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 = false;
>> 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 if (nb_mod && nb_mod >= XFS_MIN_AG_BLOCKS)
119 nb_div++; 119 nb_div++;
120 else if (nb_mod) 120 else if (nb_mod)
121 nb = nb_div * mp->m_sb.sb_agbl 121 nb = nb_div * mp->m_sb.sb_agblocks;
122 122
123 if (nb_div > XFS_MAX_AGNUMBER + 1) { 123 if (nb_div > XFS_MAX_AGNUMBER + 1) {
124 nb_div = XFS_MAX_AGNUMBER + 1; 124 nb_div = XFS_MAX_AGNUMBER + 1;
125 nb = nb_div * mp->m_sb.sb_agbl 125 nb = nb_div * mp->m_sb.sb_agblocks;
126 } 126 }
127 nagcount = nb_div; 127 nagcount = nb_div;
128 delta = nb - mp->m_sb.sb_dblocks; 128 delta = nb - mp->m_sb.sb_dblocks;
129 /* 129 /*
130 * Reject filesystems with a single AG 130 * Reject filesystems with a single AG because they are not
131 * supported, and reject a shrink oper 131 * supported, and reject a shrink operation that would cause a
132 * filesystem to become unsupported. 132 * filesystem to become unsupported.
133 */ 133 */
134 if (delta < 0 && nagcount < 2) 134 if (delta < 0 && nagcount < 2)
135 return -EINVAL; 135 return -EINVAL;
136 136
137 /* No work to do */ 137 /* No work to do */
138 if (delta == 0) 138 if (delta == 0)
139 return 0; 139 return 0;
140 140
141 /* TODO: shrinking the entire AGs hasn !! 141 oagcount = mp->m_sb.sb_agcount;
142 if (nagcount < oagcount) <<
143 return -EINVAL; <<
144 <<
145 /* allocate the new per-ag structures 142 /* allocate the new per-ag structures */
146 error = xfs_initialize_perag(mp, oagco !! 143 if (nagcount > oagcount) {
147 if (error) !! 144 error = xfs_initialize_perag(mp, nagcount, nb, &nagimax);
148 return error; !! 145 if (error)
>> 146 return error;
>> 147 } else if (nagcount < oagcount) {
>> 148 /* TODO: shrinking the entire AGs hasn't yet completed */
>> 149 return -EINVAL;
>> 150 }
149 151
150 if (delta > 0) 152 if (delta > 0)
151 error = xfs_trans_alloc(mp, &M 153 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
152 XFS_GROWFS_SPA 154 XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
153 &tp); 155 &tp);
154 else 156 else
155 error = xfs_trans_alloc(mp, &M 157 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata, -delta, 0,
156 0, &tp); 158 0, &tp);
157 if (error) 159 if (error)
158 goto out_free_unused_perag; 160 goto out_free_unused_perag;
159 161
160 last_pag = xfs_perag_get(mp, oagcount 162 last_pag = xfs_perag_get(mp, oagcount - 1);
161 if (delta > 0) { 163 if (delta > 0) {
162 error = xfs_resizefs_init_new_ 164 error = xfs_resizefs_init_new_ags(tp, &id, oagcount, nagcount,
163 delta, last_pa 165 delta, last_pag, &lastag_extended);
164 } else { 166 } else {
165 xfs_warn_mount(mp, XFS_OPSTATE 167 xfs_warn_mount(mp, XFS_OPSTATE_WARNED_SHRINK,
166 "EXPERIMENTAL online shrink feature in 168 "EXPERIMENTAL online shrink feature in use. Use at your own risk!");
167 169
168 error = xfs_ag_shrink_space(la 170 error = xfs_ag_shrink_space(last_pag, &tp, -delta);
169 } 171 }
170 xfs_perag_put(last_pag); 172 xfs_perag_put(last_pag);
171 if (error) 173 if (error)
172 goto out_trans_cancel; 174 goto out_trans_cancel;
173 175
174 /* 176 /*
175 * Update changed superblock fields tr 177 * Update changed superblock fields transactionally. These are not
176 * seen by the rest of the world until 178 * seen by the rest of the world until the transaction commit applies
177 * them atomically to the superblock. 179 * them atomically to the superblock.
