]>
Commit | Line | Data |
---|---|---|
7336d0e6 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
b3b94faa DT |
2 | /* |
3 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
da6dd40d | 4 | * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
5 | */ |
6 | ||
d77d1b58 JP |
7 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
8 | ||
9e6e0a12 | 9 | #include <linux/bio.h> |
174cd4b1 | 10 | #include <linux/sched/signal.h> |
b3b94faa DT |
11 | #include <linux/slab.h> |
12 | #include <linux/spinlock.h> | |
13 | #include <linux/completion.h> | |
14 | #include <linux/buffer_head.h> | |
9e6e0a12 SW |
15 | #include <linux/statfs.h> |
16 | #include <linux/seq_file.h> | |
17 | #include <linux/mount.h> | |
18 | #include <linux/kthread.h> | |
19 | #include <linux/delay.h> | |
5c676f6d | 20 | #include <linux/gfs2_ondisk.h> |
9e6e0a12 SW |
21 | #include <linux/crc32.h> |
22 | #include <linux/time.h> | |
e402746a | 23 | #include <linux/wait.h> |
a9185b41 | 24 | #include <linux/writeback.h> |
4667a0ec | 25 | #include <linux/backing-dev.h> |
2e60d768 | 26 | #include <linux/kernel.h> |
b3b94faa DT |
27 | |
28 | #include "gfs2.h" | |
5c676f6d | 29 | #include "incore.h" |
b3b94faa DT |
30 | #include "bmap.h" |
31 | #include "dir.h" | |
b3b94faa DT |
32 | #include "glock.h" |
33 | #include "glops.h" | |
34 | #include "inode.h" | |
35 | #include "log.h" | |
36 | #include "meta_io.h" | |
37 | #include "quota.h" | |
38 | #include "recovery.h" | |
39 | #include "rgrp.h" | |
40 | #include "super.h" | |
41 | #include "trans.h" | |
5c676f6d | 42 | #include "util.h" |
9e6e0a12 | 43 | #include "sys.h" |
307cf6e6 | 44 | #include "xattr.h" |
f4686c26 | 45 | #include "lops.h" |
9e6e0a12 | 46 | |
53dbc27e BP |
47 | enum dinode_demise { |
48 | SHOULD_DELETE_DINODE, | |
49 | SHOULD_NOT_DELETE_DINODE, | |
50 | SHOULD_DEFER_EVICTION, | |
51 | }; | |
52 | ||
fefc03bf SW |
53 | /** |
54 | * gfs2_jindex_free - Clear all the journal index information | |
55 | * @sdp: The GFS2 superblock | |
56 | * | |
57 | */ | |
58 | ||
59 | void gfs2_jindex_free(struct gfs2_sbd *sdp) | |
60 | { | |
b50f227b | 61 | struct list_head list; |
fefc03bf | 62 | struct gfs2_jdesc *jd; |
fefc03bf SW |
63 | |
64 | spin_lock(&sdp->sd_jindex_spin); | |
65 | list_add(&list, &sdp->sd_jindex_list); | |
66 | list_del_init(&sdp->sd_jindex_list); | |
67 | sdp->sd_journals = 0; | |
68 | spin_unlock(&sdp->sd_jindex_spin); | |
69 | ||
601ef0d5 | 70 | sdp->sd_jdesc = NULL; |
fefc03bf | 71 | while (!list_empty(&list)) { |
969183bc | 72 | jd = list_first_entry(&list, struct gfs2_jdesc, jd_list); |
b50f227b | 73 | gfs2_free_journal_extents(jd); |
fefc03bf SW |
74 | list_del(&jd->jd_list); |
75 | iput(jd->jd_inode); | |
601ef0d5 | 76 | jd->jd_inode = NULL; |
fefc03bf SW |
77 | kfree(jd); |
78 | } | |
79 | } | |
80 | ||
b3b94faa DT |
81 | static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid) |
82 | { | |
83 | struct gfs2_jdesc *jd; | |
b3b94faa DT |
84 | |
85 | list_for_each_entry(jd, head, jd_list) { | |
736b2f77 AG |
86 | if (jd->jd_jid == jid) |
87 | return jd; | |
b3b94faa | 88 | } |
736b2f77 | 89 | return NULL; |
b3b94faa DT |
90 | } |
91 | ||
92 | struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid) | |
93 | { | |
94 | struct gfs2_jdesc *jd; | |
95 | ||
96 | spin_lock(&sdp->sd_jindex_spin); | |
