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 {
36 struct xfs_trans_res *res; 35 struct xfs_trans_res *res;
37 struct xfs_trans_res *end_res; 36 struct xfs_trans_res *end_res;
38 int i; 37 int i;
39 38
40 res = (struct xfs_trans_res *)M_RES(mp 39 res = (struct xfs_trans_res *)M_RES(mp);
41 end_res = (struct xfs_trans_res *)(M_R 40 end_res = (struct xfs_trans_res *)(M_RES(mp) + 1);
42 for (i = 0; res < end_res; i++, res++) 41 for (i = 0; res < end_res; i++, res++)
43 trace_xfs_trans_resv_calc(mp, 42 trace_xfs_trans_resv_calc(mp, i, res);
44 } 43 }
45 #else 44 #else
46 # define xfs_trans_trace_reservations(mp) 45 # define xfs_trans_trace_reservations(mp)
47 #endif 46 #endif
48 47
49 /* 48 /*
50 * Initialize the precomputed transaction rese 49 * Initialize the precomputed transaction reservation values
51 * in the mount structure. 50 * in the mount structure.
52 */ 51 */
53 void 52 void
54 xfs_trans_init( 53 xfs_trans_init(
55 struct xfs_mount *mp) 54 struct xfs_mount *mp)
56 { 55 {
57 xfs_trans_resv_calc(mp, M_RES(mp)); 56 xfs_trans_resv_calc(mp, M_RES(mp));
58 xfs_trans_trace_reservations(mp); 57 xfs_trans_trace_reservations(mp);
59 } 58 }
60 59
61 /* 60 /*
62 * Free the transaction structure. If there i 61 * Free the transaction structure. If there is more clean up
63 * to do when the structure is freed, add it h 62 * to do when the structure is freed, add it here.
64 */ 63 */
65 STATIC void 64 STATIC void
66 xfs_trans_free( 65 xfs_trans_free(
67 struct xfs_trans *tp) 66 struct xfs_trans *tp)
68 { 67 {
69 xfs_extent_busy_sort(&tp->t_busy); 68 xfs_extent_busy_sort(&tp->t_busy);
70 xfs_extent_busy_clear(tp->t_mountp, &t 69 xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false);
71 70
72 trace_xfs_trans_free(tp, _RET_IP_); 71 trace_xfs_trans_free(tp, _RET_IP_);
73 xfs_trans_clear_context(tp); 72 xfs_trans_clear_context(tp);
74 if (!(tp->t_flags & XFS_TRANS_NO_WRITE 73 if (!(tp->t_flags & XFS_TRANS_NO_WRITECOUNT))
75 sb_end_intwrite(tp->t_mountp-> 74 sb_end_intwrite(tp->t_mountp->m_super);
76 xfs_trans_free_dqinfo(tp); 75 xfs_trans_free_dqinfo(tp);
77 kmem_cache_free(xfs_trans_cache, tp); 76 kmem_cache_free(xfs_trans_cache, tp);
78 } 77 }
79 78
80 /* 79 /*
81 * This is called to create a new transaction 80 * This is called to create a new transaction which will share the
82 * permanent log reservation of the given tran 81 * permanent log reservation of the given transaction. The remaining
83 * unused block and rt extent reservations are 82 * unused block and rt extent reservations are also inherited. This
84 * implies that the original transaction is no 83 * implies that the original transaction is no longer allowed to allocate
85 * blocks. Locks and log items, however, are 84 * blocks. Locks and log items, however, are no inherited. They must
86 * be added to the new transaction explicitly. 85 * be added to the new transaction explicitly.
87 */ 86 */
88 STATIC struct xfs_trans * 87 STATIC struct xfs_trans *
89 xfs_trans_dup( 88 xfs_trans_dup(
90 struct xfs_trans *tp) 89 struct xfs_trans *tp)
91 { 90 {
92 struct xfs_trans *ntp; 91 struct xfs_trans *ntp;
93 92
94 trace_xfs_trans_dup(tp, _RET_IP_); 93 trace_xfs_trans_dup(tp, _RET_IP_);
95 94
96 ntp = kmem_cache_zalloc(xfs_trans_cach 95 ntp = kmem_cache_zalloc(xfs_trans_cache, GFP_KERNEL | __GFP_NOFAIL);
97 96
98 /* 97 /*
99 * Initialize the new transaction stru 98 * Initialize the new transaction structure.
100 */ 99 */
101 ntp->t_magic = XFS_TRANS_HEADER_MAGIC; 100 ntp->t_magic = XFS_TRANS_HEADER_MAGIC;
102 ntp->t_mountp = tp->t_mountp; 101 ntp->t_mountp = tp->t_mountp;
103 INIT_LIST_HEAD(&ntp->t_items); 102 INIT_LIST_HEAD(&ntp->t_items);
104 INIT_LIST_HEAD(&ntp->t_busy); 103 INIT_LIST_HEAD(&ntp->t_busy);
105 INIT_LIST_HEAD(&ntp->t_dfops); 104 INIT_LIST_HEAD(&ntp->t_dfops);
106 ntp->t_highest_agno = NULLAGNUMBER; !! 105 ntp->t_firstblock = NULLFSBLOCK;
107 106
108 ASSERT(tp->t_flags & XFS_TRANS_PERM_LO 107 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
109 ASSERT(tp->t_ticket != NULL); 108 ASSERT(tp->t_ticket != NULL);
110 109
111 ntp->t_flags = XFS_TRANS_PERM_LOG_RES 110 ntp->t_flags = XFS_TRANS_PERM_LOG_RES |
112 (tp->t_flags & XFS_TRAN 111 (tp->t_flags & XFS_TRANS_RESERVE) |
113 (tp->t_flags & XFS_TRAN 112 (tp->t_flags & XFS_TRANS_NO_WRITECOUNT) |
114 (tp->t_flags & XFS_TRAN 113 (tp->t_flags & XFS_TRANS_RES_FDBLKS);
115 /* We gave our writer reference to the 114 /* We gave our writer reference to the new transaction */
116 tp->t_flags |= XFS_TRANS_NO_WRITECOUNT 115 tp->t_flags |= XFS_TRANS_NO_WRITECOUNT;
117 ntp->t_ticket = xfs_log_ticket_get(tp- 116 ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
118 117
119 ASSERT(tp->t_blk_res >= tp->t_blk_res_ 118 ASSERT(tp->t_blk_res >= tp->t_blk_res_used);
120 ntp->t_blk_res = tp->t_blk_res - tp->t 119 ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
121 tp->t_blk_res = tp->t_blk_res_used; 120 tp->t_blk_res = tp->t_blk_res_used;
122 121
123 ntp->t_rtx_res = tp->t_rtx_res - tp->t 122 ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
124 tp->t_rtx_res = tp->t_rtx_res_used; 123 tp->t_rtx_res = tp->t_rtx_res_used;
125 124
126 xfs_trans_switch_context(tp, ntp); 125 xfs_trans_switch_context(tp, ntp);
127 126
128 /* move deferred ops over to the new t 127 /* move deferred ops over to the new tp */
129 xfs_defer_move(ntp, tp); 128 xfs_defer_move(ntp, tp);
130 129
131 xfs_trans_dup_dqinfo(tp, ntp); 130 xfs_trans_dup_dqinfo(tp, ntp);
132 return ntp; 131 return ntp;
133 } 132 }
134 133
135 /* 134 /*
136 * This is called to reserve free disk blocks 135 * This is called to reserve free disk blocks and log space for the
137 * given transaction. This must be done befor 136 * given transaction. This must be done before allocating any resources
138 * within the transaction. 137 * within the transaction.
139 * 138 *
140 * This will return ENOSPC if there are not en 139 * This will return ENOSPC if there are not enough blocks available.
141 * It will sleep waiting for available log spa 140 * It will sleep waiting for available log space.
142 * The only valid value for the flags paramete 141 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
143 * is used by long running transactions. If a 142 * is used by long running transactions. If any one of the reservations
144 * fails then they will all be backed out. 143 * fails then they will all be backed out.
145 * 144 *
146 * This does not do quota reservations. That t 145 * This does not do quota reservations. That typically is done by the
147 * caller afterwards. 146 * caller afterwards.
148 */ 147 */
149 static int 148 static int
150 xfs_trans_reserve( 149 xfs_trans_reserve(
151 struct xfs_trans *tp, 150 struct xfs_trans *tp,
152 struct xfs_trans_res *resp, 151 struct xfs_trans_res *resp,
153 uint blocks, 152 uint blocks,
154 uint rtextents) 153 uint rtextents)
155 { 154 {
156 struct xfs_mount *mp = tp->t_mo 155 struct xfs_mount *mp = tp->t_mountp;
157 int error = 0; 156 int error = 0;
158 bool rsvd = (tp->t_ 157 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
159 158
160 /* 159 /*
161 * Attempt to reserve the needed disk 160 * Attempt to reserve the needed disk blocks by decrementing
162 * the number needed from the number a 161 * the number needed from the number available. This will
163 * fail if the count would go below ze 162 * fail if the count would go below zero.
164 */ 163 */
165 if (blocks > 0) { 164 if (blocks > 0) {
166 error = xfs_dec_fdblocks(mp, b !! 165 error = xfs_mod_fdblocks(mp, -((int64_t)blocks), rsvd);
167 if (error != 0) 166 if (error != 0)
168 return -ENOSPC; 167 return -ENOSPC;
169 tp->t_blk_res += blocks; 168 tp->t_blk_res += blocks;
170 } 169 }
171 170
172 /* 171 /*
173 * Reserve the log space needed for th 172 * Reserve the log space needed for this transaction.
174 */ 173 */
175 if (resp->tr_logres > 0) { 174 if (resp->tr_logres > 0) {
176 bool permanent = false; 175 bool permanent = false;
177 176
178 ASSERT(tp->t_log_res == 0 || 177 ASSERT(tp->t_log_res == 0 ||
179 tp->t_log_res == resp-> 178 tp->t_log_res == resp->tr_logres);
180 ASSERT(tp->t_log_count == 0 || 179 ASSERT(tp->t_log_count == 0 ||
181 tp->t_log_count == resp 180 tp->t_log_count == resp->tr_logcount);
182 181
183 if (resp->tr_logflags & XFS_TR 182 if (resp->tr_logflags & XFS_TRANS_PERM_LOG_RES) {
184 tp->t_flags |= XFS_TRA 183 tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
185 permanent = true; 184 permanent = true;
186 } else { 185 } else {
187 ASSERT(tp->t_ticket == 186 ASSERT(tp->t_ticket == NULL);
188 ASSERT(!(tp->t_flags & 187 ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
189 } 188 }
190 189
191 if (tp->t_ticket != NULL) { 190 if (tp->t_ticket != NULL) {
192 ASSERT(resp->tr_logfla 191 ASSERT(resp->tr_logflags & XFS_TRANS_PERM_LOG_RES);
193 error = xfs_log_regran 192 error = xfs_log_regrant(mp, tp->t_ticket);
194 } else { 193 } else {
195 error = xfs_log_reserv 194 error = xfs_log_reserve(mp, resp->tr_logres,
196 195 resp->tr_logcount,
197 196 &tp->t_ticket, permanent);
198 } 197 }
199 198
200 if (error) 199 if (error)
201 goto undo_blocks; 200 goto undo_blocks;
202 201
203 tp->t_log_res = resp->tr_logre 202 tp->t_log_res = resp->tr_logres;
204 tp->t_log_count = resp->tr_log 203 tp->t_log_count = resp->tr_logcount;
205 } 204 }
206 205
207 /* 206 /*
208 * Attempt to reserve the needed realt 207 * Attempt to reserve the needed realtime extents by decrementing
209 * the number needed from the number a 208 * the number needed from the number available. This will
210 * fail if the count would go below ze 209 * fail if the count would go below zero.
