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