97 | jd = jdesc_find_i(&sdp->sd_jindex_list, jid); | |
98 | spin_unlock(&sdp->sd_jindex_spin); | |
99 | ||
100 | return jd; | |
101 | } | |
102 | ||
b3b94faa DT |
103 | int gfs2_jdesc_check(struct gfs2_jdesc *jd) |
104 | { | |
feaa7bba SW |
105 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); |
106 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); | |
a2e0f799 | 107 | u64 size = i_size_read(jd->jd_inode); |
b3b94faa | 108 | |
47a9a527 | 109 | if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30))) |
b3b94faa | 110 | return -EIO; |
b3b94faa | 111 | |
a2e0f799 SW |
112 | jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift; |
113 | ||
114 | if (gfs2_write_alloc_required(ip, 0, size)) { | |
b3b94faa | 115 | gfs2_consist_inode(ip); |
461cb419 | 116 | return -EIO; |
b3b94faa DT |
117 | } |
118 | ||
461cb419 | 119 | return 0; |
b3b94faa DT |
120 | } |
121 | ||
b3b94faa DT |
122 | /** |
123 | * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one | |
124 | * @sdp: the filesystem | |
125 | * | |
126 | * Returns: errno | |
127 | */ | |
128 | ||
129 | int gfs2_make_fs_rw(struct gfs2_sbd *sdp) | |
130 | { | |
feaa7bba | 131 | struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode); |
5c676f6d | 132 | struct gfs2_glock *j_gl = ip->i_gl; |
55167622 | 133 | struct gfs2_log_header_host head; |
b3b94faa DT |
134 | int error; |
135 | ||
1a14d3a6 | 136 | j_gl->gl_ops->go_inval(j_gl, DIO_METADATA); |
a28dc123 BP |
137 | if (gfs2_withdrawn(sdp)) |
138 | return -EIO; | |
b3b94faa | 139 | |
f4686c26 | 140 | error = gfs2_find_jhead(sdp->sd_jdesc, &head, false); |
b66f723b AG |
141 | if (error) { |
142 | gfs2_consist(sdp); | |
a28dc123 | 143 | return error; |
b66f723b | 144 | } |
b3b94faa DT |
145 | |
146 | if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) { | |
147 | gfs2_consist(sdp); | |
a28dc123 | 148 | return -EIO; |
b3b94faa DT |
149 | } |
150 | ||
151 | /* Initialize some head of the log stuff */ | |
152 | sdp->sd_log_sequence = head.lh_sequence + 1; | |
153 | gfs2_log_pointers_init(sdp, head.lh_blkno); | |
154 | ||
b3b94faa | 155 | error = gfs2_quota_init(sdp); |
b66f723b AG |
156 | if (!error && gfs2_withdrawn(sdp)) |
157 | error = -EIO; | |
158 | if (!error) | |
a28dc123 | 159 | set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags); |
b3b94faa DT |
160 | return error; |
161 | } | |
162 | ||
1946f70a | 163 | void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf) |
bb8d8a6f SW |
164 | { |
165 | const struct gfs2_statfs_change *str = buf; | |
166 | ||
167 | sc->sc_total = be64_to_cpu(str->sc_total); | |
168 | sc->sc_free = be64_to_cpu(str->sc_free); | |
169 | sc->sc_dinodes = be64_to_cpu(str->sc_dinodes); | |
170 | } | |
171 | ||
73092698 | 172 | void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf) |
bb8d8a6f SW |
173 | { |
174 | struct gfs2_statfs_change *str = buf; | |
175 | ||
176 | str->sc_total = cpu_to_be64(sc->sc_total); | |
177 | str->sc_free = cpu_to_be64(sc->sc_free); | |
178 | str->sc_dinodes = cpu_to_be64(sc->sc_dinodes); | |
179 | } | |
180 | ||
b3b94faa DT |
181 | int gfs2_statfs_init(struct gfs2_sbd *sdp) |
182 | { | |
feaa7bba | 183 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); |
bd209cc0 | 184 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; |