211 */ 210 */
212 if (rtextents > 0) { 211 if (rtextents > 0) {
213 error = xfs_dec_frextents(mp, !! 212 error = xfs_mod_frextents(mp, -((int64_t)rtextents));
214 if (error) { 213 if (error) {
215 error = -ENOSPC; 214 error = -ENOSPC;
216 goto undo_log; 215 goto undo_log;
217 } 216 }
218 tp->t_rtx_res += rtextents; 217 tp->t_rtx_res += rtextents;
219 } 218 }
220 219
221 return 0; 220 return 0;
222 221
223 /* 222 /*
224 * Error cases jump to one of these la 223 * Error cases jump to one of these labels to undo any
225 * reservations which have already bee 224 * reservations which have already been performed.
226 */ 225 */
227 undo_log: 226 undo_log:
228 if (resp->tr_logres > 0) { 227 if (resp->tr_logres > 0) {
229 xfs_log_ticket_ungrant(mp->m_l 228 xfs_log_ticket_ungrant(mp->m_log, tp->t_ticket);
230 tp->t_ticket = NULL; 229 tp->t_ticket = NULL;
231 tp->t_log_res = 0; 230 tp->t_log_res = 0;
232 tp->t_flags &= ~XFS_TRANS_PERM 231 tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
233 } 232 }
234 233
235 undo_blocks: 234 undo_blocks:
236 if (blocks > 0) { 235 if (blocks > 0) {
237 xfs_add_fdblocks(mp, blocks); !! 236 xfs_mod_fdblocks(mp, (int64_t)blocks, rsvd);
238 tp->t_blk_res = 0; 237 tp->t_blk_res = 0;
239 } 238 }
240 return error; 239 return error;
241 } 240 }
242 241
243 int 242 int
244 xfs_trans_alloc( 243 xfs_trans_alloc(
245 struct xfs_mount *mp, 244 struct xfs_mount *mp,
246 struct xfs_trans_res *resp, 245 struct xfs_trans_res *resp,
247 uint blocks, 246 uint blocks,
248 uint rtextents, 247 uint rtextents,
249 uint flags, 248 uint flags,
250 struct xfs_trans **tpp) 249 struct xfs_trans **tpp)
251 { 250 {
252 struct xfs_trans *tp; 251 struct xfs_trans *tp;
253 bool want_retry = t 252 bool want_retry = true;
254 int error; 253 int error;
255 254
256 /* 255 /*
257 * Allocate the handle before we do ou 256 * Allocate the handle before we do our freeze accounting and setting up
258 * GFP_NOFS allocation context so that 257 * GFP_NOFS allocation context so that we avoid lockdep false positives
259 * by doing GFP_KERNEL allocations ins 258 * by doing GFP_KERNEL allocations inside sb_start_intwrite().
260 */ 259 */
261 retry: 260 retry:
262 tp = kmem_cache_zalloc(xfs_trans_cache 261 tp = kmem_cache_zalloc(xfs_trans_cache, GFP_KERNEL | __GFP_NOFAIL);
263 if (!(flags & XFS_TRANS_NO_WRITECOUNT) 262 if (!(flags & XFS_TRANS_NO_WRITECOUNT))
264 sb_start_intwrite(mp->m_super) 263 sb_start_intwrite(mp->m_super);
265 xfs_trans_set_context(tp); 264 xfs_trans_set_context(tp);
266 265
267 /* 266 /*
268 * Zero-reservation ("empty") transact 267 * Zero-reservation ("empty") transactions can't modify anything, so
269 * they're allowed to run while we're 268 * they're allowed to run while we're frozen.
270 */ 269 */
271 WARN_ON(resp->tr_logres > 0 && 270 WARN_ON(resp->tr_logres > 0 &&
272 mp->m_super->s_writers.frozen 271 mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE);
273 ASSERT(!(flags & XFS_TRANS_RES_FDBLKS) 272 ASSERT(!(flags & XFS_TRANS_RES_FDBLKS) ||
274 xfs_has_lazysbcount(mp)); 273 xfs_has_lazysbcount(mp));
275 274
276 tp->t_magic = XFS_TRANS_HEADER_MAGIC; 275 tp->t_magic = XFS_TRANS_HEADER_MAGIC;
277 tp->t_flags = flags; 276 tp->t_flags = flags;
278 tp->t_mountp = mp; 277 tp->t_mountp = mp;
279 INIT_LIST_HEAD(&tp->t_items); 278 INIT_LIST_HEAD(&tp->t_items);
280 INIT_LIST_HEAD(&tp->t_busy); 279 INIT_LIST_HEAD(&tp->t_busy);
281 INIT_LIST_HEAD(&tp->t_dfops); 280 INIT_LIST_HEAD(&tp->t_dfops);
282 tp->t_highest_agno = NULLAGNUMBER; !! 281 tp->t_firstblock = NULLFSBLOCK;
283 282
284 error = xfs_trans_reserve(tp, resp, bl 283 error = xfs_trans_reserve(tp, resp, blocks, rtextents);
285 if (error == -ENOSPC && want_retry) { 284 if (error == -ENOSPC && want_retry) {
286 xfs_trans_cancel(tp); 285 xfs_trans_cancel(tp);
287 286
288 /* 287 /*
289 * We weren't able to reserve 288 * We weren't able to reserve enough space for the transaction.
290 * Flush the other speculative 289 * Flush the other speculative space allocations to free space.
291 * Do not perform a synchronou 290 * Do not perform a synchronous scan because callers can hold
292 * other locks. 291 * other locks.
293 */ 292 */
294 error = xfs_blockgc_flush_all( !! 293 xfs_blockgc_flush_all(mp);
295 if (error) <<
296 return error; <<
297 want_retry = false; 294 want_retry = false;
298 goto retry; 295 goto retry;
299 } 296 }
300 if (error) { 297 if (error) {
301 xfs_trans_cancel(tp); 298 xfs_trans_cancel(tp);
302 return error; 299 return error;
303 } 300 }
304 301
305 trace_xfs_trans_alloc(tp, _RET_IP_); 302 trace_xfs_trans_alloc(tp, _RET_IP_);
306 303
307 *tpp = tp; 304 *tpp = tp;
308 return 0; 305 return 0;
309 } 306 }
310 307
311 /* 308 /*
312 * Create an empty transaction with no reserva 309 * Create an empty transaction with no reservation. This is a defensive
313 * mechanism for routines that query metadata 310 * mechanism for routines that query metadata without actually modifying them --
314 * if the metadata being queried is somehow cr 311 * if the metadata being queried is somehow cross-linked (think a btree block
315 * pointer that points higher in the tree), we 312 * pointer that points higher in the tree), we risk deadlock. However, blocks
316 * grabbed as part of a transaction can be re- 313 * grabbed as part of a transaction can be re-grabbed. The verifiers will
317 * notice the corrupt block and the operation 314 * notice the corrupt block and the operation will fail back to userspace
318 * without deadlocking. 315 * without deadlocking.
319 * 316 *
320 * Note the zero-length reservation; this tran 317 * Note the zero-length reservation; this transaction MUST be cancelled without
321 * any dirty data. 318 * any dirty data.
322 * 319 *
323 * Callers should obtain freeze protection to 320 * Callers should obtain freeze protection to avoid a conflict with fs freezing
324 * where we can be grabbing buffers at the sam 321 * where we can be grabbing buffers at the same time that freeze is trying to
325 * drain the buffer LRU list. 322 * drain the buffer LRU list.
326 */ 323 */
327 int 324 int
328 xfs_trans_alloc_empty( 325 xfs_trans_alloc_empty(
329 struct xfs_mount *mp, 326 struct xfs_mount *mp,
330 struct xfs_trans **tpp) 327 struct xfs_trans **tpp)
331 { 328 {
332 struct xfs_trans_res resv = 329 struct xfs_trans_res resv = {0};
333 330
334 return xfs_trans_alloc(mp, &resv, 0, 0 331 return xfs_trans_alloc(mp, &resv, 0, 0, XFS_TRANS_NO_WRITECOUNT, tpp);
335 } 332 }
336 333
337 /* 334 /*
338 * Record the indicated change to the given fi 335 * Record the indicated change to the given field for application
339 * to the file system's superblock when the tr 336 * to the file system's superblock when the transaction commits.
340 * For now, just store the change in the trans 337 * For now, just store the change in the transaction structure.
341 * 338 *
342 * Mark the transaction structure to indicate 339 * Mark the transaction structure to indicate that the superblock
343 * needs to be updated before committing. 340 * needs to be updated before committing.
344 * 341 *
345 * Because we may not be keeping track of allo 342 * Because we may not be keeping track of allocated/free inodes and
346 * used filesystem blocks in the superblock, w 343 * used filesystem blocks in the superblock, we do not mark the
347 * superblock dirty in this transaction if we 344 * superblock dirty in this transaction if we modify these fields.
348 * We still need to update the transaction del 345 * We still need to update the transaction deltas so that they get
349 * applied to the incore superblock, but we do 346 * applied to the incore superblock, but we don't want them to
350 * cause the superblock to get locked and logg 347 * cause the superblock to get locked and logged if these are the
351 * only fields in the superblock that the tran 348 * only fields in the superblock that the transaction modifies.
352 */ 349 */
353 void 350 void
354 xfs_trans_mod_sb( 351 xfs_trans_mod_sb(
355 xfs_trans_t *tp, 352 xfs_trans_t *tp,
356 uint field, 353 uint field,
357 int64_t delta) 354 int64_t delta)
358 { 355 {
359 uint32_t flags = (XFS_TRANS_DIR 356 uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
360 xfs_mount_t *mp = tp->t_mountp; 357 xfs_mount_t *mp = tp->t_mountp;
361 358
362 switch (field) { 359 switch (field) {
363 case XFS_TRANS_SB_ICOUNT: 360 case XFS_TRANS_SB_ICOUNT:
364 tp->t_icount_delta += delta; 361 tp->t_icount_delta += delta;
365 if (xfs_has_lazysbcount(mp)) 362 if (xfs_has_lazysbcount(mp))
366 flags &= ~XFS_TRANS_SB 363 flags &= ~XFS_TRANS_SB_DIRTY;
367 break; 364 break;
368 case XFS_TRANS_SB_IFREE: 365 case XFS_TRANS_SB_IFREE:
369 tp->t_ifree_delta += delta; 366 tp->t_ifree_delta += delta;
370 if (xfs_has_lazysbcount(mp)) 367 if (xfs_has_lazysbcount(mp))
371 flags &= ~XFS_TRANS_SB 368 flags &= ~XFS_TRANS_SB_DIRTY;
372 break; 369 break;
373 case XFS_TRANS_SB_FDBLOCKS: 370 case XFS_TRANS_SB_FDBLOCKS:
374 /* 371 /*
375 * Track the number of blocks 372 * Track the number of blocks allocated in the transaction.