178 */ 180 */
179 if (nagcount > oagcount) 181 if (nagcount > oagcount)
180 xfs_trans_mod_sb(tp, XFS_TRANS 182 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
181 if (delta) 183 if (delta)
182 xfs_trans_mod_sb(tp, XFS_TRANS 184 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS, delta);
183 if (id.nfree) 185 if (id.nfree)
184 xfs_trans_mod_sb(tp, XFS_TRANS 186 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, id.nfree);
185 187
186 /* 188 /*
187 * Sync sb counters now to reflect the 189 * Sync sb counters now to reflect the updated values. This is
188 * particularly important for shrink b 190 * particularly important for shrink because the write verifier
189 * will fail if sb_fdblocks is ever la 191 * will fail if sb_fdblocks is ever larger than sb_dblocks.
190 */ 192 */
191 if (xfs_has_lazysbcount(mp)) 193 if (xfs_has_lazysbcount(mp))
192 xfs_log_sb(tp); 194 xfs_log_sb(tp);
193 195
194 xfs_trans_set_sync(tp); 196 xfs_trans_set_sync(tp);
195 error = xfs_trans_commit(tp); 197 error = xfs_trans_commit(tp);
196 if (error) 198 if (error)
197 return error; 199 return error;
198 200
199 /* New allocation groups fully initial 201 /* New allocation groups fully initialized, so update mount struct */
200 if (nagimax) 202 if (nagimax)
201 mp->m_maxagi = nagimax; 203 mp->m_maxagi = nagimax;
202 xfs_set_low_space_thresholds(mp); 204 xfs_set_low_space_thresholds(mp);
203 mp->m_alloc_set_aside = xfs_alloc_set_ 205 mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
204 206
205 if (delta > 0) { 207 if (delta > 0) {
206 /* 208 /*
207 * If we expanded the last AG, 209 * If we expanded the last AG, free the per-AG reservation
208 * so we can reinitialize it w 210 * so we can reinitialize it with the new size.
209 */ 211 */
210 if (lastag_extended) { 212 if (lastag_extended) {
211 struct xfs_perag 213 struct xfs_perag *pag;
212 214
213 pag = xfs_perag_get(mp 215 pag = xfs_perag_get(mp, id.agno);
214 xfs_ag_resv_free(pag); !! 216 error = xfs_ag_resv_free(pag);
215 xfs_perag_put(pag); 217 xfs_perag_put(pag);
>> 218 if (error)
>> 219 return error;
216 } 220 }
217 /* 221 /*
218 * Reserve AG metadata blocks. 222 * Reserve AG metadata blocks. ENOSPC here does not mean there
219 * was a growfs failure, just 223 * was a growfs failure, just that there still isn't space for
220 * new user data after the gro 224 * new user data after the grow has been run.
221 */ 225 */
222 error = xfs_fs_reserve_ag_bloc 226 error = xfs_fs_reserve_ag_blocks(mp);
223 if (error == -ENOSPC) 227 if (error == -ENOSPC)
224 error = 0; 228 error = 0;
225 } 229 }
226 return error; 230 return error;
227 231
228 out_trans_cancel: 232 out_trans_cancel:
229 xfs_trans_cancel(tp); 233 xfs_trans_cancel(tp);
230 out_free_unused_perag: 234 out_free_unused_perag:
231 if (nagcount > oagcount) 235 if (nagcount > oagcount)
232 xfs_free_perag_range(mp, oagco !! 236 xfs_free_unused_perag_range(mp, oagcount, nagcount);
233 return error; 237 return error;
234 } 238 }
235 239
236 static int 240 static int
237 xfs_growfs_log_private( 241 xfs_growfs_log_private(
238 struct xfs_mount *mp, /* mou 242 struct xfs_mount *mp, /* mount point for filesystem */
239 struct xfs_growfs_log *in) /* gro 243 struct xfs_growfs_log *in) /* growfs log input struct */
240 { 244 {
241 xfs_extlen_t nb; 245 xfs_extlen_t nb;
242 246
243 nb = in->newblocks; 247 nb = in->newblocks;
244 if (nb < XFS_MIN_LOG_BLOCKS || nb < XF 248 if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
245 return -EINVAL; 249 return -EINVAL;
246 if (nb == mp->m_sb.sb_logblocks && 250 if (nb == mp->m_sb.sb_logblocks &&
247 in->isint == (mp->m_sb.sb_logstart 251 in->isint == (mp->m_sb.sb_logstart != 0))
248 return -EINVAL; 252 return -EINVAL;
249 /* 253 /*
250 * Moving the log is hard, need new in 254 * Moving the log is hard, need new interfaces to sync
251 * the log first, hold off all activit 255 * the log first, hold off all activity while moving it.
