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