bd209cc0 | 185 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; |
70c11ba8 | 186 | struct buffer_head *m_bh; |
b3b94faa DT |
187 | struct gfs2_holder gh; |
188 | int error; | |
189 | ||
190 | error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE, | |
191 | &gh); | |
192 | if (error) | |
193 | return error; | |
194 | ||
195 | error = gfs2_meta_inode_buffer(m_ip, &m_bh); | |
196 | if (error) | |
197 | goto out; | |
198 | ||
199 | if (sdp->sd_args.ar_spectator) { | |
200 | spin_lock(&sdp->sd_statfs_spin); | |
201 | gfs2_statfs_change_in(m_sc, m_bh->b_data + | |
202 | sizeof(struct gfs2_dinode)); | |
203 | spin_unlock(&sdp->sd_statfs_spin); | |
204 | } else { | |
b3b94faa DT |
205 | spin_lock(&sdp->sd_statfs_spin); |
206 | gfs2_statfs_change_in(m_sc, m_bh->b_data + | |
207 | sizeof(struct gfs2_dinode)); | |
70c11ba8 | 208 | gfs2_statfs_change_in(l_sc, sdp->sd_sc_bh->b_data + |
b3b94faa DT |
209 | sizeof(struct gfs2_dinode)); |
210 | spin_unlock(&sdp->sd_statfs_spin); | |
211 | ||
b3b94faa DT |
212 | } |
213 | ||
b3b94faa | 214 | brelse(m_bh); |
a91ea69f | 215 | out: |
b3b94faa | 216 | gfs2_glock_dq_uninit(&gh); |
b3b94faa DT |
217 | return 0; |
218 | } | |
219 | ||
cd915493 SW |
220 | void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free, |
221 | s64 dinodes) | |
b3b94faa | 222 | { |
feaa7bba | 223 | struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode); |
bd209cc0 | 224 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; |
3d3c10f2 | 225 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; |
c14f5735 BM |
226 | s64 x, y; |
227 | int need_sync = 0; | |
b3b94faa | 228 | |
70c11ba8 | 229 | gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh); |
b3b94faa DT |
230 | |
231 | spin_lock(&sdp->sd_statfs_spin); | |
232 | l_sc->sc_total += total; | |
233 | l_sc->sc_free += free; | |
234 | l_sc->sc_dinodes += dinodes; | |
70c11ba8 BP |
235 | gfs2_statfs_change_out(l_sc, sdp->sd_sc_bh->b_data + |
236 | sizeof(struct gfs2_dinode)); | |
c14f5735 BM |
237 | if (sdp->sd_args.ar_statfs_percent) { |
238 | x = 100 * l_sc->sc_free; | |
239 | y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent; | |
240 | if (x >= y || x <= -y) | |
241 | need_sync = 1; | |
242 | } | |
b3b94faa DT |
243 | spin_unlock(&sdp->sd_statfs_spin); |
244 | ||
c14f5735 | 245 | if (need_sync) |
3d3c10f2 | 246 | gfs2_wake_up_statfs(sdp); |
b3b94faa DT |
247 | } |
248 | ||
70c11ba8 | 249 | void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh) |
1946f70a BM |
250 | { |
251 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); | |
252 | struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode); | |
253 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; | |
254 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; | |
255 | ||
70c11ba8 | 256 | gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh); |
901c6c66 | 257 | gfs2_trans_add_meta(m_ip->i_gl, m_bh); |
1946f70a BM |
258 | |
259 | spin_lock(&sdp->sd_statfs_spin); | |
260 | m_sc->sc_total += l_sc->sc_total; | |
261 | m_sc->sc_free += l_sc->sc_free; | |
262 | m_sc->sc_dinodes += l_sc->sc_dinodes; | |
263 | memset(l_sc, 0, sizeof(struct gfs2_statfs_change)); | |
70c11ba8 | 264 | memset(sdp->sd_sc_bh->b_data + sizeof(struct gfs2_dinode), |
1946f70a | 265 | 0, sizeof(struct gfs2_statfs_change)); |