252 * Can have shorter or longer log in t 256 * Can have shorter or longer log in the same space,
253 * or transform internal to external l 257 * or transform internal to external log or vice versa.
254 */ 258 */
255 return -ENOSYS; 259 return -ENOSYS;
256 } 260 }
257 261
258 static int 262 static int
259 xfs_growfs_imaxpct( 263 xfs_growfs_imaxpct(
260 struct xfs_mount *mp, 264 struct xfs_mount *mp,
261 __u32 imaxpct) 265 __u32 imaxpct)
262 { 266 {
263 struct xfs_trans *tp; 267 struct xfs_trans *tp;
264 int dpct; 268 int dpct;
265 int error; 269 int error;
266 270
267 if (imaxpct > 100) 271 if (imaxpct > 100)
268 return -EINVAL; 272 return -EINVAL;
269 273
270 error = xfs_trans_alloc(mp, &M_RES(mp) 274 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
271 XFS_GROWFS_SPACE_RES(m 275 XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
272 if (error) 276 if (error)
273 return error; 277 return error;
274 278
275 dpct = imaxpct - mp->m_sb.sb_imax_pct; 279 dpct = imaxpct - mp->m_sb.sb_imax_pct;
276 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAX 280 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
277 xfs_trans_set_sync(tp); 281 xfs_trans_set_sync(tp);
278 return xfs_trans_commit(tp); 282 return xfs_trans_commit(tp);
279 } 283 }
280 284
281 /* 285 /*
282 * protected versions of growfs function acqui 286 * protected versions of growfs function acquire and release locks on the mount
283 * point - exported through ioctls: XFS_IOC_FS 287 * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
284 * XFS_IOC_FSGROWFSRT 288 * XFS_IOC_FSGROWFSRT
285 */ 289 */
286 int 290 int
287 xfs_growfs_data( 291 xfs_growfs_data(
288 struct xfs_mount *mp, 292 struct xfs_mount *mp,
289 struct xfs_growfs_data *in) 293 struct xfs_growfs_data *in)
290 { 294 {
291 int error = 0; 295 int error = 0;
292 296
293 if (!capable(CAP_SYS_ADMIN)) 297 if (!capable(CAP_SYS_ADMIN))
294 return -EPERM; 298 return -EPERM;
295 if (!mutex_trylock(&mp->m_growlock)) 299 if (!mutex_trylock(&mp->m_growlock))
296 return -EWOULDBLOCK; 300 return -EWOULDBLOCK;
297 301
298 /* update imaxpct separately to the ph 302 /* update imaxpct separately to the physical grow of the filesystem */
299 if (in->imaxpct != mp->m_sb.sb_imax_pc 303 if (in->imaxpct != mp->m_sb.sb_imax_pct) {
300 error = xfs_growfs_imaxpct(mp, 304 error = xfs_growfs_imaxpct(mp, in->imaxpct);
301 if (error) 305 if (error)
302 goto out_error; 306 goto out_error;
303 } 307 }
304 308
305 if (in->newblocks != mp->m_sb.sb_dbloc 309 if (in->newblocks != mp->m_sb.sb_dblocks) {
306 error = xfs_growfs_data_privat 310 error = xfs_growfs_data_private(mp, in);
307 if (error) 311 if (error)
308 goto out_error; 312 goto out_error;
309 } 313 }
310 314
311 /* Post growfs calculations needed to 315 /* Post growfs calculations needed to reflect new state in operations */
312 if (mp->m_sb.sb_imax_pct) { 316 if (mp->m_sb.sb_imax_pct) {
313 uint64_t icount = mp->m_sb.sb_ 317 uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
314 do_div(icount, 100); 318 do_div(icount, 100);
315 M_IGEO(mp)->maxicount = XFS_FS 319 M_IGEO(mp)->maxicount = XFS_FSB_TO_INO(mp, icount);
316 } else 320 } else
317 M_IGEO(mp)->maxicount = 0; 321 M_IGEO(mp)->maxicount = 0;
318 322
319 /* Update secondary superblocks now th 323 /* Update secondary superblocks now the physical grow has completed */
320 error = xfs_update_secondary_sbs(mp); 324 error = xfs_update_secondary_sbs(mp);
321 325
322 out_error: 326 out_error:
323 /* 327 /*
324 * Increment the generation unconditio 328 * Increment the generation unconditionally, the error could be from
325 * updating the secondary superblocks, 329 * updating the secondary superblocks, in which case the new size
326 * is live already. 330 * is live already.
