 /*
1044 * Reserve space in the log for the n 1023 * Reserve space in the log for the next transaction.
1045 * This also pushes items in the "AIL 1024 * This also pushes items in the "AIL", the list of logged items,
1046 * out to disk if they are taking up 1025 * out to disk if they are taking up space at the tail of the log
1047 * that we want to use. This require 1026 * that we want to use. This requires that either nothing be locked
1048 * across this call, or that anything 1027 * across this call, or that anything that is locked be logged in
1049 * the prior and the next transaction 1028 * the prior and the next transactions.
1050 */ 1029 */
1051 tres.tr_logflags = XFS_TRANS_PERM_LOG 1030 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1052 return xfs_trans_reserve(*tpp, &tres, 1031 return xfs_trans_reserve(*tpp, &tres, 0, 0);
1053 } 1032 }
1054 1033
1055 /* 1034 /*
1056 * Allocate an transaction, lock and join the 1035 * Allocate an transaction, lock and join the inode to it, and reserve quota.
1057 * 1036 *
1058 * The caller must ensure that the on-disk dq 1037 * The caller must ensure that the on-disk dquots attached to this inode have
1059 * already been allocated and initialized. T 1038 * already been allocated and initialized. The caller is responsible for
1060 * releasing ILOCK_EXCL if a new transaction 1039 * releasing ILOCK_EXCL if a new transaction is returned.
1061 */ 1040 */
1062 int 1041 int
1063 xfs_trans_alloc_inode( 1042 xfs_trans_alloc_inode(
1064 struct xfs_inode *ip, 1043 struct xfs_inode *ip,
1065 struct xfs_trans_res *resv, 1044 struct xfs_trans_res *resv,
1066 unsigned int dblocks, 1045 unsigned int dblocks,
1067 unsigned int rblocks, 1046 unsigned int rblocks,
1068 bool force, 1047 bool force,
1069 struct xfs_trans **tpp) 1048 struct xfs_trans **tpp)
1070 { 1049 {
1071 struct xfs_trans *tp; 1050 struct xfs_trans *tp;
1072 struct xfs_mount *mp = ip->i_m 1051 struct xfs_mount *mp = ip->i_mount;
1073 bool retried = fal 1052 bool retried = false;
1074 int error; 1053 int error;
1075 1054
1076 retry: 1055 retry:
1077 error = xfs_trans_alloc(mp, resv, dbl 1056 error = xfs_trans_alloc(mp, resv, dblocks,
1078 xfs_extlen_to_rtxlen( !! 1057 rblocks / mp->m_sb.sb_rextsize,
1079 force ? XFS_TRANS_RES 1058 force ? XFS_TRANS_RESERVE : 0, &tp);
1080 if (error) 1059 if (error)
1081 return error; 1060 return error;
1082 1061
1083 xfs_ilock(ip, XFS_ILOCK_EXCL); 1062 xfs_ilock(ip, XFS_ILOCK_EXCL);
1084 xfs_trans_ijoin(tp, ip, 0); 1063 xfs_trans_ijoin(tp, ip, 0);
1085 1064
1086 error = xfs_qm_dqattach_locked(ip, fa 1065 error = xfs_qm_dqattach_locked(ip, false);
1087 if (error) { 1066 if (error) {
1088 /* Caller should have allocat 1067 /* Caller should have allocated the dquots! */
1089 ASSERT(error != -ENOENT); 1068 ASSERT(error != -ENOENT);
1090 goto out_cancel; 1069 goto out_cancel;
1091 } 1070 }
1092 1071
1093 error = xfs_trans_reserve_quota_nblks 1072 error = xfs_trans_reserve_quota_nblks(tp, ip, dblocks, rblocks, force);
1094 if ((error == -EDQUOT || error == -EN 1073 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1095 xfs_trans_cancel(tp); 1074 xfs_trans_cancel(tp);
1096 xfs_iunlock(ip, XFS_ILOCK_EXC 1075 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1097 xfs_blockgc_free_quota(ip, 0) 1076 xfs_blockgc_free_quota(ip, 0);
1098 retried = true; 1077 retried = true;
1099 goto retry; 1078 goto retry;
1100 } 1079 }
1101 if (error) 1080 if (error)
1102 goto out_cancel; 1081 goto out_cancel;
1103 1082
1104 *tpp = tp; 1083 *tpp = tp;
1105 return 0; 1084 return 0;
1106 1085
1107 out_cancel: 1086 out_cancel:
1108 xfs_trans_cancel(tp); 1087 xfs_trans_cancel(tp);
1109 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1088 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1110 return error; 1089 return error;
1111 } 1090 }
1112 1091
1113 /* 1092 /*
