1 // SPDX-License-Identifier: GPL-2.0 1 // SPDX-License-Identifier: GPL-2.0
2 /* 2 /*
3 * Copyright (c) 2000-2005 Silicon Graphics, I 3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * All Rights Reserved. 4 * All Rights Reserved.
5 */ 5 */
6 #include "xfs.h" 6 #include "xfs.h"
7 #include "xfs_fs.h" 7 #include "xfs_fs.h"
8 #include "xfs_shared.h" 8 #include "xfs_shared.h"
9 #include "xfs_format.h" 9 #include "xfs_format.h"
10 #include "xfs_log_format.h" 10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h" 11 #include "xfs_trans_resv.h"
12 #include "xfs_sb.h" 12 #include "xfs_sb.h"
13 #include "xfs_mount.h" 13 #include "xfs_mount.h"
14 #include "xfs_trans.h" 14 #include "xfs_trans.h"
15 #include "xfs_error.h" 15 #include "xfs_error.h"
16 #include "xfs_alloc.h" 16 #include "xfs_alloc.h"
17 #include "xfs_fsops.h" 17 #include "xfs_fsops.h"
18 #include "xfs_trans_space.h" 18 #include "xfs_trans_space.h"
19 #include "xfs_log.h" 19 #include "xfs_log.h"
20 #include "xfs_log_priv.h" <<
21 #include "xfs_ag.h" 20 #include "xfs_ag.h"
22 #include "xfs_ag_resv.h" 21 #include "xfs_ag_resv.h"
23 #include "xfs_trace.h" <<
24 22
25 /* 23 /*
26 * Write new AG headers to disk. Non-transacti 24 * Write new AG headers to disk. Non-transactional, but need to be
27 * written and completed prior to the growfs t 25 * written and completed prior to the growfs transaction being logged.
28 * To do this, we use a delayed write buffer l 26 * To do this, we use a delayed write buffer list and wait for
29 * submission and IO completion of the list as 27 * submission and IO completion of the list as a whole. This allows the
30 * IO subsystem to merge all the AG headers in 28 * IO subsystem to merge all the AG headers in a single AG into a single
31 * IO and hide most of the latency of the IO f 29 * IO and hide most of the latency of the IO from us.
32 * 30 *
33 * This also means that if we get an error whi 31 * This also means that if we get an error whilst building the buffer
34 * list to write, we can cancel the entire lis 32 * list to write, we can cancel the entire list without having written
35 * anything. 33 * anything.
36 */ 34 */
37 static int 35 static int
38 xfs_resizefs_init_new_ags( 36 xfs_resizefs_init_new_ags(
39 struct xfs_trans *tp, 37 struct xfs_trans *tp,
40 struct aghdr_init_data *id, 38 struct aghdr_init_data *id,
41 xfs_agnumber_t oagcount, 39 xfs_agnumber_t oagcount,
42 xfs_agnumber_t nagcount, 40 xfs_agnumber_t nagcount,
43 xfs_rfsblock_t delta, 41 xfs_rfsblock_t delta,
44 struct xfs_perag *last_pag, <<
45 bool *lastag_extend 42 bool *lastag_extended)
46 { 43 {
47 struct xfs_mount *mp = tp->t_mo 44 struct xfs_mount *mp = tp->t_mountp;
48 xfs_rfsblock_t nb = mp->m_sb. 45 xfs_rfsblock_t nb = mp->m_sb.sb_dblocks + delta;
49 int error; 46 int error;
50 47
51 *lastag_extended = false; 48 *lastag_extended = false;
52 49
53 INIT_LIST_HEAD(&id->buffer_list); 50 INIT_LIST_HEAD(&id->buffer_list);
54 for (id->agno = nagcount - 1; 51 for (id->agno = nagcount - 1;
55 id->agno >= oagcount; 52 id->agno >= oagcount;
56 id->agno--, delta -= id->agsize) 53 id->agno--, delta -= id->agsize) {
57 54
58 if (id->agno == nagcount - 1) 55 if (id->agno == nagcount - 1)
59 id->agsize = nb - (id- 56 id->agsize = nb - (id->agno *
60 (xfs_r 57 (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