376 * Make sure it does not excee 373 * Make sure it does not exceed the number reserved. If so,
377 * shutdown as this can lead t 374 * shutdown as this can lead to accounting inconsistency.
378 */ 375 */
379 if (delta < 0) { 376 if (delta < 0) {
380 tp->t_blk_res_used += 377 tp->t_blk_res_used += (uint)-delta;
381 if (tp->t_blk_res_used 378 if (tp->t_blk_res_used > tp->t_blk_res)
382 xfs_force_shut 379 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
383 } else if (delta > 0 && (tp->t 380 } else if (delta > 0 && (tp->t_flags & XFS_TRANS_RES_FDBLKS)) {
384 int64_t blkres_delta; 381 int64_t blkres_delta;
385 382
386 /* 383 /*
387 * Return freed blocks 384 * Return freed blocks directly to the reservation
388 * instead of the glob 385 * instead of the global pool, being careful not to
389 * overflow the trans 386 * overflow the trans counter. This is used to preserve
390 * reservation across 387 * reservation across chains of transaction rolls that
391 * repeatedly free and 388 * repeatedly free and allocate blocks.
392 */ 389 */
393 blkres_delta = min_t(i 390 blkres_delta = min_t(int64_t, delta,
394 U 391 UINT_MAX - tp->t_blk_res);
395 tp->t_blk_res += blkre 392 tp->t_blk_res += blkres_delta;
396 delta -= blkres_delta; 393 delta -= blkres_delta;
397 } 394 }
398 tp->t_fdblocks_delta += delta; 395 tp->t_fdblocks_delta += delta;
399 if (xfs_has_lazysbcount(mp)) 396 if (xfs_has_lazysbcount(mp))
400 flags &= ~XFS_TRANS_SB 397 flags &= ~XFS_TRANS_SB_DIRTY;
401 break; 398 break;
402 case XFS_TRANS_SB_RES_FDBLOCKS: 399 case XFS_TRANS_SB_RES_FDBLOCKS:
403 /* 400 /*
404 * The allocation has already 401 * The allocation has already been applied to the
405 * in-core superblock's counte 402 * in-core superblock's counter. This should only
406 * be applied to the on-disk s 403 * be applied to the on-disk superblock.
407 */ 404 */
408 tp->t_res_fdblocks_delta += de 405 tp->t_res_fdblocks_delta += delta;
409 if (xfs_has_lazysbcount(mp)) 406 if (xfs_has_lazysbcount(mp))
410 flags &= ~XFS_TRANS_SB 407 flags &= ~XFS_TRANS_SB_DIRTY;
411 break; 408 break;
412 case XFS_TRANS_SB_FREXTENTS: 409 case XFS_TRANS_SB_FREXTENTS:
413 /* 410 /*
414 * Track the number of blocks 411 * Track the number of blocks allocated in the
415 * transaction. Make sure it 412 * transaction. Make sure it does not exceed the
416 * number reserved. 413 * number reserved.
417 */ 414 */
418 if (delta < 0) { 415 if (delta < 0) {
419 tp->t_rtx_res_used += 416 tp->t_rtx_res_used += (uint)-delta;
420 ASSERT(tp->t_rtx_res_u 417 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
421 } 418 }
422 tp->t_frextents_delta += delta 419 tp->t_frextents_delta += delta;
423 break; 420 break;
424 case XFS_TRANS_SB_RES_FREXTENTS: 421 case XFS_TRANS_SB_RES_FREXTENTS:
425 /* 422 /*
426 * The allocation has already 423 * The allocation has already been applied to the
427 * in-core superblock's counte 424 * in-core superblock's counter. This should only
428 * be applied to the on-disk s 425 * be applied to the on-disk superblock.
429 */ 426 */
430 ASSERT(delta < 0); 427 ASSERT(delta < 0);
431 tp->t_res_frextents_delta += d 428 tp->t_res_frextents_delta += delta;
432 break; 429 break;
433 case XFS_TRANS_SB_DBLOCKS: 430 case XFS_TRANS_SB_DBLOCKS:
434 tp->t_dblocks_delta += delta; 431 tp->t_dblocks_delta += delta;
435 break; 432 break;
436 case XFS_TRANS_SB_AGCOUNT: 433 case XFS_TRANS_SB_AGCOUNT:
437 ASSERT(delta > 0); 434 ASSERT(delta > 0);
438 tp->t_agcount_delta += delta; 435 tp->t_agcount_delta += delta;
439 break; 436 break;
440 case XFS_TRANS_SB_IMAXPCT: 437 case XFS_TRANS_SB_IMAXPCT:
441 tp->t_imaxpct_delta += delta; 438 tp->t_imaxpct_delta += delta;
442 break; 439 break;
443 case XFS_TRANS_SB_REXTSIZE: 440 case XFS_TRANS_SB_REXTSIZE:
444 tp->t_rextsize_delta += delta; 441 tp->t_rextsize_delta += delta;
445 break; 442 break;
446 case XFS_TRANS_SB_RBMBLOCKS: 443 case XFS_TRANS_SB_RBMBLOCKS:
447 tp->t_rbmblocks_delta += delta 444 tp->t_rbmblocks_delta += delta;
448 break; 445 break;
449 case XFS_TRANS_SB_RBLOCKS: 446 case XFS_TRANS_SB_RBLOCKS:
450 tp->t_rblocks_delta += delta; 447 tp->t_rblocks_delta += delta;
451 break; 448 break;
452 case XFS_TRANS_SB_REXTENTS: 449 case XFS_TRANS_SB_REXTENTS:
453 tp->t_rextents_delta += delta; 450 tp->t_rextents_delta += delta;
454 break; 451 break;
455 case XFS_TRANS_SB_REXTSLOG: 452 case XFS_TRANS_SB_REXTSLOG:
456 tp->t_rextslog_delta += delta; 453 tp->t_rextslog_delta += delta;
457 break; 454 break;
458 default: 455 default:
459 ASSERT(0); 456 ASSERT(0);
460 return; 457 return;
461 } 458 }
462 459
463 tp->t_flags |= flags; 460 tp->t_flags |= flags;
464 } 461 }
465 462
466 /* 463 /*
467 * xfs_trans_apply_sb_deltas() is called from 464 * xfs_trans_apply_sb_deltas() is called from the commit code
468 * to bring the superblock buffer into the cur 465 * to bring the superblock buffer into the current transaction
469 * and modify it as requested by earlier calls 466 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
470 * 467 *
471 * For now we just look at each field allowed 468 * For now we just look at each field allowed to change and change
472 * it if necessary. 469 * it if necessary.
473 */ 470 */
474 STATIC void 471 STATIC void
475 xfs_trans_apply_sb_deltas( 472 xfs_trans_apply_sb_deltas(
476 xfs_trans_t *tp) 473 xfs_trans_t *tp)
477 { 474 {
478 struct xfs_dsb *sbp; 475 struct xfs_dsb *sbp;
479 struct xfs_buf *bp; 476 struct xfs_buf *bp;
480 int whole = 0; 477 int whole = 0;
481 478
482 bp = xfs_trans_getsb(tp); 479 bp = xfs_trans_getsb(tp);
483 sbp = bp->b_addr; 480 sbp = bp->b_addr;
484 481
485 /* 482 /*
486 * Only update the superblock counters 483 * Only update the superblock counters if we are logging them
487 */ 484 */
488 if (!xfs_has_lazysbcount((tp->t_mountp 485 if (!xfs_has_lazysbcount((tp->t_mountp))) {
489 if (tp->t_icount_delta) 486 if (tp->t_icount_delta)
490 be64_add_cpu(&sbp->sb_ 487 be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
491 if (tp->t_ifree_delta) 488 if (tp->t_ifree_delta)
492 be64_add_cpu(&sbp->sb_ 489 be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
493 if (tp->t_fdblocks_delta) 490 if (tp->t_fdblocks_delta)
494 be64_add_cpu(&sbp->sb_ 491 be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
495 if (tp->t_res_fdblocks_delta) 492 if (tp->t_res_fdblocks_delta)
496 be64_add_cpu(&sbp->sb_ 493 be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
497 } 494 }
498 495
499 /* 496 /*
500 * Updating frextents requires careful 497 * Updating frextents requires careful handling because it does not
501 * behave like the lazysb counters bec 498 * behave like the lazysb counters because we cannot rely on log
502 * recovery in older kenels to recompu 499 * recovery in older kenels to recompute the value from the rtbitmap.
503 * This means that the ondisk frextent 500 * This means that the ondisk frextents must be consistent with the
504 * rtbitmap. 501 * rtbitmap.
505 * 502 *
506 * Therefore, log the frextents change 503 * Therefore, log the frextents change to the ondisk superblock and
507 * update the incore superblock so tha 504 * update the incore superblock so that future calls to xfs_log_sb
508 * write the correct value ondisk. 505 * write the correct value ondisk.
509 * 506 *
510 * Don't touch m_frextents because it 507 * Don't touch m_frextents because it includes incore reservations,
511 * and those are handled by the unrese 508 * and those are handled by the unreserve function.
512 */ 509 */
513 if (tp->t_frextents_delta || tp->t_res 510 if (tp->t_frextents_delta || tp->t_res_frextents_delta) {
514 struct xfs_mount *mp = 511 struct xfs_mount *mp = tp->t_mountp;
515 int64_t rtxdel 512 int64_t rtxdelta;
516 513
517 rtxdelta = tp->t_frextents_del 514 rtxdelta = tp->t_frextents_delta + tp->t_res_frextents_delta;
518 515
519 spin_lock(&mp->m_sb_lock); 516 spin_lock(&mp->m_sb_lock);
520 be64_add_cpu(&sbp->sb_frextent 517 be64_add_cpu(&sbp->sb_frextents, rtxdelta);
521 mp->m_sb.sb_frextents += rtxde 518 mp->m_sb.sb_frextents += rtxdelta;
522 spin_unlock(&mp->m_sb_lock); 519 spin_unlock(&mp->m_sb_lock);
523 } 520 }
524 521
525 if (tp->t_dblocks_delta) { 522 if (tp->t_dblocks_delta) {
526 be64_add_cpu(&sbp->sb_dblocks, 523 be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
527 whole = 1; 524 whole = 1;
528 } 525 }
529 if (tp->t_agcount_delta) { 526 if (tp->t_agcount_delta) {
530 be32_add_cpu(&sbp->sb_agcount, 527 be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
531 whole = 1; 528 whole = 1;
532 } 529 }
533 if (tp->t_imaxpct_delta) { 530 if (tp->t_imaxpct_delta) {
534 sbp->sb_imax_pct += tp->t_imax 531 sbp->sb_imax_pct += tp->t_imaxpct_delta;
535 whole = 1; 532 whole = 1;
536 } 533 }
537 if (tp->t_rextsize_delta) { 534 if (tp->t_rextsize_delta) {
538 be32_add_cpu(&sbp->sb_rextsize 535 be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
539 whole = 1; 536 whole = 1;
540 } 537 }
541 if (tp->t_rbmblocks_delta) { 538 if (tp->t_rbmblocks_delta) {
542 be32_add_cpu(&sbp->sb_rbmblock 539 be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
543 whole = 1; 540 whole = 1;
544 } 541 }
545 if (tp->t_rblocks_delta) { 542 if (tp->t_rblocks_delta) {
546 be64_add_cpu(&sbp->sb_rblocks, 543 be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
547 whole = 1; 544 whole = 1;
548 } 545 }
549 if (tp->t_rextents_delta) { 546 if (tp->t_rextents_delta) {
550 be64_add_cpu(&sbp->sb_rextents 547 be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
551 whole = 1; 548 whole = 1;
552 } 549 }
553 if (tp->t_rextslog_delta) { 550 if (tp->t_rextslog_delta) {
554 sbp->sb_rextslog += tp->t_rext 551 sbp->sb_rextslog += tp->t_rextslog_delta;
555 whole = 1; 552 whole = 1;
556 } 553 }
557 554
558 xfs_trans_buf_set_type(tp, bp, XFS_BLF 555 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
559 if (whole) 556 if (whole)
560 /* 557 /*
561 * Log the whole thing, the fi 558 * Log the whole thing, the fields are noncontiguous.
