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