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