61 else 58 else
62 id->agsize = mp->m_sb. 59 id->agsize = mp->m_sb.sb_agblocks;
63 60
64 error = xfs_ag_init_headers(mp 61 error = xfs_ag_init_headers(mp, id);
65 if (error) { 62 if (error) {
66 xfs_buf_delwri_cancel( 63 xfs_buf_delwri_cancel(&id->buffer_list);
67 return error; 64 return error;
68 } 65 }
69 } 66 }
70 67
71 error = xfs_buf_delwri_submit(&id->buf 68 error = xfs_buf_delwri_submit(&id->buffer_list);
72 if (error) 69 if (error)
73 return error; 70 return error;
74 71
75 if (delta) { 72 if (delta) {
76 *lastag_extended = true; 73 *lastag_extended = true;
77 error = xfs_ag_extend_space(la !! 74 error = xfs_ag_extend_space(mp, tp, id, delta);
78 } 75 }
79 return error; 76 return error;
80 } 77 }
81 78
82 /* 79 /*
83 * growfs operations 80 * growfs operations
84 */ 81 */
85 static int 82 static int
86 xfs_growfs_data_private( 83 xfs_growfs_data_private(
87 struct xfs_mount *mp, 84 struct xfs_mount *mp, /* mount point for filesystem */
88 struct xfs_growfs_data *in) 85 struct xfs_growfs_data *in) /* growfs data input struct */
89 { 86 {
90 xfs_agnumber_t oagcount = mp- <<
91 struct xfs_buf *bp; 87 struct xfs_buf *bp;
92 int error; 88 int error;
93 xfs_agnumber_t nagcount; 89 xfs_agnumber_t nagcount;
94 xfs_agnumber_t nagimax = 0; 90 xfs_agnumber_t nagimax = 0;
95 xfs_rfsblock_t nb, nb_div, nb 91 xfs_rfsblock_t nb, nb_div, nb_mod;
96 int64_t delta; 92 int64_t delta;
97 bool lastag_extende !! 93 bool lastag_extended;
>> 94 xfs_agnumber_t oagcount;
98 struct xfs_trans *tp; 95 struct xfs_trans *tp;
99 struct aghdr_init_data id = {}; 96 struct aghdr_init_data id = {};
100 struct xfs_perag *last_pag; <<
101 97
102 nb = in->newblocks; 98 nb = in->newblocks;
103 error = xfs_sb_validate_fsb_count(&mp- 99 error = xfs_sb_validate_fsb_count(&mp->m_sb, nb);
104 if (error) 100 if (error)
105 return error; 101 return error;
106 102
107 if (nb > mp->m_sb.sb_dblocks) { 103 if (nb > mp->m_sb.sb_dblocks) {
108 error = xfs_buf_read_uncached( 104 error = xfs_buf_read_uncached(mp->m_ddev_targp,
109 XFS_FSB_TO_BB( 105 XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
110 XFS_FSS_TO_BB( 106 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
111 if (error) 107 if (error)
112 return error; 108 return error;
113 xfs_buf_relse(bp); 109 xfs_buf_relse(bp);
114 } 110 }
115 111
116 nb_div = nb; 112 nb_div = nb;
117 nb_mod = do_div(nb_div, mp->m_sb.sb_ag 113 nb_mod = do_div(nb_div, mp->m_sb.sb_agblocks);
118 if (nb_mod && nb_mod >= XFS_MIN_AG_BLO !! 114 nagcount = nb_div + (nb_mod != 0);
119 nb_div++; !! 115 if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
120 else if (nb_mod) !! 116 nagcount--;
121 nb = nb_div * mp->m_sb.sb_agbl !! 117 nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
122 <<
123 if (nb_div > XFS_MAX_AGNUMBER + 1) { <<
124 nb_div = XFS_MAX_AGNUMBER + 1; <<
125 nb = nb_div * mp->m_sb.sb_agbl <<
126 } 118 }
127 nagcount = nb_div; <<
128 delta = nb - mp->m_sb.sb_dblocks; 119 delta = nb - mp->m_sb.sb_dblocks;
129 /* 120 /*
130 * Reject filesystems with a single AG 121 * Reject filesystems with a single AG because they are not
131 * supported, and reject a shrink oper 122 * supported, and reject a shrink operation that would cause a
132 * filesystem to become unsupported. 123 * filesystem to become unsupported.