327 */ 331 */
328 mp->m_generation++; 332 mp->m_generation++;
329 mutex_unlock(&mp->m_growlock); 333 mutex_unlock(&mp->m_growlock);
330 return error; 334 return error;
331 } 335 }
332 336
333 int 337 int
334 xfs_growfs_log( 338 xfs_growfs_log(
335 xfs_mount_t *mp, 339 xfs_mount_t *mp,
336 struct xfs_growfs_log *in) 340 struct xfs_growfs_log *in)
337 { 341 {
338 int error; 342 int error;
339 343
340 if (!capable(CAP_SYS_ADMIN)) 344 if (!capable(CAP_SYS_ADMIN))
341 return -EPERM; 345 return -EPERM;
342 if (!mutex_trylock(&mp->m_growlock)) 346 if (!mutex_trylock(&mp->m_growlock))
343 return -EWOULDBLOCK; 347 return -EWOULDBLOCK;
344 error = xfs_growfs_log_private(mp, in) 348 error = xfs_growfs_log_private(mp, in);
345 mutex_unlock(&mp->m_growlock); 349 mutex_unlock(&mp->m_growlock);
346 return error; 350 return error;
347 } 351 }
348 352
349 /* 353 /*
350 * Reserve the requested number of blocks if a 354 * Reserve the requested number of blocks if available. Otherwise return
351 * as many as possible to satisfy the request. 355 * as many as possible to satisfy the request. The actual number
352 * reserved are returned in outval. 356 * reserved are returned in outval.
353 */ 357 */
354 int 358 int
355 xfs_reserve_blocks( 359 xfs_reserve_blocks(
356 struct xfs_mount *mp, 360 struct xfs_mount *mp,
357 uint64_t request) 361 uint64_t request)
358 { 362 {
359 int64_t lcounter, delt 363 int64_t lcounter, delta;
360 int64_t fdblks_delta = 364 int64_t fdblks_delta = 0;
361 int64_t free; 365 int64_t free;
362 int error = 0; 366 int error = 0;
363 367
364 /* 368 /*
365 * With per-cpu counters, this becomes 369 * With per-cpu counters, this becomes an interesting problem. we need
366 * to work out if we are freeing or al 370 * to work out if we are freeing or allocation blocks first, then we can
367 * do the modification as necessary. 371 * do the modification as necessary.
368 * 372 *
369 * We do this under the m_sb_lock so t 373 * 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 374 * hold out any changes while we work out what to do. This means that
371 * the amount of free space can change 375 * the amount of free space can change while we do this, so we need to
372 * retry if we end up trying to reserv 376 * retry if we end up trying to reserve more space than is available.
373 */ 377 */
374 spin_lock(&mp->m_sb_lock); 378 spin_lock(&mp->m_sb_lock);
375 379
376 /* 380 /*
377 * If our previous reservation was lar 381 * If our previous reservation was larger than the current value,
378 * then move any unused blocks back to 382 * then move any unused blocks back to the free pool. Modify the resblks
379 * counters directly since we shouldn' 383 * counters directly since we shouldn't have any problems unreserving
380 * space. 384 * space.
