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" 20 #include "xfs_log_priv.h"
21 #include "xfs_ag.h" 21 #include "xfs_ag.h"
22 #include "xfs_ag_resv.h" 22 #include "xfs_ag_resv.h"
23 #include "xfs_trace.h" 23 #include "xfs_trace.h"
24 24
25 /* 25 /*
26 * Write new AG headers to disk. Non-transacti 26 * Write new AG headers to disk. Non-transactional, but need to be
27 * written and completed prior to the growfs t 27 * written and completed prior to the growfs transaction being logged.
28 * To do this, we use a delayed write buffer l 28 * To do this, we use a delayed write buffer list and wait for
29 * submission and IO completion of the list as 29 * submission and IO completion of the list as a whole. This allows the
30 * IO subsystem to merge all the AG headers in 30 * 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 31 * IO and hide most of the latency of the IO from us.
32 * 32 *
33 * This also means that if we get an error whi 33 * This also means that if we get an error whilst building the buffer
34 * list to write, we can cancel the entire lis 34 * list to write, we can cancel the entire list without having written
35 * anything. 35 * anything.
36 */ 36 */
37 static int 37 static int
38 xfs_resizefs_init_new_ags( 38 xfs_resizefs_init_new_ags(
39 struct xfs_trans *tp, 39 struct xfs_trans *tp,
40 struct aghdr_init_data *id, 40 struct aghdr_init_data *id,
41 xfs_agnumber_t oagcount, 41 xfs_agnumber_t oagcount,
42 xfs_agnumber_t nagcount, 42 xfs_agnumber_t nagcount,
43 xfs_rfsblock_t delta, 43 xfs_rfsblock_t delta,
44 struct xfs_perag *last_pag, 44 struct xfs_perag *last_pag,
45 bool *lastag_extend 45 bool *lastag_extended)
46 { 46 {
47 struct xfs_mount *mp = tp->t_mo 47 struct xfs_mount *mp = tp->t_mountp;
48 xfs_rfsblock_t nb = mp->m_sb. 48 xfs_rfsblock_t nb = mp->m_sb.sb_dblocks + delta;
49 int error; 49 int error;
50 50
51 *lastag_extended = false; 51 *lastag_extended = false;
52 52
53 INIT_LIST_HEAD(&id->buffer_list); 53 INIT_LIST_HEAD(&id->buffer_list);
54 for (id->agno = nagcount - 1; 54 for (id->agno = nagcount - 1;
55 id->agno >= oagcount; 55 id->agno >= oagcount;
56 id->agno--, delta -= id->agsize) 56 id->agno--, delta -= id->agsize) {
57 57
58 if (id->agno == nagcount - 1) 58 if (id->agno == nagcount - 1)
59 id->agsize = nb - (id- 59 id->agsize = nb - (id->agno *
60 (xfs_r 60 (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
61 else 61 else
62 id->agsize = mp->m_sb. 62 id->agsize = mp->m_sb.sb_agblocks;
63 63
64 error = xfs_ag_init_headers(mp 64 error = xfs_ag_init_headers(mp, id);
65 if (error) { 65 if (error) {
66 xfs_buf_delwri_cancel( 66 xfs_buf_delwri_cancel(&id->buffer_list);
67 return error; 67 return error;
68 } 68 }
69 } 69 }
70 70
71 error = xfs_buf_delwri_submit(&id->buf 71 error = xfs_buf_delwri_submit(&id->buffer_list);
72 if (error) 72 if (error)
73 return error; 73 return error;
74 74
75 if (delta) { 75 if (delta) {
76 *lastag_extended = true; 76 *lastag_extended = true;
77 error = xfs_ag_extend_space(la 77 error = xfs_ag_extend_space(last_pag, tp, delta);
78 } 78 }
79 return error; 79 return error;
80 } 80 }
81 81
82 /* 82 /*
83 * growfs operations 83 * growfs operations
84 */ 84 */
85 static int 85 static int
86 xfs_growfs_data_private( 86 xfs_growfs_data_private(
87 struct xfs_mount *mp, 87 struct xfs_mount *mp, /* mount point for filesystem */