133 */ 124 */
134 if (delta < 0 && nagcount < 2) 125 if (delta < 0 && nagcount < 2)
135 return -EINVAL; 126 return -EINVAL;
136 127
137 /* No work to do */ !! 128 oagcount = mp->m_sb.sb_agcount;
138 if (delta == 0) <<
139 return 0; <<
140 129
141 /* TODO: shrinking the entire AGs hasn !! 130 /* allocate the new per-ag structures */
142 if (nagcount < oagcount) !! 131 if (nagcount > oagcount) {
>> 132 error = xfs_initialize_perag(mp, nagcount, &nagimax);
>> 133 if (error)
>> 134 return error;
>> 135 } else if (nagcount < oagcount) {
>> 136 /* TODO: shrinking the entire AGs hasn't yet completed */
143 return -EINVAL; 137 return -EINVAL;
>> 138 }
144 139
145 /* allocate the new per-ag structures !! 140 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
146 error = xfs_initialize_perag(mp, oagco !! 141 (delta > 0 ? XFS_GROWFS_SPACE_RES(mp) : -delta), 0,
>> 142 XFS_TRANS_RESERVE, &tp);
147 if (error) 143 if (error)
148 return error; 144 return error;
149 145
150 if (delta > 0) <<
151 error = xfs_trans_alloc(mp, &M <<
152 XFS_GROWFS_SPA <<
153 &tp); <<
154 else <<
155 error = xfs_trans_alloc(mp, &M <<
156 0, &tp); <<
157 if (error) <<
158 goto out_free_unused_perag; <<
159 <<
160 last_pag = xfs_perag_get(mp, oagcount <<
161 if (delta > 0) { 146 if (delta > 0) {
162 error = xfs_resizefs_init_new_ 147 error = xfs_resizefs_init_new_ags(tp, &id, oagcount, nagcount,
163 delta, last_pa !! 148 delta, &lastag_extended);
164 } else { 149 } else {
165 xfs_warn_mount(mp, XFS_OPSTATE !! 150 static struct ratelimit_state shrink_warning = \
>> 151 RATELIMIT_STATE_INIT("shrink_warning", 86400 * HZ, 1);
>> 152 ratelimit_set_flags(&shrink_warning, RATELIMIT_MSG_ON_RELEASE);
>> 153
>> 154 if (__ratelimit(&shrink_warning))
>> 155 xfs_alert(mp,
166 "EXPERIMENTAL online shrink feature in 156 "EXPERIMENTAL online shrink feature in use. Use at your own risk!");
167 157
168 error = xfs_ag_shrink_space(la !! 158 error = xfs_ag_shrink_space(mp, &tp, nagcount - 1, -delta);
169 } 159 }
170 xfs_perag_put(last_pag); <<
171 if (error) 160 if (error)
172 goto out_trans_cancel; 161 goto out_trans_cancel;
173 162
174 /* 163 /*
175 * Update changed superblock fields tr 164 * Update changed superblock fields transactionally. These are not
176 * seen by the rest of the world until 165 * seen by the rest of the world until the transaction commit applies
177 * them atomically to the superblock. 166 * them atomically to the superblock.
178 */ 167 */
179 if (nagcount > oagcount) 168 if (nagcount > oagcount)
180 xfs_trans_mod_sb(tp, XFS_TRANS 169 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
181 if (delta) 170 if (delta)
182 xfs_trans_mod_sb(tp, XFS_TRANS 171 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS, delta);
183 if (id.nfree) 172 if (id.nfree)
184 xfs_trans_mod_sb(tp, XFS_TRANS 173 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, id.nfree);
185 174
186 /* 175 /*
187 * Sync sb counters now to reflect the 176 * Sync sb counters now to reflect the updated values. This is
188 * particularly important for shrink b 177 * particularly important for shrink because the write verifier
189 * will fail if sb_fdblocks is ever la 178 * will fail if sb_fdblocks is ever larger than sb_dblocks.
190 */ 179 */
191 if (xfs_has_lazysbcount(mp)) !! 180 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
192 xfs_log_sb(tp); 181 xfs_log_sb(tp);
193 182
194 xfs_trans_set_sync(tp); 183 xfs_trans_set_sync(tp);
195 error = xfs_trans_commit(tp); 184 error = xfs_trans_commit(tp);
196 if (error) 185 if (error)
197 return error; 186 return error;
198 187
199 /* New allocation groups fully initial 188 /* New allocation groups fully initialized, so update mount struct */
200 if (nagimax) 189 if (nagimax)
201 mp->m_maxagi = nagimax; 190 mp->m_maxagi = nagimax;
202 xfs_set_low_space_thresholds(mp); 191 xfs_set_low_space_thresholds(mp);
203 mp->m_alloc_set_aside = xfs_alloc_set_ 192 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
204 193
205 if (delta > 0) { 194 if (delta > 0) {
206 /* 195 /*
207 * If we expanded the last AG, 196 * If we expanded the last AG, free the per-AG reservation
208 * so we can reinitialize it w 197 * so we can reinitialize it with the new size.
209 */ 198 */
210 if (lastag_extended) { 199 if (lastag_extended) {
211 struct xfs_perag 200 struct xfs_perag *pag;
212 201
213 pag = xfs_perag_get(mp 202 pag = xfs_perag_get(mp, id.agno);
214 xfs_ag_resv_free(pag); !! 203 error = xfs_ag_resv_free(pag);
215 xfs_perag_put(pag); 204 xfs_perag_put(pag);
>> 205 if (error)
>> 206 return error;
216 } 207 }
217 /* 208 /*
218 * Reserve AG metadata blocks. 209 * Reserve AG metadata blocks. ENOSPC here does not mean there
219 * was a growfs failure, just 210 * was a growfs failure, just that there still isn't space for
220 * new user data after the gro 211 * new user data after the grow has been run.