562 */ 559 */
563 xfs_trans_log_buf(tp, bp, 0, s 560 xfs_trans_log_buf(tp, bp, 0, sizeof(struct xfs_dsb) - 1);
564 else 561 else
565 /* 562 /*
566 * Since all the modifiable fi 563 * Since all the modifiable fields are contiguous, we
567 * can get away with this. 564 * can get away with this.
568 */ 565 */
569 xfs_trans_log_buf(tp, bp, offs 566 xfs_trans_log_buf(tp, bp, offsetof(struct xfs_dsb, sb_icount),
570 offsetof(str 567 offsetof(struct xfs_dsb, sb_frextents) +
571 sizeof(sbp-> 568 sizeof(sbp->sb_frextents) - 1);
572 } 569 }
573 570
574 /* 571 /*
575 * xfs_trans_unreserve_and_mod_sb() is called 572 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations and
576 * apply superblock counter changes to the in- 573 * apply superblock counter changes to the in-core superblock. The
577 * t_res_fdblocks_delta and t_res_frextents_de 574 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
578 * applied to the in-core superblock. The ide 575 * applied to the in-core superblock. The idea is that that has already been
579 * done. 576 * done.
580 * 577 *
581 * If we are not logging superblock counters, 578 * If we are not logging superblock counters, then the inode allocated/free and
582 * used block counts are not updated in the on 579 * 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 580 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
584 * still need to update the incore superblock 581 * still need to update the incore superblock with the changes.
585 * 582 *
586 * Deltas for the inode count are +/-64, hence 583 * 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 584 * so we don't need to take the counter lock on every update.
588 */ 585 */
589 #define XFS_ICOUNT_BATCH 128 586 #define XFS_ICOUNT_BATCH 128
590 587
591 void 588 void
592 xfs_trans_unreserve_and_mod_sb( 589 xfs_trans_unreserve_and_mod_sb(
593 struct xfs_trans *tp) 590 struct xfs_trans *tp)
594 { 591 {
595 struct xfs_mount *mp = tp->t_mo 592 struct xfs_mount *mp = tp->t_mountp;
596 int64_t blkdelta = tp- !! 593 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
597 int64_t rtxdelta = tp- !! 594 int64_t blkdelta = 0;
>> 595 int64_t rtxdelta = 0;
598 int64_t idelta = 0; 596 int64_t idelta = 0;
599 int64_t ifreedelta = 0 597 int64_t ifreedelta = 0;
>> 598 int error;
600 599
601 /* !! 600 /* calculate deltas */
602 * Calculate the deltas. !! 601 if (tp->t_blk_res > 0)
603 * !! 602 blkdelta = tp->t_blk_res;
604 * t_fdblocks_delta and t_frextents_de !! 603 if ((tp->t_fdblocks_delta != 0) &&
605 * !! 604 (xfs_has_lazysbcount(mp) ||
606 * - positive values indicate blocks !! 605 (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 606 blkdelta += tp->t_fdblocks_delta;
617 ASSERT(blkdelta >= 0); <<
618 } <<
619 607
620 ASSERT(tp->t_rtx_res || tp->t_frextent !! 608 if (tp->t_rtx_res > 0)
621 if (tp->t_flags & XFS_TRANS_SB_DIRTY) !! 609 rtxdelta = tp->t_rtx_res;
>> 610 if ((tp->t_frextents_delta != 0) &&
>> 611 (tp->t_flags & XFS_TRANS_SB_DIRTY))
622 rtxdelta += tp->t_frextents_de 612 rtxdelta += tp->t_frextents_delta;
623 ASSERT(rtxdelta >= 0); <<
624 } <<
625 613
626 if (xfs_has_lazysbcount(mp) || (tp->t_ !! 614 if (xfs_has_lazysbcount(mp) ||
>> 615 (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
627 idelta = tp->t_icount_delta; 616 idelta = tp->t_icount_delta;
628 ifreedelta = tp->t_ifree_delta 617 ifreedelta = tp->t_ifree_delta;
629 } 618 }
630 619
631 /* apply the per-cpu counters */ 620 /* apply the per-cpu counters */
632 if (blkdelta) !! 621 if (blkdelta) {
633 xfs_add_fdblocks(mp, blkdelta) !! 622 error = xfs_mod_fdblocks(mp, blkdelta, rsvd);
>> 623 ASSERT(!error);
>> 624 }
634 625
635 if (idelta) 626 if (idelta)
636 percpu_counter_add_batch(&mp-> 627 percpu_counter_add_batch(&mp->m_icount, idelta,
637 XFS_I 628 XFS_ICOUNT_BATCH);
638 629
639 if (ifreedelta) 630 if (ifreedelta)
640 percpu_counter_add(&mp->m_ifre 631 percpu_counter_add(&mp->m_ifree, ifreedelta);
641 632
642 if (rtxdelta) !! 633 if (rtxdelta) {
643 xfs_add_frextents(mp, rtxdelta !! 634 error = xfs_mod_frextents(mp, rtxdelta);
>> 635 ASSERT(!error);
>> 636 }
644 637
645 if (!(tp->t_flags & XFS_TRANS_SB_DIRTY 638 if (!(tp->t_flags & XFS_TRANS_SB_DIRTY))
646 return; 639 return;
647 640
648 /* apply remaining deltas */ 641 /* apply remaining deltas */
649 spin_lock(&mp->m_sb_lock); 642 spin_lock(&mp->m_sb_lock);
650 mp->m_sb.sb_fdblocks += tp->t_fdblocks 643 mp->m_sb.sb_fdblocks += tp->t_fdblocks_delta + tp->t_res_fdblocks_delta;
651 mp->m_sb.sb_icount += idelta; 644 mp->m_sb.sb_icount += idelta;
652 mp->m_sb.sb_ifree += ifreedelta; 645 mp->m_sb.sb_ifree += ifreedelta;
653 /* 646 /*
654 * Do not touch sb_frextents here beca 647 * Do not touch sb_frextents here because we are dealing with incore
655 * reservation. sb_frextents is not p 648 * reservation. sb_frextents is not part of the lazy sb counters so it
656 * must be consistent with the ondisk 649 * must be consistent with the ondisk rtbitmap and must never include
657 * incore reservations. 650 * incore reservations.
658 */ 651 */
659 mp->m_sb.sb_dblocks += tp->t_dblocks_d 652 mp->m_sb.sb_dblocks += tp->t_dblocks_delta;
660 mp->m_sb.sb_agcount += tp->t_agcount_d 653 mp->m_sb.sb_agcount += tp->t_agcount_delta;
661 mp->m_sb.sb_imax_pct += tp->t_imaxpct_ 654 mp->m_sb.sb_imax_pct += tp->t_imaxpct_delta;
662 mp->m_sb.sb_rextsize += tp->t_rextsize 655 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 656 mp->m_sb.sb_rbmblocks += tp->t_rbmblocks_delta;
668 mp->m_sb.sb_rblocks += tp->t_rblocks_d 657 mp->m_sb.sb_rblocks += tp->t_rblocks_delta;
669 mp->m_sb.sb_rextents += tp->t_rextents 658 mp->m_sb.sb_rextents += tp->t_rextents_delta;
670 mp->m_sb.sb_rextslog += tp->t_rextslog 659 mp->m_sb.sb_rextslog += tp->t_rextslog_delta;
671 spin_unlock(&mp->m_sb_lock); 660 spin_unlock(&mp->m_sb_lock);
672 661
673 /* 662 /*
674 * Debug checks outside of the spinloc 663 * Debug checks outside of the spinlock so they don't lock up the
675 * machine if they fail. 664 * machine if they fail.
676 */ 665 */
677 ASSERT(mp->m_sb.sb_imax_pct >= 0); 666 ASSERT(mp->m_sb.sb_imax_pct >= 0);
678 ASSERT(mp->m_sb.sb_rextslog >= 0); 667 ASSERT(mp->m_sb.sb_rextslog >= 0);
>> 668 return;
679 } 669 }
680 670
681 /* Add the given log item to the transaction's 671 /* Add the given log item to the transaction's list of log items. */
682 void 672 void
683 xfs_trans_add_item( 673 xfs_trans_add_item(
684 struct xfs_trans *tp, 674 struct xfs_trans *tp,
685 struct xfs_log_item *lip) 675 struct xfs_log_item *lip)
686 { 676 {
687 ASSERT(lip->li_log == tp->t_mountp->m_ 677 ASSERT(lip->li_log == tp->t_mountp->m_log);
688 ASSERT(lip->li_ailp == tp->t_mountp->m 678 ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
689 ASSERT(list_empty(&lip->li_trans)); 679 ASSERT(list_empty(&lip->li_trans));
690 ASSERT(!test_bit(XFS_LI_DIRTY, &lip->l 680 ASSERT(!test_bit(XFS_LI_DIRTY, &lip->li_flags));
691 681
692 list_add_tail(&lip->li_trans, &tp->t_i 682 list_add_tail(&lip->li_trans, &tp->t_items);
693 trace_xfs_trans_add_item(tp, _RET_IP_) 683 trace_xfs_trans_add_item(tp, _RET_IP_);
694 } 684 }
695 685
696 /* 686 /*
697 * Unlink the log item from the transaction. t 687 * Unlink the log item from the transaction. the log item is no longer
698 * considered dirty in this transaction, as th 688 * considered dirty in this transaction, as the linked transaction has
699 * finished, either by abort or commit complet 689 * finished, either by abort or commit completion.