1946f70a | 266 | gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode)); |
901c6c66 | 267 | spin_unlock(&sdp->sd_statfs_spin); |
1946f70a BM |
268 | } |
269 | ||
8c42d637 | 270 | int gfs2_statfs_sync(struct super_block *sb, int type) |
b3b94faa | 271 | { |
8c42d637 | 272 | struct gfs2_sbd *sdp = sb->s_fs_info; |
feaa7bba | 273 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); |
bd209cc0 AV |
274 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; |
275 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; | |
b3b94faa | 276 | struct gfs2_holder gh; |
70c11ba8 | 277 | struct buffer_head *m_bh; |
b3b94faa DT |
278 | int error; |
279 | ||
280 | error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE, | |
281 | &gh); | |
282 | if (error) | |
2e60d768 | 283 | goto out; |
b3b94faa DT |
284 | |
285 | error = gfs2_meta_inode_buffer(m_ip, &m_bh); | |
286 | if (error) | |
2e60d768 | 287 | goto out_unlock; |
b3b94faa DT |
288 | |
289 | spin_lock(&sdp->sd_statfs_spin); | |
290 | gfs2_statfs_change_in(m_sc, m_bh->b_data + | |
907b9bce | 291 | sizeof(struct gfs2_dinode)); |
b3b94faa DT |
292 | if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) { |
293 | spin_unlock(&sdp->sd_statfs_spin); | |
294 | goto out_bh; | |
295 | } | |
296 | spin_unlock(&sdp->sd_statfs_spin); | |
297 | ||
b3b94faa DT |
298 | error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0); |
299 | if (error) | |
70c11ba8 | 300 | goto out_bh; |
b3b94faa | 301 | |
70c11ba8 | 302 | update_statfs(sdp, m_bh); |
3d3c10f2 | 303 | sdp->sd_statfs_force_sync = 0; |
b3b94faa DT |
304 | |
305 | gfs2_trans_end(sdp); | |
306 | ||
a91ea69f | 307 | out_bh: |
b3b94faa | 308 | brelse(m_bh); |
2e60d768 | 309 | out_unlock: |
b3b94faa | 310 | gfs2_glock_dq_uninit(&gh); |
2e60d768 | 311 | out: |
b3b94faa DT |
312 | return error; |
313 | } | |
314 | ||
b3b94faa DT |
315 | struct lfcc { |
316 | struct list_head list; | |
317 | struct gfs2_holder gh; | |
318 | }; | |
319 | ||
320 | /** | |
321 | * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all | |
322 | * journals are clean | |
323 | * @sdp: the file system | |
b3b94faa DT |
324 | * |
325 | * Returns: errno | |
326 | */ | |
327 | ||
52b1cdcb | 328 | static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp) |
b3b94faa | 329 | { |
5c676f6d | 330 | struct gfs2_inode *ip; |
b3b94faa DT |
331 | struct gfs2_jdesc *jd; |
332 | struct lfcc *lfcc; | |
333 | LIST_HEAD(list); | |
55167622 | 334 | struct gfs2_log_header_host lh; |
b77b4a48 AG |
335 | int error, error2; |
336 | ||
337 | /* | |
338 | * Grab all the journal glocks in SH mode. We are *probably* doing | |
339 | * that to prevent recovery. | |
340 | */ | |
b3b94faa | 341 | |
b3b94faa DT |
342 | list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) { |
343 | lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL); | |
344 | if (!lfcc) { | |
345 | error = -ENOMEM; | |
346 | goto out; | |
347 | } | |
feaa7bba SW |
348 | ip = GFS2_I(jd->jd_inode); |
349 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh); | |
b3b94faa DT |
350 | if (error) { |
351 | kfree(lfcc); | |
352 | goto out; | |
353 | } | |
354 | list_add(&lfcc->list, &list); | |
355 | } | |
356 | ||
b77b4a48 AG |
357 | gfs2_freeze_unlock(&sdp->sd_freeze_gh); |
358 | ||