221 */ 212 */
222 error = xfs_fs_reserve_ag_bloc 213 error = xfs_fs_reserve_ag_blocks(mp);
223 if (error == -ENOSPC) 214 if (error == -ENOSPC)
224 error = 0; 215 error = 0;
225 } 216 }
226 return error; 217 return error;
227 218
228 out_trans_cancel: 219 out_trans_cancel:
229 xfs_trans_cancel(tp); 220 xfs_trans_cancel(tp);
230 out_free_unused_perag: <<
231 if (nagcount > oagcount) <<
232 xfs_free_perag_range(mp, oagco <<
233 return error; 221 return error;
234 } 222 }
235 223
236 static int 224 static int
237 xfs_growfs_log_private( 225 xfs_growfs_log_private(
238 struct xfs_mount *mp, /* mou 226 struct xfs_mount *mp, /* mount point for filesystem */
239 struct xfs_growfs_log *in) /* gro 227 struct xfs_growfs_log *in) /* growfs log input struct */
240 { 228 {
241 xfs_extlen_t nb; 229 xfs_extlen_t nb;
242 230
243 nb = in->newblocks; 231 nb = in->newblocks;
244 if (nb < XFS_MIN_LOG_BLOCKS || nb < XF 232 if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
245 return -EINVAL; 233 return -EINVAL;
246 if (nb == mp->m_sb.sb_logblocks && 234 if (nb == mp->m_sb.sb_logblocks &&
247 in->isint == (mp->m_sb.sb_logstart 235 in->isint == (mp->m_sb.sb_logstart != 0))
248 return -EINVAL; 236 return -EINVAL;
249 /* 237 /*
250 * Moving the log is hard, need new in 238 * Moving the log is hard, need new interfaces to sync
251 * the log first, hold off all activit 239 * the log first, hold off all activity while moving it.
252 * Can have shorter or longer log in t 240 * Can have shorter or longer log in the same space,
253 * or transform internal to external l 241 * or transform internal to external log or vice versa.
254 */ 242 */
255 return -ENOSYS; 243 return -ENOSYS;
256 } 244 }
257 245
258 static int 246 static int
259 xfs_growfs_imaxpct( 247 xfs_growfs_imaxpct(
260 struct xfs_mount *mp, 248 struct xfs_mount *mp,
261 __u32 imaxpct) 249 __u32 imaxpct)
262 { 250 {
263 struct xfs_trans *tp; 251 struct xfs_trans *tp;
264 int dpct; 252 int dpct;
265 int error; 253 int error;
266 254
267 if (imaxpct > 100) 255 if (imaxpct > 100)
268 return -EINVAL; 256 return -EINVAL;
269 257
270 error = xfs_trans_alloc(mp, &M_RES(mp) 258 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
271 XFS_GROWFS_SPACE_RES(m 259 XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
272 if (error) 260 if (error)
273 return error; 261 return error;
274 262
275 dpct = imaxpct - mp->m_sb.sb_imax_pct; 263 dpct = imaxpct - mp->m_sb.sb_imax_pct;
276 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAX 264 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
277 xfs_trans_set_sync(tp); 265 xfs_trans_set_sync(tp);
278 return xfs_trans_commit(tp); 266 return xfs_trans_commit(tp);
279 } 267 }
280 268
281 /* 269 /*
282 * protected versions of growfs function acqui 270 * protected versions of growfs function acquire and release locks on the mount
283 * point - exported through ioctls: XFS_IOC_FS 271 * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
284 * XFS_IOC_FSGROWFSRT 272 * XFS_IOC_FSGROWFSRT
285 */ 273 */
286 int 274 int
287 xfs_growfs_data( 275 xfs_growfs_data(
288 struct xfs_mount *mp, 276 struct xfs_mount *mp,
289 struct xfs_growfs_data *in) 277 struct xfs_growfs_data *in)
290 { 278 {
291 int error = 0; 279 int error = 0;
292 280
293 if (!capable(CAP_SYS_ADMIN)) 281 if (!capable(CAP_SYS_ADMIN))
294 return -EPERM; 282 return -EPERM;
295 if (!mutex_trylock(&mp->m_growlock)) 283 if (!mutex_trylock(&mp->m_growlock))
296 return -EWOULDBLOCK; 284 return -EWOULDBLOCK;
297 285
298 /* update imaxpct separately to the ph 286 /* update imaxpct separately to the physical grow of the filesystem */