700 */ 690 */
701 void 691 void
702 xfs_trans_del_item( 692 xfs_trans_del_item(
703 struct xfs_log_item *lip) 693 struct xfs_log_item *lip)
704 { 694 {
705 clear_bit(XFS_LI_DIRTY, &lip->li_flags 695 clear_bit(XFS_LI_DIRTY, &lip->li_flags);
706 list_del_init(&lip->li_trans); 696 list_del_init(&lip->li_trans);
707 } 697 }
708 698
709 /* Detach and unlock all of the items in a tra 699 /* Detach and unlock all of the items in a transaction */
710 static void 700 static void
711 xfs_trans_free_items( 701 xfs_trans_free_items(
712 struct xfs_trans *tp, 702 struct xfs_trans *tp,
713 bool abort) 703 bool abort)
714 { 704 {
715 struct xfs_log_item *lip, *next; 705 struct xfs_log_item *lip, *next;
716 706
717 trace_xfs_trans_free_items(tp, _RET_IP 707 trace_xfs_trans_free_items(tp, _RET_IP_);
718 708
719 list_for_each_entry_safe(lip, next, &t 709 list_for_each_entry_safe(lip, next, &tp->t_items, li_trans) {
720 xfs_trans_del_item(lip); 710 xfs_trans_del_item(lip);
721 if (abort) 711 if (abort)
722 set_bit(XFS_LI_ABORTED 712 set_bit(XFS_LI_ABORTED, &lip->li_flags);
723 if (lip->li_ops->iop_release) 713 if (lip->li_ops->iop_release)
724 lip->li_ops->iop_relea 714 lip->li_ops->iop_release(lip);
725 } 715 }
726 } 716 }
727 717
>> 718 static inline void
>> 719 xfs_log_item_batch_insert(
>> 720 struct xfs_ail *ailp,
>> 721 struct xfs_ail_cursor *cur,
>> 722 struct xfs_log_item **log_items,
>> 723 int nr_items,
>> 724 xfs_lsn_t commit_lsn)
>> 725 {
>> 726 int i;
>> 727
>> 728 spin_lock(&ailp->ail_lock);
>> 729 /* xfs_trans_ail_update_bulk drops ailp->ail_lock */
>> 730 xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
>> 731
>> 732 for (i = 0; i < nr_items; i++) {
>> 733 struct xfs_log_item *lip = log_items[i];
>> 734
>> 735 if (lip->li_ops->iop_unpin)
>> 736 lip->li_ops->iop_unpin(lip, 0);
>> 737 }
>> 738 }
>> 739
>> 740 /*
>> 741 * Bulk operation version of xfs_trans_committed that takes a log vector of
>> 742 * items to insert into the AIL. This uses bulk AIL insertion techniques to
>> 743 * minimise lock traffic.
>> 744 *
>> 745 * If we are called with the aborted flag set, it is because a log write during
>> 746 * a CIL checkpoint commit has failed. In this case, all the items in the
>> 747 * checkpoint have already gone through iop_committed and iop_committing, which
>> 748 * means that checkpoint commit abort handling is treated exactly the same
>> 749 * as an iclog write error even though we haven't started any IO yet. Hence in
>> 750 * this case all we need to do is iop_committed processing, followed by an
>> 751 * iop_unpin(aborted) call.
>> 752 *
>> 753 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
>> 754 * at the end of the AIL, the insert cursor avoids the need to walk
>> 755 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
>> 756 * call. This saves a lot of needless list walking and is a net win, even
>> 757 * though it slightly increases that amount of AIL lock traffic to set it up
>> 758 * and tear it down.
>> 759 */
>> 760 void
>> 761 xfs_trans_committed_bulk(
>> 762 struct xfs_ail *ailp,
>> 763 struct list_head *lv_chain,
>> 764 xfs_lsn_t commit_lsn,
>> 765 bool aborted)
>> 766 {
>> 767 #define LOG_ITEM_BATCH_SIZE 32
>> 768 struct xfs_log_item *log_items[LOG_ITEM_BATCH_SIZE];
>> 769 struct xfs_log_vec *lv;
>> 770 struct xfs_ail_cursor cur;
>> 771 int i = 0;
>> 772
>> 773 spin_lock(&ailp->ail_lock);
>> 774 xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
>> 775 spin_unlock(&ailp->ail_lock);
>> 776
>> 777 /* unpin all the log items */
>> 778 list_for_each_entry(lv, lv_chain, lv_list) {
>> 779 struct xfs_log_item *lip = lv->lv_item;
>> 780 xfs_lsn_t item_lsn;
>> 781
>> 782 if (aborted)
>> 783 set_bit(XFS_LI_ABORTED, &lip->li_flags);
>> 784
>> 785 if (lip->li_ops->flags & XFS_ITEM_RELEASE_WHEN_COMMITTED) {
>> 786 lip->li_ops->iop_release(lip);
>> 787 continue;
>> 788 }
>> 789
>> 790 if (lip->li_ops->iop_committed)
>> 791 item_lsn = lip->li_ops->iop_committed(lip, commit_lsn);
>> 792 else
>> 793 item_lsn = commit_lsn;
>> 794
>> 795 /* item_lsn of -1 means the item needs no further processing */
>> 796 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
>> 797 continue;
>> 798
>> 799 /*
>> 800 * if we are aborting the operation, no point in inserting the
>> 801 * object into the AIL as we are in a shutdown situation.
>> 802 */
>> 803 if (aborted) {
>> 804 ASSERT(xlog_is_shutdown(ailp->ail_log));
>> 805 if (lip->li_ops->iop_unpin)
>> 806 lip->li_ops->iop_unpin(lip, 1);
>> 807 continue;
>> 808 }
>> 809
>> 810 if (item_lsn != commit_lsn) {
>> 811
>> 812 /*
>> 813 * Not a bulk update option due to unusual item_lsn.
>> 814 * Push into AIL immediately, rechecking the lsn once
>> 815 * we have the ail lock. Then unpin the item. This does
>> 816 * not affect the AIL cursor the bulk insert path is
>> 817 * using.
>> 818 */
>> 819 spin_lock(&ailp->ail_lock);
>> 820 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
>> 821 xfs_trans_ail_update(ailp, lip, item_lsn);
>> 822 else
>> 823 spin_unlock(&ailp->ail_lock);
>> 824 if (lip->li_ops->iop_unpin)
>> 825 lip->li_ops->iop_unpin(lip, 0);
>> 826 continue;
>> 827 }
>> 828
>> 829 /* Item is a candidate for bulk AIL insert. */
>> 830 log_items[i++] = lv->lv_item;
>> 831 if (i >= LOG_ITEM_BATCH_SIZE) {
>> 832 xfs_log_item_batch_insert(ailp, &cur, log_items,
>> 833 LOG_ITEM_BATCH_SIZE, commit_lsn);
>> 834 i = 0;
>> 835 }
>> 836 }
>> 837
>> 838 /* make sure we insert the remainder! */
>> 839 if (i)
>> 840 xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
>> 841
>> 842 spin_lock(&ailp->ail_lock);
>> 843 xfs_trans_ail_cursor_done(&cur);
>> 844 spin_unlock(&ailp->ail_lock);
>> 845 }
>> 846
728 /* 847 /*
729 * Sort transaction items prior to running pre 848 * Sort transaction items prior to running precommit operations. This will
730 * attempt to order the items such that they w 849 * attempt to order the items such that they will always be locked in the same
731 * order. Items that have no sort function are 850 * order. Items that have no sort function are moved to the end of the list
732 * and so are locked last. 851 * and so are locked last.
733 * 852 *
734 * This may need refinement as different types 853 * This may need refinement as different types of objects add sort functions.
735 * 854 *
736 * Function is more complex than it needs to b 855 * Function is more complex than it needs to be because we are comparing 64 bit
737 * values and the function only returns 32 bit 856 * values and the function only returns 32 bit values.
738 */ 857 */
739 static int 858 static int
740 xfs_trans_precommit_sort( 859 xfs_trans_precommit_sort(
741 void *unused_arg, 860 void *unused_arg,
742 const struct list_head *a, 861 const struct list_head *a,
743 const struct list_head *b) 862 const struct list_head *b)
744 { 863 {
745 struct xfs_log_item *lia = contain 864 struct xfs_log_item *lia = container_of(a,
746 struct 865 struct xfs_log_item, li_trans);
747 struct xfs_log_item *lib = contain 866 struct xfs_log_item *lib = container_of(b,
748 struct 867 struct xfs_log_item, li_trans);
749 int64_t diff; 868 int64_t diff;
750 869
751 /* 870 /*
752 * If both items are non-sortable, lea 871 * If both items are non-sortable, leave them alone. If only one is
753 * sortable, move the non-sortable ite 872 * sortable, move the non-sortable item towards the end of the list.
754 */ 873 */
755 if (!lia->li_ops->iop_sort && !lib->li 874 if (!lia->li_ops->iop_sort && !lib->li_ops->iop_sort)
756 return 0; 875 return 0;
757 if (!lia->li_ops->iop_sort) 876 if (!lia->li_ops->iop_sort)
758 return 1; 877 return 1;
759 if (!lib->li_ops->iop_sort) 878 if (!lib->li_ops->iop_sort)
760 return -1; 879 return -1;
761 880
762 diff = lia->li_ops->iop_sort(lia) - li 881 diff = lia->li_ops->iop_sort(lia) - lib->li_ops->iop_sort(lib);
763 if (diff < 0) 882 if (diff < 0)
764 return -1; 883 return -1;
765 if (diff > 0) 884 if (diff > 0)
766 return 1; 885 return 1;
767 return 0; 886 return 0;
768 } 887 }
769 888
770 /* 889 /*
771 * Run transaction precommit functions. 890 * Run transaction precommit functions.
772 * 891 *
773 * If there is an error in any of the callouts 892 * If there is an error in any of the callouts, then stop immediately and
774 * trigger a shutdown to abort the transaction 893 * trigger a shutdown to abort the transaction. There is no recovery possible
775 * from errors at this point as the transactio 894 * from errors at this point as the transaction is dirty....
776 */ 895 */
777 static int 896 static int
778 xfs_trans_run_precommits( 897 xfs_trans_run_precommits(
779 struct xfs_trans *tp) 898 struct xfs_trans *tp)
780 { 899 {
781 struct xfs_mount *mp = tp->t_mo 900 struct xfs_mount *mp = tp->t_mountp;
782 struct xfs_log_item *lip, *n; 901 struct xfs_log_item *lip, *n;
783 int error = 0; 902 int error = 0;
784 903
785 /* 904 /*
786 * Sort the item list to avoid ABBA de 905 * Sort the item list to avoid ABBA deadlocks with other transactions
787 * running precommit operations that l 906 * running precommit operations that lock multiple shared items such as
788 * inode cluster buffers. 907 * inode cluster buffers.
789 */ 908 */
790 list_sort(NULL, &tp->t_items, xfs_tran 909 list_sort(NULL, &tp->t_items, xfs_trans_precommit_sort);
791 910
792 /* 911 /*
793 * Precommit operations can remove the 912 * Precommit operations can remove the log item from the transaction
794 * if the log item exists purely to de 913 * if the log item exists purely to delay modifications until they
795 * can be ordered against other operat 914 * can be ordered against other operations. Hence we have to use
796 * list_for_each_entry_safe() here. 915 * list_for_each_entry_safe() here.
797 */ 916 */
798 list_for_each_entry_safe(lip, n, &tp-> 917 list_for_each_entry_safe(lip, n, &tp->t_items, li_trans) {
799 if (!test_bit(XFS_LI_DIRTY, &l 918 if (!test_bit(XFS_LI_DIRTY, &lip->li_flags))
800 continue; 919 continue;
801 if (lip->li_ops->iop_precommit 920 if (lip->li_ops->iop_precommit) {
802 error = lip->li_ops->i 921 error = lip->li_ops->iop_precommit(tp, lip);
803 if (error) 922 if (error)
804 break; 923 break;
805 } 924 }
806 } 925 }
807 if (error) 926 if (error)
808 xfs_force_shutdown(mp, SHUTDOW 927 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
809 return error; 928 return error;
810 } 929 }
811 930
812 /* 931 /*
813 * Commit the given transaction to the log. 932 * Commit the given transaction to the log.