381 */ 385 */
382 if (mp->m_resblks > request) { 386 if (mp->m_resblks > request) {
383 lcounter = mp->m_resblks_avail 387 lcounter = mp->m_resblks_avail - request;
384 if (lcounter > 0) { !! 388 if (lcounter > 0) { /* release unused blocks */
385 fdblks_delta = lcounte 389 fdblks_delta = lcounter;
386 mp->m_resblks_avail -= 390 mp->m_resblks_avail -= lcounter;
387 } 391 }
388 mp->m_resblks = request; 392 mp->m_resblks = request;
389 if (fdblks_delta) { 393 if (fdblks_delta) {
390 spin_unlock(&mp->m_sb_ 394 spin_unlock(&mp->m_sb_lock);
391 xfs_add_fdblocks(mp, f !! 395 error = xfs_mod_fdblocks(mp, fdblks_delta, 0);
392 spin_lock(&mp->m_sb_lo 396 spin_lock(&mp->m_sb_lock);
393 } 397 }
394 398
395 goto out; 399 goto out;
396 } 400 }
397 401
398 /* 402 /*
399 * If the request is larger than the c 403 * If the request is larger than the current reservation, reserve the
400 * blocks before we update the reserve 404 * blocks before we update the reserve counters. Sample m_fdblocks and
401 * perform a partial reservation if th 405 * perform a partial reservation if the request exceeds free space.
402 * 406 *
403 * The code below estimates how many b 407 * The code below estimates how many blocks it can request from
404 * fdblocks to stash in the reserve po 408 * fdblocks to stash in the reserve pool. This is a classic TOCTOU
405 * race since fdblocks updates are not 409 * race since fdblocks updates are not always coordinated via
406 * m_sb_lock. Set the reserve size ev 410 * m_sb_lock. Set the reserve size even if there's not enough free
407 * space to fill it because mod_fdbloc 411 * space to fill it because mod_fdblocks will refill an undersized
408 * reserve when it can. 412 * reserve when it can.
409 */ 413 */
410 free = percpu_counter_sum(&mp->m_fdblo 414 free = percpu_counter_sum(&mp->m_fdblocks) -
411 415 xfs_fdblocks_unavailable(mp);
412 delta = request - mp->m_resblks; 416 delta = request - mp->m_resblks;
413 mp->m_resblks = request; 417 mp->m_resblks = request;
414 if (delta > 0 && free > 0) { 418 if (delta > 0 && free > 0) {
415 /* 419 /*
416 * We'll either succeed in get 420 * We'll either succeed in getting space from the free block
417 * count or we'll get an ENOSP 421 * count or we'll get an ENOSPC. Don't set the reserved flag
418 * here - we don't want to res 422 * here - we don't want to reserve the extra reserve blocks
419 * from the reserve. 423 * from the reserve.
420 * 424 *
421 * The desired reserve size ca 425 * The desired reserve size can change after we drop the lock.
422 * Use mod_fdblocks to put the 426 * Use mod_fdblocks to put the space into the reserve or into
423 * fdblocks as appropriate. 427 * fdblocks as appropriate.
424 */ 428 */
425 fdblks_delta = min(free, delta 429 fdblks_delta = min(free, delta);
426 spin_unlock(&mp->m_sb_lock); 430 spin_unlock(&mp->m_sb_lock);
427 error = xfs_dec_fdblocks(mp, f !! 431 error = xfs_mod_fdblocks(mp, -fdblks_delta, 0);
428 if (!error) 432 if (!error)
429 xfs_add_fdblocks(mp, f !! 433 xfs_mod_fdblocks(mp, fdblks_delta, 0);
430 spin_lock(&mp->m_sb_lock); 434 spin_lock(&mp->m_sb_lock);
431 } 435 }
432 out: 436 out:
433 spin_unlock(&mp->m_sb_lock); 437 spin_unlock(&mp->m_sb_lock);
434 return error; 438 return error;
435 } 439 }
436 440
437 int 441 int
438 xfs_fs_goingdown( 442 xfs_fs_goingdown(