299 if (in->imaxpct != mp->m_sb.sb_imax_pc 287 if (in->imaxpct != mp->m_sb.sb_imax_pct) {
300 error = xfs_growfs_imaxpct(mp, 288 error = xfs_growfs_imaxpct(mp, in->imaxpct);
301 if (error) 289 if (error)
302 goto out_error; 290 goto out_error;
303 } 291 }
304 292
305 if (in->newblocks != mp->m_sb.sb_dbloc 293 if (in->newblocks != mp->m_sb.sb_dblocks) {
306 error = xfs_growfs_data_privat 294 error = xfs_growfs_data_private(mp, in);
307 if (error) 295 if (error)
308 goto out_error; 296 goto out_error;
309 } 297 }
310 298
311 /* Post growfs calculations needed to 299 /* Post growfs calculations needed to reflect new state in operations */
312 if (mp->m_sb.sb_imax_pct) { 300 if (mp->m_sb.sb_imax_pct) {
313 uint64_t icount = mp->m_sb.sb_ 301 uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
314 do_div(icount, 100); 302 do_div(icount, 100);
315 M_IGEO(mp)->maxicount = XFS_FS 303 M_IGEO(mp)->maxicount = XFS_FSB_TO_INO(mp, icount);
316 } else 304 } else
317 M_IGEO(mp)->maxicount = 0; 305 M_IGEO(mp)->maxicount = 0;
318 306
319 /* Update secondary superblocks now th 307 /* Update secondary superblocks now the physical grow has completed */
320 error = xfs_update_secondary_sbs(mp); 308 error = xfs_update_secondary_sbs(mp);
321 309
322 out_error: 310 out_error:
323 /* 311 /*
324 * Increment the generation unconditio 312 * Increment the generation unconditionally, the error could be from
325 * updating the secondary superblocks, 313 * updating the secondary superblocks, in which case the new size
326 * is live already. 314 * is live already.
327 */ 315 */
328 mp->m_generation++; 316 mp->m_generation++;
329 mutex_unlock(&mp->m_growlock); 317 mutex_unlock(&mp->m_growlock);
330 return error; 318 return error;
331 } 319 }
332 320
333 int 321 int
334 xfs_growfs_log( 322 xfs_growfs_log(
335 xfs_mount_t *mp, 323 xfs_mount_t *mp,
336 struct xfs_growfs_log *in) 324 struct xfs_growfs_log *in)
337 { 325 {
338 int error; 326 int error;
339 327
340 if (!capable(CAP_SYS_ADMIN)) 328 if (!capable(CAP_SYS_ADMIN))
341 return -EPERM; 329 return -EPERM;
342 if (!mutex_trylock(&mp->m_growlock)) 330 if (!mutex_trylock(&mp->m_growlock))
343 return -EWOULDBLOCK; 331 return -EWOULDBLOCK;
344 error = xfs_growfs_log_private(mp, in) 332 error = xfs_growfs_log_private(mp, in);
345 mutex_unlock(&mp->m_growlock); 333 mutex_unlock(&mp->m_growlock);
346 return error; 334 return error;
347 } 335 }
348 336
349 /* 337 /*
>> 338 * exported through ioctl XFS_IOC_FSCOUNTS
>> 339 */
>> 340
>> 341 void
>> 342 xfs_fs_counts(
>> 343 xfs_mount_t *mp,
>> 344 xfs_fsop_counts_t *cnt)
>> 345 {
>> 346 cnt->allocino = percpu_counter_read_positive(&mp->m_icount);
>> 347 cnt->freeino = percpu_counter_read_positive(&mp->m_ifree);
>> 348 cnt->freedata = percpu_counter_read_positive(&mp->m_fdblocks) -
>> 349 mp->m_alloc_set_aside;
>> 350
>> 351 spin_lock(&mp->m_sb_lock);
>> 352 cnt->freertx = mp->m_sb.sb_frextents;
>> 353 spin_unlock(&mp->m_sb_lock);
>> 354 }
>> 355
>> 356 /*
>> 357 * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
>> 358 *
>> 359 * xfs_reserve_blocks is called to set m_resblks
>> 360 * in the in-core mount table. The number of unused reserved blocks
>> 361 * is kept in m_resblks_avail.
>> 362 *
350 * Reserve the requested number of blocks if a 363 * Reserve the requested number of blocks if available. Otherwise return
351 * as many as possible to satisfy the request. 364 * as many as possible to satisfy the request. The actual number
352 * reserved are returned in outval. !! 365 * reserved are returned in outval
>> 366 *
>> 367 * A null inval pointer indicates that only the current reserved blocks
>> 368 * available should be returned no settings are changed.