814 * 933 *
815 * XFS disk error handling mechanism is not ba 934 * XFS disk error handling mechanism is not based on a typical
816 * transaction abort mechanism. Logically afte 935 * transaction abort mechanism. Logically after the filesystem
817 * gets marked 'SHUTDOWN', we can't let any ne 936 * gets marked 'SHUTDOWN', we can't let any new transactions
818 * be durable - ie. committed to disk - becaus 937 * be durable - ie. committed to disk - because some metadata might
819 * be inconsistent. In such cases, this return 938 * be inconsistent. In such cases, this returns an error, and the
820 * caller may assume that all locked objects j 939 * caller may assume that all locked objects joined to the transaction
821 * have already been unlocked as if the commit 940 * have already been unlocked as if the commit had succeeded.
822 * Do not reference the transaction structure 941 * Do not reference the transaction structure after this call.
823 */ 942 */
824 static int 943 static int
825 __xfs_trans_commit( 944 __xfs_trans_commit(
826 struct xfs_trans *tp, 945 struct xfs_trans *tp,
827 bool regrant) 946 bool regrant)
828 { 947 {
829 struct xfs_mount *mp = tp->t_mo 948 struct xfs_mount *mp = tp->t_mountp;
830 struct xlog *log = mp->m_l 949 struct xlog *log = mp->m_log;
831 xfs_csn_t commit_seq = 0 950 xfs_csn_t commit_seq = 0;
832 int error = 0; 951 int error = 0;
833 int sync = tp->t_f 952 int sync = tp->t_flags & XFS_TRANS_SYNC;
834 953
835 trace_xfs_trans_commit(tp, _RET_IP_); 954 trace_xfs_trans_commit(tp, _RET_IP_);
836 955
837 error = xfs_trans_run_precommits(tp); 956 error = xfs_trans_run_precommits(tp);
838 if (error) { 957 if (error) {
839 if (tp->t_flags & XFS_TRANS_PE 958 if (tp->t_flags & XFS_TRANS_PERM_LOG_RES)
840 xfs_defer_cancel(tp); 959 xfs_defer_cancel(tp);
841 goto out_unreserve; 960 goto out_unreserve;
842 } 961 }
843 962
844 /* 963 /*
845 * Finish deferred items on final comm 964 * Finish deferred items on final commit. Only permanent transactions
846 * should ever have deferred ops. 965 * should ever have deferred ops.
847 */ 966 */
848 WARN_ON_ONCE(!list_empty(&tp->t_dfops) 967 WARN_ON_ONCE(!list_empty(&tp->t_dfops) &&
849 !(tp->t_flags & XFS_TRANS 968 !(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
850 if (!regrant && (tp->t_flags & XFS_TRA 969 if (!regrant && (tp->t_flags & XFS_TRANS_PERM_LOG_RES)) {
851 error = xfs_defer_finish_norol 970 error = xfs_defer_finish_noroll(&tp);
852 if (error) 971 if (error)
853 goto out_unreserve; 972 goto out_unreserve;
854 <<
855 /* Run precommits from final t <<
856 error = xfs_trans_run_precommi <<
857 if (error) <<
858 goto out_unreserve; <<
859 } 973 }
860 974
861 /* 975 /*
862 * If there is nothing to be logged by 976 * If there is nothing to be logged by the transaction,
863 * then unlock all of the items associ 977 * then unlock all of the items associated with the
864 * transaction and free the transactio 978 * transaction and free the transaction structure.
865 * Also make sure to return any reserv 979 * Also make sure to return any reserved blocks to
866 * the free pool. 980 * the free pool.
867 */ 981 */
868 if (!(tp->t_flags & XFS_TRANS_DIRTY)) 982 if (!(tp->t_flags & XFS_TRANS_DIRTY))
869 goto out_unreserve; 983 goto out_unreserve;
870 984
871 /* 985 /*
872 * We must check against log shutdown 986 * We must check against log shutdown here because we cannot abort log
873 * items and leave them dirty, inconsi 987 * items and leave them dirty, inconsistent and unpinned in memory while
874 * the log is active. This leaves them 988 * the log is active. This leaves them open to being written back to
875 * disk, and that will lead to on-disk 989 * disk, and that will lead to on-disk corruption.
876 */ 990 */
877 if (xlog_is_shutdown(log)) { 991 if (xlog_is_shutdown(log)) {
878 error = -EIO; 992 error = -EIO;
879 goto out_unreserve; 993 goto out_unreserve;
880 } 994 }
881 995
882 ASSERT(tp->t_ticket != NULL); 996 ASSERT(tp->t_ticket != NULL);
883 997
884 /* 998 /*
885 * If we need to update the superblock 999 * If we need to update the superblock, then do it now.
886 */ 1000 */
887 if (tp->t_flags & XFS_TRANS_SB_DIRTY) 1001 if (tp->t_flags & XFS_TRANS_SB_DIRTY)
888 xfs_trans_apply_sb_deltas(tp); 1002 xfs_trans_apply_sb_deltas(tp);
889 xfs_trans_apply_dquot_deltas(tp); 1003 xfs_trans_apply_dquot_deltas(tp);
890 1004
891 xlog_cil_commit(log, tp, &commit_seq, 1005 xlog_cil_commit(log, tp, &commit_seq, regrant);
892 1006
893 xfs_trans_free(tp); 1007 xfs_trans_free(tp);
894 1008
895 /* 1009 /*
896 * If the transaction needs to be sync 1010 * If the transaction needs to be synchronous, then force the
897 * log out now and wait for it. 1011 * log out now and wait for it.
898 */ 1012 */
899 if (sync) { 1013 if (sync) {
900 error = xfs_log_force_seq(mp, 1014 error = xfs_log_force_seq(mp, commit_seq, XFS_LOG_SYNC, NULL);
901 XFS_STATS_INC(mp, xs_trans_syn 1015 XFS_STATS_INC(mp, xs_trans_sync);
902 } else { 1016 } else {
903 XFS_STATS_INC(mp, xs_trans_asy 1017 XFS_STATS_INC(mp, xs_trans_async);
904 } 1018 }
905 1019
906 return error; 1020 return error;
907 1021
908 out_unreserve: 1022 out_unreserve:
909 xfs_trans_unreserve_and_mod_sb(tp); 1023 xfs_trans_unreserve_and_mod_sb(tp);
910 1024
911 /* 1025 /*
912 * It is indeed possible for the trans 1026 * It is indeed possible for the transaction to be not dirty but
913 * the dqinfo portion to be. All that 1027 * the dqinfo portion to be. All that means is that we have some
914 * (non-persistent) quota reservations 1028 * (non-persistent) quota reservations that need to be unreserved.
915 */ 1029 */
916 xfs_trans_unreserve_and_mod_dquots(tp) 1030 xfs_trans_unreserve_and_mod_dquots(tp);
917 if (tp->t_ticket) { 1031 if (tp->t_ticket) {
918 if (regrant && !xlog_is_shutdo 1032 if (regrant && !xlog_is_shutdown(log))
919 xfs_log_ticket_regrant 1033 xfs_log_ticket_regrant(log, tp->t_ticket);
920 else 1034 else
921 xfs_log_ticket_ungrant 1035 xfs_log_ticket_ungrant(log, tp->t_ticket);
922 tp->t_ticket = NULL; 1036 tp->t_ticket = NULL;
923 } 1037 }
924 xfs_trans_free_items(tp, !!error); 1038 xfs_trans_free_items(tp, !!error);
925 xfs_trans_free(tp); 1039 xfs_trans_free(tp);
926 1040
927 XFS_STATS_INC(mp, xs_trans_empty); 1041 XFS_STATS_INC(mp, xs_trans_empty);
928 return error; 1042 return error;
929 } 1043 }
930 1044
931 int 1045 int
932 xfs_trans_commit( 1046 xfs_trans_commit(
933 struct xfs_trans *tp) 1047 struct xfs_trans *tp)
934 { 1048 {
935 return __xfs_trans_commit(tp, false); 1049 return __xfs_trans_commit(tp, false);
936 } 1050 }
937 1051
938 /* 1052 /*
939 * Unlock all of the transaction's items and f 1053 * Unlock all of the transaction's items and free the transaction. If the
940 * transaction is dirty, we must shut down the 1054 * transaction is dirty, we must shut down the filesystem because there is no
941 * way to restore them to their previous state 1055 * way to restore them to their previous state.
942 * 1056 *
943 * If the transaction has made a log reservati 1057 * If the transaction has made a log reservation, make sure to release it as
944 * well. 1058 * well.
945 * 1059 *
946 * This is a high level function (equivalent t 1060 * This is a high level function (equivalent to xfs_trans_commit()) and so can
947 * be called after the transaction has effecti 1061 * be called after the transaction has effectively been aborted due to the mount
948 * being shut down. However, if the mount has 1062 * being shut down. However, if the mount has not been shut down and the
949 * transaction is dirty we will shut the mount 1063 * transaction is dirty we will shut the mount down and, in doing so, that
950 * guarantees that the log is shut down, too. 1064 * guarantees that the log is shut down, too. Hence we don't need to be as
951 * careful with shutdown state and dirty items 1065 * careful with shutdown state and dirty items here as we need to be in
952 * xfs_trans_commit(). 1066 * xfs_trans_commit().
953 */ 1067 */
954 void 1068 void
955 xfs_trans_cancel( 1069 xfs_trans_cancel(
956 struct xfs_trans *tp) 1070 struct xfs_trans *tp)
957 { 1071 {
958 struct xfs_mount *mp = tp->t_mo 1072 struct xfs_mount *mp = tp->t_mountp;
959 struct xlog *log = mp->m_l 1073 struct xlog *log = mp->m_log;
960 bool dirty = (tp->t 1074 bool dirty = (tp->t_flags & XFS_TRANS_DIRTY);
961 1075
962 trace_xfs_trans_cancel(tp, _RET_IP_); 1076 trace_xfs_trans_cancel(tp, _RET_IP_);
963 1077
964 /* 1078 /*
965 * It's never valid to cancel a transa 1079 * It's never valid to cancel a transaction with deferred ops attached,
966 * because the transaction is effectiv 1080 * because the transaction is effectively dirty. Complain about this
967 * loudly before freeing the in-memory !! 1081 * loudly before freeing the in-memory defer items.
968 * filesystem. <<
969 */ 1082 */
970 if (!list_empty(&tp->t_dfops)) { 1083 if (!list_empty(&tp->t_dfops)) {
>> 1084 ASSERT(xfs_is_shutdown(mp) || list_empty(&tp->t_dfops));
971 ASSERT(tp->t_flags & XFS_TRANS 1085 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
972 dirty = true; 1086 dirty = true;
973 xfs_defer_cancel(tp); 1087 xfs_defer_cancel(tp);
974 } 1088 }
975 1089
976 /* 1090 /*
977 * See if the caller is relying on us 1091 * See if the caller is relying on us to shut down the filesystem. We
978 * only want an error report if there 1092 * only want an error report if there isn't already a shutdown in
979 * progress, so we only need to check 1093 * progress, so we only need to check against the mount shutdown state
980 * here. 1094 * here.