24972557 | 359 | error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE, |
ebdc416c AG |
360 | LM_FLAG_NOEXP | GL_NOPID, |
361 | &sdp->sd_freeze_gh); | |
52b1cdcb | 362 | if (error) |
b77b4a48 | 363 | goto relock_shared; |
b3b94faa DT |
364 | |
365 | list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) { | |
366 | error = gfs2_jdesc_check(jd); | |
367 | if (error) | |
368 | break; | |
f4686c26 | 369 | error = gfs2_find_jhead(jd, &lh, false); |
b3b94faa DT |
370 | if (error) |
371 | break; | |
372 | if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) { | |
373 | error = -EBUSY; | |
374 | break; | |
375 | } | |
376 | } | |
377 | ||
b77b4a48 AG |
378 | if (!error) |
379 | goto out; /* success */ | |
380 | ||
381 | gfs2_freeze_unlock(&sdp->sd_freeze_gh); | |
382 | ||
383 | relock_shared: | |
6c7410f4 | 384 | error2 = gfs2_freeze_lock_shared(sdp); |
b77b4a48 | 385 | gfs2_assert_withdraw(sdp, !error2); |
b3b94faa | 386 | |
a91ea69f | 387 | out: |
b3b94faa | 388 | while (!list_empty(&list)) { |
969183bc | 389 | lfcc = list_first_entry(&list, struct lfcc, list); |
b3b94faa DT |
390 | list_del(&lfcc->list); |
391 | gfs2_glock_dq_uninit(&lfcc->gh); | |
392 | kfree(lfcc); | |
393 | } | |
b3b94faa DT |
394 | return error; |
395 | } | |
396 | ||
9eed04cd SW |
397 | void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf) |
398 | { | |
7db35444 | 399 | const struct inode *inode = &ip->i_inode; |
9eed04cd SW |
400 | struct gfs2_dinode *str = buf; |
401 | ||
402 | str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC); | |
403 | str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI); | |
404 | str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI); | |
405 | str->di_num.no_addr = cpu_to_be64(ip->i_no_addr); | |
406 | str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino); | |
7db35444 AG |
407 | str->di_mode = cpu_to_be32(inode->i_mode); |
408 | str->di_uid = cpu_to_be32(i_uid_read(inode)); | |
409 | str->di_gid = cpu_to_be32(i_gid_read(inode)); | |
410 | str->di_nlink = cpu_to_be32(inode->i_nlink); | |
411 | str->di_size = cpu_to_be64(i_size_read(inode)); | |
412 | str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(inode)); | |
413 | str->di_atime = cpu_to_be64(inode->i_atime.tv_sec); | |
414 | str->di_mtime = cpu_to_be64(inode->i_mtime.tv_sec); | |
8a8b8d91 | 415 | str->di_ctime = cpu_to_be64(inode_get_ctime(inode).tv_sec); |
9eed04cd SW |
416 | |
417 | str->di_goal_meta = cpu_to_be64(ip->i_goal); | |
418 | str->di_goal_data = cpu_to_be64(ip->i_goal); | |
419 | str->di_generation = cpu_to_be64(ip->i_generation); | |
420 | ||
421 | str->di_flags = cpu_to_be32(ip->i_diskflags); | |
422 | str->di_height = cpu_to_be16(ip->i_height); | |
7db35444 | 423 | str->di_payload_format = cpu_to_be32(S_ISDIR(inode->i_mode) && |
9eed04cd SW |
424 | !(ip->i_diskflags & GFS2_DIF_EXHASH) ? |
425 | GFS2_FORMAT_DE : 0); | |
426 | str->di_depth = cpu_to_be16(ip->i_depth); | |
427 | str->di_entries = cpu_to_be32(ip->i_entries); | |
428 | ||
429 | str->di_eattr = cpu_to_be64(ip->i_eattr); | |
7db35444 AG |
430 | str->di_atime_nsec = cpu_to_be32(inode->i_atime.tv_nsec); |
431 | str->di_mtime_nsec = cpu_to_be32(inode->i_mtime.tv_nsec); | |
8a8b8d91 | 432 | str->di_ctime_nsec = cpu_to_be32(inode_get_ctime(inode).tv_nsec); |