353 */ 369 */
>> 370
354 int 371 int
355 xfs_reserve_blocks( 372 xfs_reserve_blocks(
356 struct xfs_mount *mp, !! 373 xfs_mount_t *mp,
357 uint64_t request) !! 374 uint64_t *inval,
>> 375 xfs_fsop_resblks_t *outval)
358 { 376 {
359 int64_t lcounter, delt 377 int64_t lcounter, delta;
360 int64_t fdblks_delta = 378 int64_t fdblks_delta = 0;
>> 379 uint64_t request;
361 int64_t free; 380 int64_t free;
362 int error = 0; 381 int error = 0;
363 382
>> 383 /* If inval is null, report current values and return */
>> 384 if (inval == (uint64_t *)NULL) {
>> 385 if (!outval)
>> 386 return -EINVAL;
>> 387 outval->resblks = mp->m_resblks;
>> 388 outval->resblks_avail = mp->m_resblks_avail;
>> 389 return 0;
>> 390 }
>> 391
>> 392 request = *inval;
>> 393
364 /* 394 /*
365 * With per-cpu counters, this becomes 395 * With per-cpu counters, this becomes an interesting problem. we need
366 * to work out if we are freeing or al 396 * to work out if we are freeing or allocation blocks first, then we can
367 * do the modification as necessary. 397 * do the modification as necessary.
368 * 398 *
369 * We do this under the m_sb_lock so t 399 * We do this under the m_sb_lock so that if we are near ENOSPC, we will
370 * hold out any changes while we work 400 * hold out any changes while we work out what to do. This means that
371 * the amount of free space can change 401 * the amount of free space can change while we do this, so we need to
372 * retry if we end up trying to reserv 402 * retry if we end up trying to reserve more space than is available.
373 */ 403 */
374 spin_lock(&mp->m_sb_lock); 404 spin_lock(&mp->m_sb_lock);
375 405
376 /* 406 /*
377 * If our previous reservation was lar 407 * If our previous reservation was larger than the current value,
378 * then move any unused blocks back to 408 * then move any unused blocks back to the free pool. Modify the resblks
379 * counters directly since we shouldn' 409 * counters directly since we shouldn't have any problems unreserving
380 * space. 410 * space.
381 */ 411 */
382 if (mp->m_resblks > request) { 412 if (mp->m_resblks > request) {
383 lcounter = mp->m_resblks_avail 413 lcounter = mp->m_resblks_avail - request;
384 if (lcounter > 0) { !! 414 if (lcounter > 0) { /* release unused blocks */
385 fdblks_delta = lcounte 415 fdblks_delta = lcounter;
386 mp->m_resblks_avail -= 416 mp->m_resblks_avail -= lcounter;
387 } 417 }
388 mp->m_resblks = request; 418 mp->m_resblks = request;
389 if (fdblks_delta) { 419 if (fdblks_delta) {
390 spin_unlock(&mp->m_sb_ 420 spin_unlock(&mp->m_sb_lock);
391 xfs_add_fdblocks(mp, f !! 421 error = xfs_mod_fdblocks(mp, fdblks_delta, 0);
392 spin_lock(&mp->m_sb_lo 422 spin_lock(&mp->m_sb_lock);
393 } 423 }
394 424
395 goto out; 425 goto out;
396 } 426 }
397 427
398 /* 428 /*
399 * If the request is larger than the c 429 * If the request is larger than the current reservation, reserve the
400 * blocks before we update the reserve 430 * blocks before we update the reserve counters. Sample m_fdblocks and
401 * perform a partial reservation if th 431 * perform a partial reservation if the request exceeds free space.
402 * !! 432 */
403 * The code below estimates how many b !! 433 error = -ENOSPC;
404 * fdblocks to stash in the reserve po !! 434 do {
405 * race since fdblocks updates are not !! 435 free = percpu_counter_sum(&mp->m_fdblocks) -
406 * m_sb_lock. Set the reserve size ev !! 436 mp->m_alloc_set_aside;
407 * space to fill it because mod_fdbloc !! 437 if (free <= 0)
408 * reserve when it can. !! 438 break;
409 */ !! 439
410 free = percpu_counter_sum(&mp->m_fdblo !! 440 delta = request - mp->m_resblks;
411 !! 441 lcounter = free - delta;
412 delta = request - mp->m_resblks; !! 442 if (lcounter < 0)
413 mp->m_resblks = request; !! 443 /* We can't satisfy the request, just get what we can */
414 if (delta > 0 && free > 0) { !! 444 fdblks_delta = free;
>> 445 else
>> 446 fdblks_delta = delta;
>> 447
415 /* 448 /*
416 * We'll either succeed in get 449 * We'll either succeed in getting space from the free block
417 * count or we'll get an ENOSP !! 450 * count or we'll get an ENOSPC. If we get a ENOSPC, it means
418 * here - we don't want to res !! 451 * things changed while we were calculating fdblks_delta and so
419 * from the reserve. !! 452 * we should try again to see if there is anything left to
>> 453 * reserve.
420 * 454 *
421 * The desired reserve size ca !! 455 * Don't set the reserved flag here - we don't want to reserve
422 * Use mod_fdblocks to put the !! 456 * the extra reserve blocks from the reserve.....