1114 * Try to reserve more blocks for a transacti <<
1115 * <<
1116 * This is for callers that need to attach re <<
1117 * those resources to determine the space res <<
1118 * modify the attached resources. In other w <<
1119 * fail due to ENOSPC, so the caller must be <<
1120 * without shutting down the fs. <<
1121 */ <<
1122 int <<
1123 xfs_trans_reserve_more( <<
1124 struct xfs_trans *tp, <<
1125 unsigned int blocks, <<
1126 unsigned int rtextents) <<
1127 { <<
1128 struct xfs_trans_res resv = { }; <<
1129 <<
1130 return xfs_trans_reserve(tp, &resv, b <<
1131 } <<
1132 <<
1133 /* <<
1134 * Try to reserve more blocks and file quota <<
1135 * conditions of usage as xfs_trans_reserve_m <<
1136 */ <<
1137 int <<
1138 xfs_trans_reserve_more_inode( <<
1139 struct xfs_trans *tp, <<
1140 struct xfs_inode *ip, <<
1141 unsigned int dblocks, <<
1142 unsigned int rblocks, <<
1143 bool force_quota) <<
1144 { <<
1145 struct xfs_trans_res resv = { }; <<
1146 struct xfs_mount *mp = ip->i_m <<
1147 unsigned int rtx = xfs_ext <<
1148 int error; <<
1149 <<
1150 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL <<
1151 <<
1152 error = xfs_trans_reserve(tp, &resv, <<
1153 if (error) <<
1154 return error; <<
1155 <<
1156 if (!XFS_IS_QUOTA_ON(mp) || xfs_is_qu <<
1157 return 0; <<
1158 <<
1159 if (tp->t_flags & XFS_TRANS_RESERVE) <<
1160 force_quota = true; <<
1161 <<
1162 error = xfs_trans_reserve_quota_nblks <<
1163 force_quota); <<
1164 if (!error) <<
1165 return 0; <<
1166 <<
1167 /* Quota failed, give back the new re <<
1168 xfs_add_fdblocks(mp, dblocks); <<
1169 tp->t_blk_res -= dblocks; <<
1170 xfs_add_frextents(mp, rtx); <<
1171 tp->t_rtx_res -= rtx; <<
1172 return error; <<
1173 } <<
1174 <<
1175 /* <<
1176 * Allocate an transaction in preparation for 1093 * Allocate an transaction in preparation for inode creation by reserving quota
1177 * against the given dquots. Callers are not 1094 * against the given dquots. Callers are not required to hold any inode locks.
1178 */ 1095 */
1179 int 1096 int
1180 xfs_trans_alloc_icreate( 1097 xfs_trans_alloc_icreate(
1181 struct xfs_mount *mp, 1098 struct xfs_mount *mp,
1182 struct xfs_trans_res *resv, 1099 struct xfs_trans_res *resv,
1183 struct xfs_dquot *udqp, 1100 struct xfs_dquot *udqp,
1184 struct xfs_dquot *gdqp, 1101 struct xfs_dquot *gdqp,
1185 struct xfs_dquot *pdqp, 1102 struct xfs_dquot *pdqp,
1186 unsigned int dblocks, 1103 unsigned int dblocks,
1187 struct xfs_trans **tpp) 1104 struct xfs_trans **tpp)
1188 { 1105 {
1189 struct xfs_trans *tp; 1106 struct xfs_trans *tp;
1190 bool retried = fal 1107 bool retried = false;
1191 int error; 1108 int error;
1192 1109
1193 retry: 1110 retry:
1194 error = xfs_trans_alloc(mp, resv, dbl 1111 error = xfs_trans_alloc(mp, resv, dblocks, 0, 0, &tp);
1195 if (error) 1112 if (error)
1196 return error; 1113 return error;
1197 1114
1198 error = xfs_trans_reserve_quota_icrea 1115 error = xfs_trans_reserve_quota_icreate(tp, udqp, gdqp, pdqp, dblocks);
1199 if ((error == -EDQUOT || error == -EN 1116 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1200 xfs_trans_cancel(tp); 1117 xfs_trans_cancel(tp);
1201 xfs_blockgc_free_dquots(mp, u 1118 xfs_blockgc_free_dquots(mp, udqp, gdqp, pdqp, 0);
1202 retried = true; 1119 retried = true;
1203 goto retry; 1120 goto retry;
1204 } 1121 }
1205 if (error) { 1122 if (error) {
1206 xfs_trans_cancel(tp); 1123 xfs_trans_cancel(tp);
1207 return error; 1124 return error;
1208 } 1125 }
1209 1126
1210 *tpp = tp; 1127 *tpp = tp;
1211 return 0; 1128 return 0;
1212 } 1129 }
1213 1130
1214 /* 1131 /*
1215 * Allocate an transaction, lock and join the 1132 * Allocate an transaction, lock and join the inode to it, and reserve quota
1216 * in preparation for inode attribute changes 1133 * in preparation for inode attribute changes that include uid, gid, or prid
1217 * changes. 1134 * changes.