88 struct xfs_growfs_data *in) 88 struct xfs_growfs_data *in) /* growfs data input struct */
89 { 89 {
90 xfs_agnumber_t oagcount = mp- <<
91 struct xfs_buf *bp; 90 struct xfs_buf *bp;
92 int error; 91 int error;
93 xfs_agnumber_t nagcount; 92 xfs_agnumber_t nagcount;
94 xfs_agnumber_t nagimax = 0; 93 xfs_agnumber_t nagimax = 0;
95 xfs_rfsblock_t nb, nb_div, nb 94 xfs_rfsblock_t nb, nb_div, nb_mod;
96 int64_t delta; 95 int64_t delta;
97 bool lastag_extende !! 96 bool lastag_extended;
>> 97 xfs_agnumber_t oagcount;
98 struct xfs_trans *tp; 98 struct xfs_trans *tp;
99 struct aghdr_init_data id = {}; 99 struct aghdr_init_data id = {};
100 struct xfs_perag *last_pag; 100 struct xfs_perag *last_pag;
101 101
102 nb = in->newblocks; 102 nb = in->newblocks;
103 error = xfs_sb_validate_fsb_count(&mp- 103 error = xfs_sb_validate_fsb_count(&mp->m_sb, nb);
104 if (error) 104 if (error)
105 return error; 105 return error;
106 106
107 if (nb > mp->m_sb.sb_dblocks) { 107 if (nb > mp->m_sb.sb_dblocks) {
108 error = xfs_buf_read_uncached( 108 error = xfs_buf_read_uncached(mp->m_ddev_targp,
109 XFS_FSB_TO_BB( 109 XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
110 XFS_FSS_TO_BB( 110 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
111 if (error) 111 if (error)
112 return error; 112 return error;
113 xfs_buf_relse(bp); 113 xfs_buf_relse(bp);
114 } 114 }
115 115
116 nb_div = nb; 116 nb_div = nb;
117 nb_mod = do_div(nb_div, mp->m_sb.sb_ag 117 nb_mod = do_div(nb_div, mp->m_sb.sb_agblocks);
118 if (nb_mod && nb_mod >= XFS_MIN_AG_BLO !! 118 nagcount = nb_div + (nb_mod != 0);
119 nb_div++; !! 119 if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
120 else if (nb_mod) !! 120 nagcount--;
121 nb = nb_div * mp->m_sb.sb_agbl !! 121 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 } 122 }
127 nagcount = nb_div; <<
128 delta = nb - mp->m_sb.sb_dblocks; 123 delta = nb - mp->m_sb.sb_dblocks;
129 /* 124 /*
130 * Reject filesystems with a single AG 125 * Reject filesystems with a single AG because they are not
131 * supported, and reject a shrink oper 126 * supported, and reject a shrink operation that would cause a
132 * filesystem to become unsupported. 127 * filesystem to become unsupported.
133 */ 128 */
134 if (delta < 0 && nagcount < 2) 129 if (delta < 0 && nagcount < 2)
135 return -EINVAL; 130 return -EINVAL;
136 131
137 /* No work to do */ !! 132 oagcount = mp->m_sb.sb_agcount;
138 if (delta == 0) !! 133 /* allocate the new per-ag structures */
139 return 0; !! 134 if (nagcount > oagcount) {
140 !! 135 error = xfs_initialize_perag(mp, nagcount, nb, &nagimax);
141 /* TODO: shrinking the entire AGs hasn !! 136 if (error)
142 if (nagcount < oagcount) !! 137 return error;
>> 138 } else if (nagcount < oagcount) {
>> 139 /* TODO: shrinking the entire AGs hasn't yet completed */
143 return -EINVAL; 140 return -EINVAL;
>> 141 }
144 142
145 /* allocate the new per-ag structures !! 143 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
146 error = xfs_initialize_perag(mp, oagco !! 144 (delta > 0 ? XFS_GROWFS_SPACE_RES(mp) : -delta), 0,
>> 145 XFS_TRANS_RESERVE, &tp);
147 if (error) 146 if (error)
148 return error; 147 return error;
149 148
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 149 last_pag = xfs_perag_get(mp, oagcount - 1);
161 if (delta > 0) { 150 if (delta > 0) {
162 error = xfs_resizefs_init_new_ 151 error = xfs_resizefs_init_new_ags(tp, &id, oagcount, nagcount,