423 * fdblocks as appropriate. <<
424 */ 457 */
425 fdblks_delta = min(free, delta <<
426 spin_unlock(&mp->m_sb_lock); 458 spin_unlock(&mp->m_sb_lock);
427 error = xfs_dec_fdblocks(mp, f !! 459 error = xfs_mod_fdblocks(mp, -fdblks_delta, 0);
428 if (!error) <<
429 xfs_add_fdblocks(mp, f <<
430 spin_lock(&mp->m_sb_lock); 460 spin_lock(&mp->m_sb_lock);
>> 461 } while (error == -ENOSPC);
>> 462
>> 463 /*
>> 464 * Update the reserve counters if blocks have been successfully
>> 465 * allocated.
>> 466 */
>> 467 if (!error && fdblks_delta) {
>> 468 mp->m_resblks += fdblks_delta;
>> 469 mp->m_resblks_avail += fdblks_delta;
431 } 470 }
>> 471
432 out: 472 out:
>> 473 if (outval) {
>> 474 outval->resblks = mp->m_resblks;
>> 475 outval->resblks_avail = mp->m_resblks_avail;
>> 476 }
>> 477
433 spin_unlock(&mp->m_sb_lock); 478 spin_unlock(&mp->m_sb_lock);
434 return error; 479 return error;
435 } 480 }
436 481
437 int 482 int
438 xfs_fs_goingdown( 483 xfs_fs_goingdown(
439 xfs_mount_t *mp, 484 xfs_mount_t *mp,
440 uint32_t inflags) 485 uint32_t inflags)
441 { 486 {
442 switch (inflags) { 487 switch (inflags) {
443 case XFS_FSOP_GOING_FLAGS_DEFAULT: { 488 case XFS_FSOP_GOING_FLAGS_DEFAULT: {
444 if (!bdev_freeze(mp->m_super-> !! 489 if (!freeze_bdev(mp->m_super->s_bdev)) {
445 xfs_force_shutdown(mp, 490 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
446 bdev_thaw(mp->m_super- !! 491 thaw_bdev(mp->m_super->s_bdev);
447 } 492 }
448 break; 493 break;
449 } 494 }
450 case XFS_FSOP_GOING_FLAGS_LOGFLUSH: 495 case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
451 xfs_force_shutdown(mp, SHUTDOW 496 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
452 break; 497 break;
453 case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH: 498 case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
454 xfs_force_shutdown(mp, 499 xfs_force_shutdown(mp,
455 SHUTDOWN_FORCE 500 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
456 break; 501 break;
457 default: 502 default:
458 return -EINVAL; 503 return -EINVAL;
459 } 504 }
460 505
461 return 0; 506 return 0;
462 } 507 }
463 508
464 /* 509 /*
465 * Force a shutdown of the filesystem instantl 510 * Force a shutdown of the filesystem instantly while keeping the filesystem
466 * consistent. We don't do an unmount here; ju 511 * consistent. We don't do an unmount here; just shutdown the shop, make sure
467 * that absolutely nothing persistent happens 512 * that absolutely nothing persistent happens to this filesystem after this
468 * point. 513 * point.
469 * <<
470 * The shutdown state change is atomic, result <<
471 * first shutdown call processing the shutdown <<
472 * log once as it requires, and we don't spam <<
473 * shutdowns race to set the shutdown flags. <<
474 */ 514 */
475 void 515 void
476 xfs_do_force_shutdown( 516 xfs_do_force_shutdown(
477 struct xfs_mount *mp, 517 struct xfs_mount *mp,
478 uint32_t flags, !! 518 int flags,
479 char *fname, 519 char *fname,
480 int lnnum) 520 int lnnum)
481 { 521 {
482 int tag; !! 522 bool logerror = flags & SHUTDOWN_LOG_IO_ERROR;
483 const char *why; <<
484 523
>> 524 /*
>> 525 * No need to duplicate efforts.
>> 526 */
>> 527 if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
>> 528 return;
485 529
486 if (xfs_set_shutdown(mp)) { !! 530 /*
487 xlog_shutdown_wait(mp->m_log); !! 531 * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
>> 532 * queue up anybody new on the log reservations, and wakes up
>> 533 * everybody who's sleeping on log reservations to tell them
>> 534 * the bad news.