1218 * 1135 *
1219 * The caller must ensure that the on-disk dq 1136 * The caller must ensure that the on-disk dquots attached to this inode have
1220 * already been allocated and initialized. T 1137 * already been allocated and initialized. The ILOCK will be dropped when the
1221 * transaction is committed or cancelled. 1138 * transaction is committed or cancelled.
1222 */ 1139 */
1223 int 1140 int
1224 xfs_trans_alloc_ichange( 1141 xfs_trans_alloc_ichange(
1225 struct xfs_inode *ip, 1142 struct xfs_inode *ip,
1226 struct xfs_dquot *new_udqp, 1143 struct xfs_dquot *new_udqp,
1227 struct xfs_dquot *new_gdqp, 1144 struct xfs_dquot *new_gdqp,
1228 struct xfs_dquot *new_pdqp, 1145 struct xfs_dquot *new_pdqp,
1229 bool force, 1146 bool force,
1230 struct xfs_trans **tpp) 1147 struct xfs_trans **tpp)
1231 { 1148 {
1232 struct xfs_trans *tp; 1149 struct xfs_trans *tp;
1233 struct xfs_mount *mp = ip->i_m 1150 struct xfs_mount *mp = ip->i_mount;
1234 struct xfs_dquot *udqp; 1151 struct xfs_dquot *udqp;
1235 struct xfs_dquot *gdqp; 1152 struct xfs_dquot *gdqp;
1236 struct xfs_dquot *pdqp; 1153 struct xfs_dquot *pdqp;
1237 bool retried = fal 1154 bool retried = false;
1238 int error; 1155 int error;
1239 1156
1240 retry: 1157 retry:
1241 error = xfs_trans_alloc(mp, &M_RES(mp 1158 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
1242 if (error) 1159 if (error)
1243 return error; 1160 return error;
1244 1161
1245 xfs_ilock(ip, XFS_ILOCK_EXCL); 1162 xfs_ilock(ip, XFS_ILOCK_EXCL);
1246 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXC 1163 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1247 1164
1248 error = xfs_qm_dqattach_locked(ip, fa 1165 error = xfs_qm_dqattach_locked(ip, false);
1249 if (error) { 1166 if (error) {
1250 /* Caller should have allocat 1167 /* Caller should have allocated the dquots! */
1251 ASSERT(error != -ENOENT); 1168 ASSERT(error != -ENOENT);
1252 goto out_cancel; 1169 goto out_cancel;
1253 } 1170 }
1254 1171
1255 /* 1172 /*
1256 * For each quota type, skip quota re 1173 * For each quota type, skip quota reservations if the inode's dquots
1257 * now match the ones that came from 1174 * now match the ones that came from the caller, or the caller didn't
1258 * pass one in. The inode's dquots c 1175 * pass one in. The inode's dquots can change if we drop the ILOCK to
1259 * perform a blockgc scan, so we must 1176 * perform a blockgc scan, so we must preserve the caller's arguments.
1260 */ 1177 */
1261 udqp = (new_udqp != ip->i_udquot) ? n 1178 udqp = (new_udqp != ip->i_udquot) ? new_udqp : NULL;
1262 gdqp = (new_gdqp != ip->i_gdquot) ? n 1179 gdqp = (new_gdqp != ip->i_gdquot) ? new_gdqp : NULL;
1263 pdqp = (new_pdqp != ip->i_pdquot) ? n 1180 pdqp = (new_pdqp != ip->i_pdquot) ? new_pdqp : NULL;
1264 if (udqp || gdqp || pdqp) { 1181 if (udqp || gdqp || pdqp) {
1265 unsigned int qflags = XFS_ 1182 unsigned int qflags = XFS_QMOPT_RES_REGBLKS;
1266 1183
1267 if (force) 1184 if (force)
1268 qflags |= XFS_QMOPT_F 1185 qflags |= XFS_QMOPT_FORCE_RES;
1269 1186
1270 /* 1187 /*
1271 * Reserve enough quota to ha 1188 * Reserve enough quota to handle blocks on disk and reserved
1272 * for a delayed allocation. 1189 * for a delayed allocation. We'll actually transfer the
1273 * delalloc reservation betwe 1190 * delalloc reservation between dquots at chown time, even
1274 * though that part is only s 1191 * though that part is only semi-transactional.