9eed04cd | 433 | } |
9e6e0a12 SW |
434 | |
435 | /** | |
436 | * gfs2_write_inode - Make sure the inode is stable on the disk | |
437 | * @inode: The inode | |
1027efaa | 438 | * @wbc: The writeback control structure |
9e6e0a12 SW |
439 | * |
440 | * Returns: errno | |
441 | */ | |
442 | ||
a9185b41 | 443 | static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc) |
9e6e0a12 SW |
444 | { |
445 | struct gfs2_inode *ip = GFS2_I(inode); | |
446 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1027efaa | 447 | struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl); |
de1414a6 | 448 | struct backing_dev_info *bdi = inode_to_bdi(metamapping->host); |
ab9bbda0 | 449 | int ret = 0; |
adbc3ddf | 450 | bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip)); |
ab9bbda0 | 451 | |
adbc3ddf | 452 | if (flush_all) |
c1696fb8 | 453 | gfs2_log_flush(GFS2_SB(inode), ip->i_gl, |
805c0907 BP |
454 | GFS2_LOG_HEAD_FLUSH_NORMAL | |
455 | GFS2_LFC_WRITE_INODE); | |
a88a341a | 456 | if (bdi->wb.dirty_exceeded) |
4667a0ec | 457 | gfs2_ail1_flush(sdp, wbc); |
1d4ec642 SW |
458 | else |
459 | filemap_fdatawrite(metamapping); | |
adbc3ddf | 460 | if (flush_all) |
1027efaa SW |
461 | ret = filemap_fdatawait(metamapping); |
462 | if (ret) | |
463 | mark_inode_dirty_sync(inode); | |
957a7acd AD |
464 | else { |
465 | spin_lock(&inode->i_lock); | |
466 | if (!(inode->i_flags & I_DIRTY)) | |
467 | gfs2_ordered_del_inode(ip); | |
468 | spin_unlock(&inode->i_lock); | |
469 | } | |
9e6e0a12 SW |
470 | return ret; |
471 | } | |
472 | ||
ab9bbda0 SW |
473 | /** |
474 | * gfs2_dirty_inode - check for atime updates | |
475 | * @inode: The inode in question | |
476 | * @flags: The type of dirty | |
477 | * | |
478 | * Unfortunately it can be called under any combination of inode | |
af1abe11 | 479 | * glock and freeze glock, so we have to check carefully. |
ab9bbda0 SW |
480 | * |
481 | * At the moment this deals only with atime - it should be possible | |
482 | * to expand that role in future, once a review of the locking has | |
483 | * been carried out. | |
484 | */ | |
485 | ||
486 | static void gfs2_dirty_inode(struct inode *inode, int flags) | |
487 | { | |
488 | struct gfs2_inode *ip = GFS2_I(inode); | |
489 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
490 | struct buffer_head *bh; | |
491 | struct gfs2_holder gh; | |
492 | int need_unlock = 0; | |
493 | int need_endtrans = 0; | |
494 | int ret; | |
495 | ||
38552ff6 AG |
496 | if (unlikely(!ip->i_gl)) { |
497 | /* This can only happen during incomplete inode creation. */ | |
498 | BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags)); | |
499 | return; | |
500 | } | |
501 | ||
eb43e660 | 502 | if (unlikely(gfs2_withdrawn(sdp))) |
0d1c7ae9 | 503 | return; |
ab9bbda0 SW |
504 | if (!gfs2_glock_is_locked_by_me(ip->i_gl)) { |
505 | ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); | |
506 | if (ret) { | |
507 | fs_err(sdp, "dirty_inode: glock %d\n", ret); | |
e28c02b9 | 508 | gfs2_dump_glock(NULL, ip->i_gl, true); |
ab9bbda0 SW |
509 | return; |
510 | } | |
511 | need_unlock = 1; | |
3d162688 BM |
512 | } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE)) |
513 | return; | |
ab9bbda0 SW |
514 | |
515 | if (current->journal_info == NULL) { | |