>> 535 */
>> 536 if (xfs_log_force_umount(mp, logerror))
488 return; 537 return;
489 } <<
490 if (mp->m_sb_bp) <<
491 mp->m_sb_bp->b_flags |= XBF_DO <<
492 538
493 if (flags & SHUTDOWN_FORCE_UMOUNT) !! 539 if (flags & SHUTDOWN_FORCE_UMOUNT) {
494 xfs_alert(mp, "User initiated !! 540 xfs_alert(mp,
>> 541 "User initiated shutdown (0x%x) received. Shutting down filesystem",
>> 542 flags);
>> 543 return;
>> 544 }
495 545
496 if (xlog_force_shutdown(mp->m_log, fla !! 546 if (flags & SHUTDOWN_CORRUPT_INCORE) {
497 tag = XFS_PTAG_SHUTDOWN_LOGERR !! 547 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT,
498 why = "Log I/O Error"; !! 548 "Corruption of in-memory data (0x%x) detected at %pS (%s:%d). Shutting down filesystem",
499 } else if (flags & SHUTDOWN_CORRUPT_IN !! 549 flags, __return_address, fname, lnnum);
500 tag = XFS_PTAG_SHUTDOWN_CORRUP !! 550 if (XFS_ERRLEVEL_HIGH <= xfs_error_level)
501 why = "Corruption of in-memory !! 551 xfs_stack_trace();
502 } else if (flags & SHUTDOWN_CORRUPT_ON !! 552 } else if (logerror) {
503 tag = XFS_PTAG_SHUTDOWN_CORRUP !! 553 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR,
504 why = "Corruption of on-disk m !! 554 "Log I/O error (0x%x) detected at %pS (%s:%d). Shutting down filesystem",
505 } else if (flags & SHUTDOWN_DEVICE_REM !! 555 flags, __return_address, fname, lnnum);
506 tag = XFS_PTAG_SHUTDOWN_IOERRO <<
507 why = "Block device removal"; <<
508 } else { 556 } else {
509 tag = XFS_PTAG_SHUTDOWN_IOERRO !! 557 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
510 why = "Metadata I/O Error"; !! 558 "I/O error (0x%x) detected at %pS (%s:%d). Shutting down filesystem",
>> 559 flags, __return_address, fname, lnnum);
511 } 560 }
512 561
513 trace_xfs_force_shutdown(mp, tag, flag <<
514 <<
515 xfs_alert_tag(mp, tag, <<
516 "%s (0x%x) detected at %pS (%s:%d). Shutting <<
517 why, flags, __return_a <<
518 xfs_alert(mp, 562 xfs_alert(mp,
519 "Please unmount the filesystem 563 "Please unmount the filesystem and rectify the problem(s)");
520 if (xfs_error_level >= XFS_ERRLEVEL_HI <<
521 xfs_stack_trace(); <<
522 } 564 }
523 565
524 /* 566 /*
525 * Reserve free space for per-AG metadata. 567 * Reserve free space for per-AG metadata.
526 */ 568 */
527 int 569 int
528 xfs_fs_reserve_ag_blocks( 570 xfs_fs_reserve_ag_blocks(
529 struct xfs_mount *mp) 571 struct xfs_mount *mp)
530 { 572 {
531 xfs_agnumber_t agno; 573 xfs_agnumber_t agno;
532 struct xfs_perag *pag; 574 struct xfs_perag *pag;
533 int error = 0; 575 int error = 0;
534 int err2; 576 int err2;
535 577
536 mp->m_finobt_nores = false; 578 mp->m_finobt_nores = false;
537 for_each_perag(mp, agno, pag) { 579 for_each_perag(mp, agno, pag) {
538 err2 = xfs_ag_resv_init(pag, N 580 err2 = xfs_ag_resv_init(pag, NULL);
539 if (err2 && !error) 581 if (err2 && !error)
540 error = err2; 582 error = err2;
541 } 583 }
542 584
543 if (error && error != -ENOSPC) { 585 if (error && error != -ENOSPC) {
544 xfs_warn(mp, 586 xfs_warn(mp,
545 "Error %d reserving per-AG metadata re 587 "Error %d reserving per-AG metadata reserve pool.", error);
546 xfs_force_shutdown(mp, SHUTDOW 588 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
547 } 589 }
548 590
549 return error; 591 return error;
550 } 592 }
551 593
552 /* 594 /*
553 * Free space reserved for per-AG metadata. 595 * Free space reserved for per-AG metadata.
554 */ 596 */
555 void !! 597 int
556 xfs_fs_unreserve_ag_blocks( 598 xfs_fs_unreserve_ag_blocks(
557 struct xfs_mount *mp) 599 struct xfs_mount *mp)
558 { 600 {
559 xfs_agnumber_t agno; 601 xfs_agnumber_t agno;
560 struct xfs_perag *pag; 602 struct xfs_perag *pag;
>> 603 int error = 0;
>> 604 int err2;
>> 605
>> 606 for_each_perag(mp, agno, pag) {
>> 607 err2 = xfs_ag_resv_free(pag);
>> 608 if (err2 && !error)
>> 609 error = err2;
>> 610 }
561 611
562 for_each_perag(mp, agno, pag) !! 612 if (error)
563 xfs_ag_resv_free(pag); !! 613 xfs_warn(mp,
>> 614 "Error %d freeing per-AG metadata reserve pool.", error);
>> 615
>> 616 return error;
564 } 617 }
565 618
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.