981 */ 1095 */
982 if (dirty && !xfs_is_shutdown(mp)) { 1096 if (dirty && !xfs_is_shutdown(mp)) {
983 XFS_ERROR_REPORT("xfs_trans_ca 1097 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
984 xfs_force_shutdown(mp, SHUTDOW 1098 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
985 } 1099 }
986 #ifdef DEBUG 1100 #ifdef DEBUG
987 /* Log items need to be consistent unt 1101 /* Log items need to be consistent until the log is shut down. */
988 if (!dirty && !xlog_is_shutdown(log)) 1102 if (!dirty && !xlog_is_shutdown(log)) {
989 struct xfs_log_item *lip; 1103 struct xfs_log_item *lip;
990 1104
991 list_for_each_entry(lip, &tp-> 1105 list_for_each_entry(lip, &tp->t_items, li_trans)
992 ASSERT(!xlog_item_is_i 1106 ASSERT(!xlog_item_is_intent_done(lip));
993 } 1107 }
994 #endif 1108 #endif
995 xfs_trans_unreserve_and_mod_sb(tp); 1109 xfs_trans_unreserve_and_mod_sb(tp);
996 xfs_trans_unreserve_and_mod_dquots(tp) 1110 xfs_trans_unreserve_and_mod_dquots(tp);
997 1111
998 if (tp->t_ticket) { 1112 if (tp->t_ticket) {
999 xfs_log_ticket_ungrant(log, tp 1113 xfs_log_ticket_ungrant(log, tp->t_ticket);
1000 tp->t_ticket = NULL; 1114 tp->t_ticket = NULL;
1001 } 1115 }
1002 1116
1003 xfs_trans_free_items(tp, dirty); 1117 xfs_trans_free_items(tp, dirty);
1004 xfs_trans_free(tp); 1118 xfs_trans_free(tp);
1005 } 1119 }
1006 1120
1007 /* 1121 /*
1008 * Roll from one trans in the sequence of PER 1122 * Roll from one trans in the sequence of PERMANENT transactions to
1009 * the next: permanent transactions are only 1123 * the next: permanent transactions are only flushed out when
1010 * committed with xfs_trans_commit(), but we 1124 * committed with xfs_trans_commit(), but we still want as soon
1011 * as possible to let chunks of it go to the 1125 * 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 1126 * chunk we've been working on and get a new transaction to continue.
1013 */ 1127 */
1014 int 1128 int
1015 xfs_trans_roll( 1129 xfs_trans_roll(
1016 struct xfs_trans **tpp) 1130 struct xfs_trans **tpp)
1017 { 1131 {
1018 struct xfs_trans *trans = *tpp 1132 struct xfs_trans *trans = *tpp;
1019 struct xfs_trans_res tres; 1133 struct xfs_trans_res tres;
1020 int error; 1134 int error;
1021 1135
1022 trace_xfs_trans_roll(trans, _RET_IP_) 1136 trace_xfs_trans_roll(trans, _RET_IP_);
1023 1137
1024 /* 1138 /*
1025 * Copy the critical parameters from 1139 * Copy the critical parameters from one trans to the next.
1026 */ 1140 */
1027 tres.tr_logres = trans->t_log_res; 1141 tres.tr_logres = trans->t_log_res;
1028 tres.tr_logcount = trans->t_log_count 1142 tres.tr_logcount = trans->t_log_count;
1029 1143
1030 *tpp = xfs_trans_dup(trans); 1144 *tpp = xfs_trans_dup(trans);
1031 1145
1032 /* 1146 /*
1033 * Commit the current transaction. 1147 * Commit the current transaction.
1034 * If this commit failed, then it'd j 1148 * If this commit failed, then it'd just unlock those items that
1035 * are not marked ihold. That also me 1149 * are not marked ihold. That also means that a filesystem shutdown
1036 * is in progress. The caller takes t 1150 * is in progress. The caller takes the responsibility to cancel
1037 * the duplicate transaction that get 1151 * the duplicate transaction that gets returned.
1038 */ 1152 */
1039 error = __xfs_trans_commit(trans, tru 1153 error = __xfs_trans_commit(trans, true);
1040 if (error) 1154 if (error)
1041 return error; 1155 return error;
1042 1156
1043 /* 1157 /*
1044 * Reserve space in the log for the n 1158 * Reserve space in the log for the next transaction.
1045 * This also pushes items in the "AIL 1159 * This also pushes items in the "AIL", the list of logged items,
1046 * out to disk if they are taking up 1160 * out to disk if they are taking up space at the tail of the log
1047 * that we want to use. This require 1161 * that we want to use. This requires that either nothing be locked
1048 * across this call, or that anything 1162 * across this call, or that anything that is locked be logged in
1049 * the prior and the next transaction 1163 * the prior and the next transactions.
1050 */ 1164 */
1051 tres.tr_logflags = XFS_TRANS_PERM_LOG 1165 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1052 return xfs_trans_reserve(*tpp, &tres, 1166 return xfs_trans_reserve(*tpp, &tres, 0, 0);
1053 } 1167 }
1054 1168
1055 /* 1169 /*
1056 * Allocate an transaction, lock and join the 1170 * Allocate an transaction, lock and join the inode to it, and reserve quota.
1057 * 1171 *
1058 * The caller must ensure that the on-disk dq 1172 * The caller must ensure that the on-disk dquots attached to this inode have
1059 * already been allocated and initialized. T 1173 * already been allocated and initialized. The caller is responsible for
1060 * releasing ILOCK_EXCL if a new transaction 1174 * releasing ILOCK_EXCL if a new transaction is returned.
1061 */ 1175 */
1062 int 1176 int
1063 xfs_trans_alloc_inode( 1177 xfs_trans_alloc_inode(
1064 struct xfs_inode *ip, 1178 struct xfs_inode *ip,
1065 struct xfs_trans_res *resv, 1179 struct xfs_trans_res *resv,
1066 unsigned int dblocks, 1180 unsigned int dblocks,
1067 unsigned int rblocks, 1181 unsigned int rblocks,
1068 bool force, 1182 bool force,
1069 struct xfs_trans **tpp) 1183 struct xfs_trans **tpp)
1070 { 1184 {
1071 struct xfs_trans *tp; 1185 struct xfs_trans *tp;
1072 struct xfs_mount *mp = ip->i_m 1186 struct xfs_mount *mp = ip->i_mount;
1073 bool retried = fal 1187 bool retried = false;
1074 int error; 1188 int error;
1075 1189
1076 retry: 1190 retry:
1077 error = xfs_trans_alloc(mp, resv, dbl 1191 error = xfs_trans_alloc(mp, resv, dblocks,
1078 xfs_extlen_to_rtxlen( !! 1192 rblocks / mp->m_sb.sb_rextsize,
1079 force ? XFS_TRANS_RES 1193 force ? XFS_TRANS_RESERVE : 0, &tp);
1080 if (error) 1194 if (error)
1081 return error; 1195 return error;
1082 1196
1083 xfs_ilock(ip, XFS_ILOCK_EXCL); 1197 xfs_ilock(ip, XFS_ILOCK_EXCL);
1084 xfs_trans_ijoin(tp, ip, 0); 1198 xfs_trans_ijoin(tp, ip, 0);
1085 1199
1086 error = xfs_qm_dqattach_locked(ip, fa 1200 error = xfs_qm_dqattach_locked(ip, false);
1087 if (error) { 1201 if (error) {
1088 /* Caller should have allocat 1202 /* Caller should have allocated the dquots! */
1089 ASSERT(error != -ENOENT); 1203 ASSERT(error != -ENOENT);
1090 goto out_cancel; 1204 goto out_cancel;
1091 } 1205 }
1092 1206
1093 error = xfs_trans_reserve_quota_nblks 1207 error = xfs_trans_reserve_quota_nblks(tp, ip, dblocks, rblocks, force);
1094 if ((error == -EDQUOT || error == -EN 1208 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1095 xfs_trans_cancel(tp); 1209 xfs_trans_cancel(tp);
1096 xfs_iunlock(ip, XFS_ILOCK_EXC 1210 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1097 xfs_blockgc_free_quota(ip, 0) 1211 xfs_blockgc_free_quota(ip, 0);
1098 retried = true; 1212 retried = true;
1099 goto retry; 1213 goto retry;
1100 } 1214 }
1101 if (error) 1215 if (error)
1102 goto out_cancel; 1216 goto out_cancel;
1103 1217
1104 *tpp = tp; 1218 *tpp = tp;
1105 return 0; 1219 return 0;
1106 1220
1107 out_cancel: 1221 out_cancel:
1108 xfs_trans_cancel(tp); 1222 xfs_trans_cancel(tp);
1109 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1223 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1110 return error; 1224 return error;
1111 } 1225 }
1112 1226
1113 /* 1227 /*
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 1228 * Allocate an transaction in preparation for inode creation by reserving quota
1177 * against the given dquots. Callers are not 1229 * against the given dquots. Callers are not required to hold any inode locks.
1178 */ 1230 */
1179 int 1231 int
1180 xfs_trans_alloc_icreate( 1232 xfs_trans_alloc_icreate(
1181 struct xfs_mount *mp, 1233 struct xfs_mount *mp,
1182 struct xfs_trans_res *resv, 1234 struct xfs_trans_res *resv,
1183 struct xfs_dquot *udqp, 1235 struct xfs_dquot *udqp,
1184 struct xfs_dquot *gdqp, 1236 struct xfs_dquot *gdqp,
1185 struct xfs_dquot *pdqp, 1237 struct xfs_dquot *pdqp,
1186 unsigned int dblocks, 1238 unsigned int dblocks,
1187 struct xfs_trans **tpp) 1239 struct xfs_trans **tpp)
1188 { 1240 {
1189 struct xfs_trans *tp; 1241 struct xfs_trans *tp;
1190 bool retried = fal 1242 bool retried = false;
1191 int error; 1243 int error;
1192 1244
1193 retry: 1245 retry:
1194 error = xfs_trans_alloc(mp, resv, dbl 1246 error = xfs_trans_alloc(mp, resv, dblocks, 0, 0, &tp);
1195 if (error) 1247 if (error)
1196 return error; 1248 return error;
1197 1249
1198 error = xfs_trans_reserve_quota_icrea 1250 error = xfs_trans_reserve_quota_icreate(tp, udqp, gdqp, pdqp, dblocks);
1199 if ((error == -EDQUOT || error == -EN 1251 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1200 xfs_trans_cancel(tp); 1252 xfs_trans_cancel(tp);
1201 xfs_blockgc_free_dquots(mp, u 1253 xfs_blockgc_free_dquots(mp, udqp, gdqp, pdqp, 0);
1202 retried = true; 1254 retried = true;
1203 goto retry; 1255 goto retry;
1204 } 1256 }
1205 if (error) { 1257 if (error) {
1206 xfs_trans_cancel(tp); 1258 xfs_trans_cancel(tp);
1207 return error; 1259 return error;
1208 } 1260 }
1209 1261
1210 *tpp = tp; 1262 *tpp = tp;
1211 return 0; 1263 return 0;
1212 } 1264 }
1213 1265
1214 /* 1266 /*
1215 * Allocate an transaction, lock and join the 1267 * Allocate an transaction, lock and join the inode to it, and reserve quota
1216 * in preparation for inode attribute changes 1268 * in preparation for inode attribute changes that include uid, gid, or prid
1217 * changes. 1269 * changes.