163 delta, last_pa 152 delta, last_pag, &lastag_extended);
164 } else { 153 } else {
165 xfs_warn_mount(mp, XFS_OPSTATE 154 xfs_warn_mount(mp, XFS_OPSTATE_WARNED_SHRINK,
166 "EXPERIMENTAL online shrink feature in 155 "EXPERIMENTAL online shrink feature in use. Use at your own risk!");
167 156
168 error = xfs_ag_shrink_space(la 157 error = xfs_ag_shrink_space(last_pag, &tp, -delta);
169 } 158 }
170 xfs_perag_put(last_pag); 159 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_has_lazysbcount(mp))
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 xfs_fdblocks_unavailable(mp);
>> 350 cnt->freertx = percpu_counter_read_positive(&mp->m_frextents);
>> 351 }
>> 352
>> 353 /*
>> 354 * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
>> 355 *
>> 356 * xfs_reserve_blocks is called to set m_resblks
>> 357 * in the in-core mount table. The number of unused reserved blocks
>> 358 * is kept in m_resblks_avail.
>> 359 *
350 * Reserve the requested number of blocks if a 360 * Reserve the requested number of blocks if available. Otherwise return
351 * as many as possible to satisfy the request. 361 * as many as possible to satisfy the request. The actual number
352 * reserved are returned in outval. !! 362 * reserved are returned in outval
>> 363 *
>> 364 * A null inval pointer indicates that only the current reserved blocks
>> 365 * available should be returned no settings are changed.
353 */ 366 */
>> 367
354 int 368 int
355 xfs_reserve_blocks( 369 xfs_reserve_blocks(
356 struct xfs_mount *mp, !! 370 xfs_mount_t *mp,
357 uint64_t request) !! 371 uint64_t *inval,
>> 372 xfs_fsop_resblks_t *outval)
358 { 373 {
359 int64_t lcounter, delt 374 int64_t lcounter, delta;
360 int64_t fdblks_delta = 375 int64_t fdblks_delta = 0;
>> 376 uint64_t request;
361 int64_t free; 377 int64_t free;
362 int error = 0; 378 int error = 0;
363 379
>> 380 /* If inval is null, report current values and return */
>> 381 if (inval == (uint64_t *)NULL) {
>> 382 if (!outval)
>> 383 return -EINVAL;
>> 384 outval->resblks = mp->m_resblks;
>> 385 outval->resblks_avail = mp->m_resblks_avail;
>> 386 return 0;
>> 387 }
>> 388
>> 389 request = *inval;
>> 390
364 /* 391 /*
365 * With per-cpu counters, this becomes 392 * With per-cpu counters, this becomes an interesting problem. we need
366 * to work out if we are freeing or al 393 * to work out if we are freeing or allocation blocks first, then we can
367 * do the modification as necessary. 394 * do the modification as necessary.
368 * 395 *
369 * We do this under the m_sb_lock so t 396 * 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 397 * hold out any changes while we work out what to do. This means that
371 * the amount of free space can change 398 * the amount of free space can change while we do this, so we need to
372 * retry if we end up trying to reserv 399 * retry if we end up trying to reserve more space than is available.
373 */ 400 */
374 spin_lock(&mp->m_sb_lock); 401 spin_lock(&mp->m_sb_lock);
375 402
376 /* 403 /*
377 * If our previous reservation was lar 404 * If our previous reservation was larger than the current value,
378 * then move any unused blocks back to 405 * then move any unused blocks back to the free pool. Modify the resblks
379 * counters directly since we shouldn' 406 * counters directly since we shouldn't have any problems unreserving
380 * space. 407 * space.