439 xfs_mount_t *mp, 443 xfs_mount_t *mp,
440 uint32_t inflags) 444 uint32_t inflags)
441 { 445 {
442 switch (inflags) { 446 switch (inflags) {
443 case XFS_FSOP_GOING_FLAGS_DEFAULT: { 447 case XFS_FSOP_GOING_FLAGS_DEFAULT: {
444 if (!bdev_freeze(mp->m_super-> 448 if (!bdev_freeze(mp->m_super->s_bdev)) {
445 xfs_force_shutdown(mp, 449 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
446 bdev_thaw(mp->m_super- 450 bdev_thaw(mp->m_super->s_bdev);
447 } 451 }
448 break; 452 break;
449 } 453 }
450 case XFS_FSOP_GOING_FLAGS_LOGFLUSH: 454 case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
451 xfs_force_shutdown(mp, SHUTDOW 455 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
452 break; 456 break;
453 case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH: 457 case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
454 xfs_force_shutdown(mp, 458 xfs_force_shutdown(mp,
455 SHUTDOWN_FORCE 459 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
456 break; 460 break;
457 default: 461 default:
458 return -EINVAL; 462 return -EINVAL;
459 } 463 }
460 464
461 return 0; 465 return 0;
462 } 466 }
463 467
464 /* 468 /*
465 * Force a shutdown of the filesystem instantl 469 * Force a shutdown of the filesystem instantly while keeping the filesystem
466 * consistent. We don't do an unmount here; ju 470 * consistent. We don't do an unmount here; just shutdown the shop, make sure
467 * that absolutely nothing persistent happens 471 * that absolutely nothing persistent happens to this filesystem after this
468 * point. 472 * point.
469 * 473 *
470 * The shutdown state change is atomic, result 474 * The shutdown state change is atomic, resulting in the first and only the
471 * first shutdown call processing the shutdown 475 * first shutdown call processing the shutdown. This means we only shutdown the
472 * log once as it requires, and we don't spam 476 * log once as it requires, and we don't spam the logs when multiple concurrent
473 * shutdowns race to set the shutdown flags. 477 * shutdowns race to set the shutdown flags.
474 */ 478 */
475 void 479 void
476 xfs_do_force_shutdown( 480 xfs_do_force_shutdown(
477 struct xfs_mount *mp, 481 struct xfs_mount *mp,
478 uint32_t flags, 482 uint32_t flags,
479 char *fname, 483 char *fname,
480 int lnnum) 484 int lnnum)
481 { 485 {
482 int tag; 486 int tag;
483 const char *why; 487 const char *why;
484 488
485 489
486 if (xfs_set_shutdown(mp)) { !! 490 if (test_and_set_bit(XFS_OPSTATE_SHUTDOWN, &mp->m_opstate)) {
487 xlog_shutdown_wait(mp->m_log); 491 xlog_shutdown_wait(mp->m_log);
488 return; 492 return;
489 } 493 }
490 if (mp->m_sb_bp) 494 if (mp->m_sb_bp)
491 mp->m_sb_bp->b_flags |= XBF_DO 495 mp->m_sb_bp->b_flags |= XBF_DONE;
492 496
493 if (flags & SHUTDOWN_FORCE_UMOUNT) 497 if (flags & SHUTDOWN_FORCE_UMOUNT)
494 xfs_alert(mp, "User initiated 498 xfs_alert(mp, "User initiated shutdown received.");
495 499
496 if (xlog_force_shutdown(mp->m_log, fla 500 if (xlog_force_shutdown(mp->m_log, flags)) {
497 tag = XFS_PTAG_SHUTDOWN_LOGERR 501 tag = XFS_PTAG_SHUTDOWN_LOGERROR;
498 why = "Log I/O Error"; 502 why = "Log I/O Error";
499 } else if (flags & SHUTDOWN_CORRUPT_IN 503 } else if (flags & SHUTDOWN_CORRUPT_INCORE) {
500 tag = XFS_PTAG_SHUTDOWN_CORRUP 504 tag = XFS_PTAG_SHUTDOWN_CORRUPT;