1275 */ 1192 */
1276 error = xfs_trans_reserve_quo 1193 error = xfs_trans_reserve_quota_bydquots(tp, mp, udqp, gdqp,
1277 pdqp, ip->i_n 1194 pdqp, ip->i_nblocks + ip->i_delayed_blks,
1278 1, qflags); 1195 1, qflags);
1279 if ((error == -EDQUOT || erro 1196 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1280 xfs_trans_cancel(tp); 1197 xfs_trans_cancel(tp);
1281 xfs_blockgc_free_dquo 1198 xfs_blockgc_free_dquots(mp, udqp, gdqp, pdqp, 0);
1282 retried = true; 1199 retried = true;
1283 goto retry; 1200 goto retry;
1284 } 1201 }
1285 if (error) 1202 if (error)
1286 goto out_cancel; 1203 goto out_cancel;
1287 } 1204 }
1288 1205
1289 *tpp = tp; 1206 *tpp = tp;
1290 return 0; <<
1291 <<
1292 out_cancel: <<
1293 xfs_trans_cancel(tp); <<
1294 return error; <<
1295 } <<
1296 <<
1297 /* <<
1298 * Allocate an transaction, lock and join the <<
1299 * and reserve quota for a directory update. <<
1300 * @dblocks will be set to zero for a reserva <<
1301 * @nospace_error will be set to a negative e <<
1302 * constraint we hit. <<
1303 * <<
1304 * The caller must ensure that the on-disk dq <<
1305 * already been allocated and initialized. T <<
1306 * transaction is committed or cancelled. <<
1307 * <<
1308 * Caller is responsible for unlocking the in <<
1309 */ <<
1310 int <<
1311 xfs_trans_alloc_dir( <<
1312 struct xfs_inode *dp, <<
1313 struct xfs_trans_res *resv, <<
1314 struct xfs_inode *ip, <<
1315 unsigned int *dblocks, <<
1316 struct xfs_trans **tpp, <<
1317 int *nospace_erro <<
1318 { <<
1319 struct xfs_trans *tp; <<
1320 struct xfs_mount *mp = ip->i_m <<
1321 unsigned int resblks; <<
1322 bool retried = fal <<
1323 int error; <<
1324 <<
1325 retry: <<
1326 *nospace_error = 0; <<
1327 resblks = *dblocks; <<
1328 error = xfs_trans_alloc(mp, resv, res <<
1329 if (error == -ENOSPC) { <<
1330 *nospace_error = error; <<
1331 resblks = 0; <<
1332 error = xfs_trans_alloc(mp, r <<
1333 } <<
1334 if (error) <<
1335 return error; <<
1336 <<
1337 xfs_lock_two_inodes(dp, XFS_ILOCK_EXC <<
1338 <<
1339 xfs_trans_ijoin(tp, dp, 0); <<
1340 xfs_trans_ijoin(tp, ip, 0); <<
1341 <<
1342 error = xfs_qm_dqattach_locked(dp, fa <<
1343 if (error) { <<
1344 /* Caller should have allocat <<
1345 ASSERT(error != -ENOENT); <<
1346 goto out_cancel; <<
1347 } <<
1348 <<
1349 error = xfs_qm_dqattach_locked(ip, fa <<
1350 if (error) { <<
1351 /* Caller should have allocat <<
1352 ASSERT(error != -ENOENT); <<
1353 goto out_cancel; <<
1354 } <<
1355 <<
1356 if (resblks == 0) <<
1357 goto done; <<
1358 <<
1359 error = xfs_trans_reserve_quota_nblks <<
1360 if (error == -EDQUOT || error == -ENO <<
1361 if (!retried) { <<
1362 xfs_trans_cancel(tp); <<
1363 xfs_iunlock(dp, XFS_I <<
1364 if (dp != ip) <<
1365 xfs_iunlock(i <<
1366 xfs_blockgc_free_quot <<
1367 retried = true; <<
1368 goto retry; <<
1369 } <<
1370 <<
1371 *nospace_error = error; <<
1372 resblks = 0; <<
1373 error = 0; <<
1374 } <<
1375 if (error) <<
1376 goto out_cancel; <<
1377 <<
1378 done: <<
1379 *tpp = tp; <<
1380 *dblocks = resblks; <<
1381 return 0; 1207 return 0;
1382 1208
1383 out_cancel: 1209 out_cancel:
1384 xfs_trans_cancel(tp); 1210 xfs_trans_cancel(tp);
1385 return error; 1211 return error;
1386 } 1212 }
1387 1213
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.