381 */ 408 */
382 if (mp->m_resblks > request) { 409 if (mp->m_resblks > request) {
383 lcounter = mp->m_resblks_avail 410 lcounter = mp->m_resblks_avail - request;
384 if (lcounter > 0) { !! 411 if (lcounter > 0) { /* release unused blocks */
385 fdblks_delta = lcounte 412 fdblks_delta = lcounter;
386 mp->m_resblks_avail -= 413 mp->m_resblks_avail -= lcounter;
387 } 414 }
388 mp->m_resblks = request; 415 mp->m_resblks = request;
389 if (fdblks_delta) { 416 if (fdblks_delta) {
390 spin_unlock(&mp->m_sb_ 417 spin_unlock(&mp->m_sb_lock);
391 xfs_add_fdblocks(mp, f !! 418 error = xfs_mod_fdblocks(mp, fdblks_delta, 0);
392 spin_lock(&mp->m_sb_lo 419 spin_lock(&mp->m_sb_lock);
393 } 420 }
394 421
395 goto out; 422 goto out;
396 } 423 }
397 424
398 /* 425 /*
399 * If the request is larger than the c 426 * If the request is larger than the current reservation, reserve the
400 * blocks before we update the reserve 427 * blocks before we update the reserve counters. Sample m_fdblocks and
401 * perform a partial reservation if th 428 * perform a partial reservation if the request exceeds free space.
402 * 429 *
403 * The code below estimates how many b 430 * The code below estimates how many blocks it can request from
404 * fdblocks to stash in the reserve po 431 * fdblocks to stash in the reserve pool. This is a classic TOCTOU
405 * race since fdblocks updates are not 432 * race since fdblocks updates are not always coordinated via
406 * m_sb_lock. Set the reserve size ev 433 * m_sb_lock. Set the reserve size even if there's not enough free
407 * space to fill it because mod_fdbloc 434 * space to fill it because mod_fdblocks will refill an undersized
408 * reserve when it can. 435 * reserve when it can.
409 */ 436 */
410 free = percpu_counter_sum(&mp->m_fdblo 437 free = percpu_counter_sum(&mp->m_fdblocks) -
411 438 xfs_fdblocks_unavailable(mp);
412 delta = request - mp->m_resblks; 439 delta = request - mp->m_resblks;
413 mp->m_resblks = request; 440 mp->m_resblks = request;
414 if (delta > 0 && free > 0) { 441 if (delta > 0 && free > 0) {
415 /* 442 /*
416 * We'll either succeed in get 443 * We'll either succeed in getting space from the free block
417 * count or we'll get an ENOSP 444 * count or we'll get an ENOSPC. Don't set the reserved flag
418 * here - we don't want to res 445 * here - we don't want to reserve the extra reserve blocks
419 * from the reserve. 446 * from the reserve.
420 * 447 *
421 * The desired reserve size ca 448 * The desired reserve size can change after we drop the lock.
422 * Use mod_fdblocks to put the 449 * Use mod_fdblocks to put the space into the reserve or into
423 * fdblocks as appropriate. 450 * fdblocks as appropriate.
424 */ 451 */
425 fdblks_delta = min(free, delta 452 fdblks_delta = min(free, delta);
426 spin_unlock(&mp->m_sb_lock); 453 spin_unlock(&mp->m_sb_lock);
427 error = xfs_dec_fdblocks(mp, f !! 454 error = xfs_mod_fdblocks(mp, -fdblks_delta, 0);
428 if (!error) 455 if (!error)
429 xfs_add_fdblocks(mp, f !! 456 xfs_mod_fdblocks(mp, fdblks_delta, 0);
430 spin_lock(&mp->m_sb_lock); 457 spin_lock(&mp->m_sb_lock);
431 } 458 }
432 out: 459 out:
>> 460 if (outval) {
>> 461 outval->resblks = mp->m_resblks;
>> 462 outval->resblks_avail = mp->m_resblks_avail;
>> 463 }
>> 464
433 spin_unlock(&mp->m_sb_lock); 465 spin_unlock(&mp->m_sb_lock);