1218 * 1270 *
1219 * The caller must ensure that the on-disk dq 1271 * The caller must ensure that the on-disk dquots attached to this inode have
1220 * already been allocated and initialized. T 1272 * already been allocated and initialized. The ILOCK will be dropped when the
1221 * transaction is committed or cancelled. 1273 * transaction is committed or cancelled.
1222 */ 1274 */
1223 int 1275 int
1224 xfs_trans_alloc_ichange( 1276 xfs_trans_alloc_ichange(
1225 struct xfs_inode *ip, 1277 struct xfs_inode *ip,
1226 struct xfs_dquot *new_udqp, 1278 struct xfs_dquot *new_udqp,
1227 struct xfs_dquot *new_gdqp, 1279 struct xfs_dquot *new_gdqp,
1228 struct xfs_dquot *new_pdqp, 1280 struct xfs_dquot *new_pdqp,
1229 bool force, 1281 bool force,
1230 struct xfs_trans **tpp) 1282 struct xfs_trans **tpp)
1231 { 1283 {
1232 struct xfs_trans *tp; 1284 struct xfs_trans *tp;
1233 struct xfs_mount *mp = ip->i_m 1285 struct xfs_mount *mp = ip->i_mount;
1234 struct xfs_dquot *udqp; 1286 struct xfs_dquot *udqp;
1235 struct xfs_dquot *gdqp; 1287 struct xfs_dquot *gdqp;
1236 struct xfs_dquot *pdqp; 1288 struct xfs_dquot *pdqp;
1237 bool retried = fal 1289 bool retried = false;
1238 int error; 1290 int error;
1239 1291
1240 retry: 1292 retry:
1241 error = xfs_trans_alloc(mp, &M_RES(mp 1293 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
1242 if (error) 1294 if (error)
1243 return error; 1295 return error;
1244 1296
1245 xfs_ilock(ip, XFS_ILOCK_EXCL); 1297 xfs_ilock(ip, XFS_ILOCK_EXCL);
1246 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXC 1298 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1247 1299
1248 error = xfs_qm_dqattach_locked(ip, fa 1300 error = xfs_qm_dqattach_locked(ip, false);
1249 if (error) { 1301 if (error) {
1250 /* Caller should have allocat 1302 /* Caller should have allocated the dquots! */
1251 ASSERT(error != -ENOENT); 1303 ASSERT(error != -ENOENT);
1252 goto out_cancel; 1304 goto out_cancel;
1253 } 1305 }
1254 1306
1255 /* 1307 /*
1256 * For each quota type, skip quota re 1308 * For each quota type, skip quota reservations if the inode's dquots
1257 * now match the ones that came from 1309 * now match the ones that came from the caller, or the caller didn't
1258 * pass one in. The inode's dquots c 1310 * pass one in. The inode's dquots can change if we drop the ILOCK to
1259 * perform a blockgc scan, so we must 1311 * perform a blockgc scan, so we must preserve the caller's arguments.
1260 */ 1312 */
1261 udqp = (new_udqp != ip->i_udquot) ? n 1313 udqp = (new_udqp != ip->i_udquot) ? new_udqp : NULL;
1262 gdqp = (new_gdqp != ip->i_gdquot) ? n 1314 gdqp = (new_gdqp != ip->i_gdquot) ? new_gdqp : NULL;
1263 pdqp = (new_pdqp != ip->i_pdquot) ? n 1315 pdqp = (new_pdqp != ip->i_pdquot) ? new_pdqp : NULL;
1264 if (udqp || gdqp || pdqp) { 1316 if (udqp || gdqp || pdqp) {
1265 unsigned int qflags = XFS_ 1317 unsigned int qflags = XFS_QMOPT_RES_REGBLKS;
1266 1318
1267 if (force) 1319 if (force)
1268 qflags |= XFS_QMOPT_F 1320 qflags |= XFS_QMOPT_FORCE_RES;
1269 1321
1270 /* 1322 /*
1271 * Reserve enough quota to ha 1323 * Reserve enough quota to handle blocks on disk and reserved
1272 * for a delayed allocation. 1324 * for a delayed allocation. We'll actually transfer the
1273 * delalloc reservation betwe 1325 * delalloc reservation between dquots at chown time, even
1274 * though that part is only s 1326 * though that part is only semi-transactional.
1275 */ 1327 */
1276 error = xfs_trans_reserve_quo 1328 error = xfs_trans_reserve_quota_bydquots(tp, mp, udqp, gdqp,
1277 pdqp, ip->i_n 1329 pdqp, ip->i_nblocks + ip->i_delayed_blks,
1278 1, qflags); 1330 1, qflags);
1279 if ((error == -EDQUOT || erro 1331 if ((error == -EDQUOT || error == -ENOSPC) && !retried) {
1280 xfs_trans_cancel(tp); 1332 xfs_trans_cancel(tp);
1281 xfs_blockgc_free_dquo 1333 xfs_blockgc_free_dquots(mp, udqp, gdqp, pdqp, 0);
1282 retried = true; 1334 retried = true;
1283 goto retry; 1335 goto retry;
1284 } 1336 }
1285 if (error) 1337 if (error)
1286 goto out_cancel; 1338 goto out_cancel;
1287 } 1339 }
1288 1340
1289 *tpp = tp; 1341 *tpp = tp;
1290 return 0; 1342 return 0;
1291 1343
1292 out_cancel: 1344 out_cancel:
1293 xfs_trans_cancel(tp); 1345 xfs_trans_cancel(tp);
1294 return error; 1346 return error;
1295 } 1347 }
1296 1348
1297 /* 1349 /*
1298 * Allocate an transaction, lock and join the 1350 * Allocate an transaction, lock and join the directory and child inodes to it,
1299 * and reserve quota for a directory update. 1351 * and reserve quota for a directory update. If there isn't sufficient space,
1300 * @dblocks will be set to zero for a reserva 1352 * @dblocks will be set to zero for a reservationless directory update and
1301 * @nospace_error will be set to a negative e 1353 * @nospace_error will be set to a negative errno describing the space
1302 * constraint we hit. 1354 * constraint we hit.
1303 * 1355 *
1304 * The caller must ensure that the on-disk dq 1356 * The caller must ensure that the on-disk dquots attached to this inode have
1305 * already been allocated and initialized. T 1357 * already been allocated and initialized. The ILOCKs will be dropped when the
1306 * transaction is committed or cancelled. 1358 * transaction is committed or cancelled.
1307 * <<
1308 * Caller is responsible for unlocking the in <<
1309 */ 1359 */
1310 int 1360 int
1311 xfs_trans_alloc_dir( 1361 xfs_trans_alloc_dir(
1312 struct xfs_inode *dp, 1362 struct xfs_inode *dp,
1313 struct xfs_trans_res *resv, 1363 struct xfs_trans_res *resv,
1314 struct xfs_inode *ip, 1364 struct xfs_inode *ip,
1315 unsigned int *dblocks, 1365 unsigned int *dblocks,
1316 struct xfs_trans **tpp, 1366 struct xfs_trans **tpp,
1317 int *nospace_erro 1367 int *nospace_error)
1318 { 1368 {
1319 struct xfs_trans *tp; 1369 struct xfs_trans *tp;
1320 struct xfs_mount *mp = ip->i_m 1370 struct xfs_mount *mp = ip->i_mount;
1321 unsigned int resblks; 1371 unsigned int resblks;
1322 bool retried = fal 1372 bool retried = false;
1323 int error; 1373 int error;
1324 1374
1325 retry: 1375 retry:
1326 *nospace_error = 0; 1376 *nospace_error = 0;
1327 resblks = *dblocks; 1377 resblks = *dblocks;
1328 error = xfs_trans_alloc(mp, resv, res 1378 error = xfs_trans_alloc(mp, resv, resblks, 0, 0, &tp);
1329 if (error == -ENOSPC) { 1379 if (error == -ENOSPC) {
1330 *nospace_error = error; 1380 *nospace_error = error;
1331 resblks = 0; 1381 resblks = 0;
1332 error = xfs_trans_alloc(mp, r 1382 error = xfs_trans_alloc(mp, resv, resblks, 0, 0, &tp);
1333 } 1383 }
1334 if (error) 1384 if (error)
1335 return error; 1385 return error;
1336 1386
1337 xfs_lock_two_inodes(dp, XFS_ILOCK_EXC 1387 xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL);
1338 1388
1339 xfs_trans_ijoin(tp, dp, 0); !! 1389 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1340 xfs_trans_ijoin(tp, ip, 0); !! 1390 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1341 1391
1342 error = xfs_qm_dqattach_locked(dp, fa 1392 error = xfs_qm_dqattach_locked(dp, false);
1343 if (error) { 1393 if (error) {
1344 /* Caller should have allocat 1394 /* Caller should have allocated the dquots! */
1345 ASSERT(error != -ENOENT); 1395 ASSERT(error != -ENOENT);
1346 goto out_cancel; 1396 goto out_cancel;
1347 } 1397 }
1348 1398
1349 error = xfs_qm_dqattach_locked(ip, fa 1399 error = xfs_qm_dqattach_locked(ip, false);
1350 if (error) { 1400 if (error) {
1351 /* Caller should have allocat 1401 /* Caller should have allocated the dquots! */
1352 ASSERT(error != -ENOENT); 1402 ASSERT(error != -ENOENT);
1353 goto out_cancel; 1403 goto out_cancel;
1354 } 1404 }
1355 1405
1356 if (resblks == 0) 1406 if (resblks == 0)
1357 goto done; 1407 goto done;
1358 1408
1359 error = xfs_trans_reserve_quota_nblks 1409 error = xfs_trans_reserve_quota_nblks(tp, dp, resblks, 0, false);
1360 if (error == -EDQUOT || error == -ENO 1410 if (error == -EDQUOT || error == -ENOSPC) {
1361 if (!retried) { 1411 if (!retried) {
1362 xfs_trans_cancel(tp); 1412 xfs_trans_cancel(tp);
1363 xfs_iunlock(dp, XFS_I <<
1364 if (dp != ip) <<
1365 xfs_iunlock(i <<
1366 xfs_blockgc_free_quot 1413 xfs_blockgc_free_quota(dp, 0);
1367 retried = true; 1414 retried = true;
1368 goto retry; 1415 goto retry;
1369 } 1416 }
1370 1417
1371 *nospace_error = error; 1418 *nospace_error = error;
1372 resblks = 0; 1419 resblks = 0;
1373 error = 0; 1420 error = 0;
1374 } 1421 }
1375 if (error) 1422 if (error)
1376 goto out_cancel; 1423 goto out_cancel;
1377 1424
1378 done: 1425 done:
1379 *tpp = tp; 1426 *tpp = tp;
1380 *dblocks = resblks; 1427 *dblocks = resblks;
1381 return 0; 1428 return 0;
1382 1429
1383 out_cancel: 1430 out_cancel:
1384 xfs_trans_cancel(tp); 1431 xfs_trans_cancel(tp);
1385 return error; 1432 return error;
1386 } 1433 }
1387 1434
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.