501 why = "Corruption of in-memory 505 why = "Corruption of in-memory data";
502 } else if (flags & SHUTDOWN_CORRUPT_ON 506 } else if (flags & SHUTDOWN_CORRUPT_ONDISK) {
503 tag = XFS_PTAG_SHUTDOWN_CORRUP 507 tag = XFS_PTAG_SHUTDOWN_CORRUPT;
504 why = "Corruption of on-disk m 508 why = "Corruption of on-disk metadata";
505 } else if (flags & SHUTDOWN_DEVICE_REM 509 } else if (flags & SHUTDOWN_DEVICE_REMOVED) {
506 tag = XFS_PTAG_SHUTDOWN_IOERRO 510 tag = XFS_PTAG_SHUTDOWN_IOERROR;
507 why = "Block device removal"; 511 why = "Block device removal";
508 } else { 512 } else {
509 tag = XFS_PTAG_SHUTDOWN_IOERRO 513 tag = XFS_PTAG_SHUTDOWN_IOERROR;
510 why = "Metadata I/O Error"; 514 why = "Metadata I/O Error";
511 } 515 }
512 516
513 trace_xfs_force_shutdown(mp, tag, flag 517 trace_xfs_force_shutdown(mp, tag, flags, fname, lnnum);
514 518
515 xfs_alert_tag(mp, tag, 519 xfs_alert_tag(mp, tag,
516 "%s (0x%x) detected at %pS (%s:%d). Shutting 520 "%s (0x%x) detected at %pS (%s:%d). Shutting down filesystem.",
517 why, flags, __return_a 521 why, flags, __return_address, fname, lnnum);
518 xfs_alert(mp, 522 xfs_alert(mp,
519 "Please unmount the filesystem 523 "Please unmount the filesystem and rectify the problem(s)");
520 if (xfs_error_level >= XFS_ERRLEVEL_HI 524 if (xfs_error_level >= XFS_ERRLEVEL_HIGH)
521 xfs_stack_trace(); 525 xfs_stack_trace();
522 } 526 }
523 527
524 /* 528 /*
525 * Reserve free space for per-AG metadata. 529 * Reserve free space for per-AG metadata.
526 */ 530 */
527 int 531 int
528 xfs_fs_reserve_ag_blocks( 532 xfs_fs_reserve_ag_blocks(
529 struct xfs_mount *mp) 533 struct xfs_mount *mp)
530 { 534 {
531 xfs_agnumber_t agno; 535 xfs_agnumber_t agno;
532 struct xfs_perag *pag; 536 struct xfs_perag *pag;
533 int error = 0; 537 int error = 0;
534 int err2; 538 int err2;
535 539
536 mp->m_finobt_nores = false; 540 mp->m_finobt_nores = false;
537 for_each_perag(mp, agno, pag) { 541 for_each_perag(mp, agno, pag) {
538 err2 = xfs_ag_resv_init(pag, N 542 err2 = xfs_ag_resv_init(pag, NULL);
539 if (err2 && !error) 543 if (err2 && !error)
540 error = err2; 544 error = err2;
541 } 545 }
542 546
543 if (error && error != -ENOSPC) { 547 if (error && error != -ENOSPC) {
544 xfs_warn(mp, 548 xfs_warn(mp,
545 "Error %d reserving per-AG metadata re 549 "Error %d reserving per-AG metadata reserve pool.", error);
546 xfs_force_shutdown(mp, SHUTDOW 550 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
547 } 551 }
548 552
549 return error; 553 return error;
550 } 554 }
551 555
552 /* 556 /*
553 * Free space reserved for per-AG metadata. 557 * Free space reserved for per-AG metadata.
554 */ 558 */
555 void !! 559 int
556 xfs_fs_unreserve_ag_blocks( 560 xfs_fs_unreserve_ag_blocks(
557 struct xfs_mount *mp) 561 struct xfs_mount *mp)
558 { 562 {
559 xfs_agnumber_t agno; 563 xfs_agnumber_t agno;
560 struct xfs_perag *pag; 564 struct xfs_perag *pag;
>> 565 int error = 0;
>> 566 int err2;
>> 567
>> 568 for_each_perag(mp, agno, pag) {
>> 569 err2 = xfs_ag_resv_free(pag);
>> 570 if (err2 && !error)
>> 571 error = err2;
>> 572 }
561 573
562 for_each_perag(mp, agno, pag) !! 574 if (error)
563 xfs_ag_resv_free(pag); !! 575 xfs_warn(mp,
>> 576 "Error %d freeing per-AG metadata reserve pool.", error);
>> 577
>> 578 return error;
564 } 579 }
565 580
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.