434 return error; 466 return error;
435 } 467 }
436 468
437 int 469 int
438 xfs_fs_goingdown( 470 xfs_fs_goingdown(
439 xfs_mount_t *mp, 471 xfs_mount_t *mp,
440 uint32_t inflags) 472 uint32_t inflags)
441 { 473 {
442 switch (inflags) { 474 switch (inflags) {
443 case XFS_FSOP_GOING_FLAGS_DEFAULT: { 475 case XFS_FSOP_GOING_FLAGS_DEFAULT: {
444 if (!bdev_freeze(mp->m_super-> !! 476 if (!freeze_bdev(mp->m_super->s_bdev)) {
445 xfs_force_shutdown(mp, 477 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
446 bdev_thaw(mp->m_super- !! 478 thaw_bdev(mp->m_super->s_bdev);
447 } 479 }
448 break; 480 break;
449 } 481 }
450 case XFS_FSOP_GOING_FLAGS_LOGFLUSH: 482 case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
451 xfs_force_shutdown(mp, SHUTDOW 483 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
452 break; 484 break;
453 case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH: 485 case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
454 xfs_force_shutdown(mp, 486 xfs_force_shutdown(mp,
455 SHUTDOWN_FORCE 487 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
456 break; 488 break;
457 default: 489 default:
458 return -EINVAL; 490 return -EINVAL;
459 } 491 }
460 492
461 return 0; 493 return 0;
462 } 494 }
463 495
464 /* 496 /*
465 * Force a shutdown of the filesystem instantl 497 * Force a shutdown of the filesystem instantly while keeping the filesystem
466 * consistent. We don't do an unmount here; ju 498 * consistent. We don't do an unmount here; just shutdown the shop, make sure
467 * that absolutely nothing persistent happens 499 * that absolutely nothing persistent happens to this filesystem after this
468 * point. 500 * point.
469 * 501 *
470 * The shutdown state change is atomic, result 502 * The shutdown state change is atomic, resulting in the first and only the
471 * first shutdown call processing the shutdown 503 * first shutdown call processing the shutdown. This means we only shutdown the
472 * log once as it requires, and we don't spam 504 * log once as it requires, and we don't spam the logs when multiple concurrent
473 * shutdowns race to set the shutdown flags. 505 * shutdowns race to set the shutdown flags.
474 */ 506 */
475 void 507 void
476 xfs_do_force_shutdown( 508 xfs_do_force_shutdown(
477 struct xfs_mount *mp, 509 struct xfs_mount *mp,
478 uint32_t flags, 510 uint32_t flags,
479 char *fname, 511 char *fname,
480 int lnnum) 512 int lnnum)
481 { 513 {
482 int tag; 514 int tag;
483 const char *why; 515 const char *why;
484 516
485 517
486 if (xfs_set_shutdown(mp)) { !! 518 if (test_and_set_bit(XFS_OPSTATE_SHUTDOWN, &mp->m_opstate)) {
487 xlog_shutdown_wait(mp->m_log); 519 xlog_shutdown_wait(mp->m_log);
488 return; 520 return;
489 } 521 }
490 if (mp->m_sb_bp) 522 if (mp->m_sb_bp)
491 mp->m_sb_bp->b_flags |= XBF_DO 523 mp->m_sb_bp->b_flags |= XBF_DONE;
492 524
493 if (flags & SHUTDOWN_FORCE_UMOUNT) 525 if (flags & SHUTDOWN_FORCE_UMOUNT)
494 xfs_alert(mp, "User initiated 526 xfs_alert(mp, "User initiated shutdown received.");
495 527
496 if (xlog_force_shutdown(mp->m_log, fla 528 if (xlog_force_shutdown(mp->m_log, flags)) {
497 tag = XFS_PTAG_SHUTDOWN_LOGERR 529 tag = XFS_PTAG_SHUTDOWN_LOGERROR;
498 why = "Log I/O Error"; 530 why = "Log I/O Error";
499 } else if (flags & SHUTDOWN_CORRUPT_IN 531 } else if (flags & SHUTDOWN_CORRUPT_INCORE) {
500 tag = XFS_PTAG_SHUTDOWN_CORRUP 532 tag = XFS_PTAG_SHUTDOWN_CORRUPT;
501 why = "Corruption of in-memory 533 why = "Corruption of in-memory data";
502 } else if (flags & SHUTDOWN_CORRUPT_ON 534 } else if (flags & SHUTDOWN_CORRUPT_ONDISK) {
503 tag = XFS_PTAG_SHUTDOWN_CORRUP 535 tag = XFS_PTAG_SHUTDOWN_CORRUPT;
504 why = "Corruption of on-disk m 536 why = "Corruption of on-disk metadata";
505 } else if (flags & SHUTDOWN_DEVICE_REM <<
506 tag = XFS_PTAG_SHUTDOWN_IOERRO <<
507 why = "Block device removal"; <<
508 } else { 537 } else {
509 tag = XFS_PTAG_SHUTDOWN_IOERRO 538 tag = XFS_PTAG_SHUTDOWN_IOERROR;
510 why = "Metadata I/O Error"; 539 why = "Metadata I/O Error";
511 } 540 }
512 541
513 trace_xfs_force_shutdown(mp, tag, flag 542 trace_xfs_force_shutdown(mp, tag, flags, fname, lnnum);
514 543
515 xfs_alert_tag(mp, tag, 544 xfs_alert_tag(mp, tag,
516 "%s (0x%x) detected at %pS (%s:%d). Shutting 545 "%s (0x%x) detected at %pS (%s:%d). Shutting down filesystem.",
517 why, flags, __return_a 546 why, flags, __return_address, fname, lnnum);
518 xfs_alert(mp, 547 xfs_alert(mp,
519 "Please unmount the filesystem 548 "Please unmount the filesystem and rectify the problem(s)");
520 if (xfs_error_level >= XFS_ERRLEVEL_HI 549 if (xfs_error_level >= XFS_ERRLEVEL_HIGH)
521 xfs_stack_trace(); 550 xfs_stack_trace();
522 } 551 }
523 552
524 /* 553 /*
525 * Reserve free space for per-AG metadata. 554 * Reserve free space for per-AG metadata.
526 */ 555 */
527 int 556 int
528 xfs_fs_reserve_ag_blocks( 557 xfs_fs_reserve_ag_blocks(
529 struct xfs_mount *mp) 558 struct xfs_mount *mp)
530 { 559 {
531 xfs_agnumber_t agno; 560 xfs_agnumber_t agno;
532 struct xfs_perag *pag; 561 struct xfs_perag *pag;
533 int error = 0; 562 int error = 0;
534 int err2; 563 int err2;
535 564
536 mp->m_finobt_nores = false; 565 mp->m_finobt_nores = false;
537 for_each_perag(mp, agno, pag) { 566 for_each_perag(mp, agno, pag) {
538 err2 = xfs_ag_resv_init(pag, N 567 err2 = xfs_ag_resv_init(pag, NULL);
539 if (err2 && !error) 568 if (err2 && !error)
540 error = err2; 569 error = err2;
541 } 570 }
542 571
543 if (error && error != -ENOSPC) { 572 if (error && error != -ENOSPC) {
544 xfs_warn(mp, 573 xfs_warn(mp,
545 "Error %d reserving per-AG metadata re 574 "Error %d reserving per-AG metadata reserve pool.", error);
546 xfs_force_shutdown(mp, SHUTDOW 575 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
547 } 576 }
548 577
549 return error; 578 return error;
550 } 579 }
551 580
552 /* 581 /*
553 * Free space reserved for per-AG metadata. 582 * Free space reserved for per-AG metadata.
554 */ 583 */
555 void !! 584 int
556 xfs_fs_unreserve_ag_blocks( 585 xfs_fs_unreserve_ag_blocks(
557 struct xfs_mount *mp) 586 struct xfs_mount *mp)
558 { 587 {
559 xfs_agnumber_t agno; 588 xfs_agnumber_t agno;
560 struct xfs_perag *pag; 589 struct xfs_perag *pag;
>> 590 int error = 0;
>> 591 int err2;
561 592
562 for_each_perag(mp, agno, pag) !! 593 for_each_perag(mp, agno, pag) {
563 xfs_ag_resv_free(pag); !! 594 err2 = xfs_ag_resv_free(pag);
>> 595 if (err2 && !error)
>> 596 error = err2;
>> 597 }
>> 598
>> 599 if (error)
>> 600 xfs_warn(mp,
>> 601 "Error %d freeing per-AG metadata reserve pool.", error);
>> 602
>> 603 return error;
564 } 604 }
565 605
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.