516 | ret = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
517 | if (ret) { | |
518 | fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret); | |
519 | goto out; | |
520 | } | |
521 | need_endtrans = 1; | |
522 | } | |
523 | ||
524 | ret = gfs2_meta_inode_buffer(ip, &bh); | |
525 | if (ret == 0) { | |
350a9b0a | 526 | gfs2_trans_add_meta(ip->i_gl, bh); |
ab9bbda0 SW |
527 | gfs2_dinode_out(ip, bh->b_data); |
528 | brelse(bh); | |
529 | } | |
530 | ||
531 | if (need_endtrans) | |
532 | gfs2_trans_end(sdp); | |
533 | out: | |
534 | if (need_unlock) | |
535 | gfs2_glock_dq_uninit(&gh); | |
536 | } | |
537 | ||
9e6e0a12 SW |
538 | /** |
539 | * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one | |
540 | * @sdp: the filesystem | |
541 | * | |
542 | * Returns: errno | |
543 | */ | |
544 | ||
eb602521 | 545 | void gfs2_make_fs_ro(struct gfs2_sbd *sdp) |
9e6e0a12 | 546 | { |
601ef0d5 BP |
547 | int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags); |
548 | ||
e7beb8b6 | 549 | if (!test_bit(SDF_KILL, &sdp->sd_flags)) |
6b388abc AG |
550 | gfs2_flush_delete_work(sdp); |
551 | ||
fe4f7940 | 552 | gfs2_destroy_threads(sdp); |
8ad151c2 | 553 | |
601ef0d5 BP |
554 | if (log_write_allowed) { |
555 | gfs2_quota_sync(sdp->sd_vfs, 0); | |
556 | gfs2_statfs_sync(sdp->sd_vfs, 0); | |
9e6e0a12 | 557 | |
68ca088d BP |
558 | /* We do two log flushes here. The first one commits dirty inodes |
559 | * and rgrps to the journal, but queues up revokes to the ail list. | |
560 | * The second flush writes out and removes the revokes. | |
561 | * | |
562 | * The first must be done before the FLUSH_SHUTDOWN code | |
563 | * clears the LIVE flag, otherwise it will not be able to start | |
564 | * a transaction to write its revokes, and the error will cause | |
565 | * a withdraw of the file system. */ | |
566 | gfs2_log_flush(sdp, NULL, GFS2_LFC_MAKE_FS_RO); | |
601ef0d5 BP |
567 | gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN | |
568 | GFS2_LFC_MAKE_FS_RO); | |
f3708fb5 AG |
569 | wait_event_timeout(sdp->sd_log_waitq, |
570 | gfs2_log_is_empty(sdp), | |
571 | HZ * 5); | |
572 | gfs2_assert_warn(sdp, gfs2_log_is_empty(sdp)); | |
601ef0d5 | 573 | } |
9e6e0a12 | 574 | gfs2_quota_cleanup(sdp); |
9e6e0a12 SW |
575 | } |
576 | ||
9e6e0a12 SW |
577 | /** |
578 | * gfs2_put_super - Unmount the filesystem | |
579 | * @sb: The VFS superblock | |
580 | * | |
581 | */ | |
582 | ||
583 | static void gfs2_put_super(struct super_block *sb) | |
584 | { | |
585 | struct gfs2_sbd *sdp = sb->s_fs_info; | |
9e6e0a12 SW |
586 | struct gfs2_jdesc *jd; |
587 | ||
9e6e0a12 SW |
588 | /* No more recovery requests */ |
589 | set_bit(SDF_NORECOVERY, &sdp->sd_flags); | |
590 | smp_mb(); | |
591 | ||
592 | /* Wait on outstanding recovery */ | |
593 | restart: | |
594 | spin_lock(&sdp->sd_jindex_spin); | |
595 | list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) { | |
596 | if (!test_bit(JDF_RECOVERY, &jd->jd_flags)) | |
597 | continue; | |
598 | spin_unlock(&sdp->sd_jindex_spin); | |
599 | wait_on_bit(&jd->jd_flags, JDF_RECOVERY, | |
74316201 | 600 | TASK_UNINTERRUPTIBLE); |
9e6e0a12 SW |
601 | goto restart; |
602 | } | |
603 | spin_unlock(&sdp->sd_jindex_spin); | |
604 | ||
bc98a42c | 605 | if (!sb_rdonly(sb)) { |
eb602521 | 606 | gfs2_make_fs_ro(sdp); |
9e6e0a12 | 607 | } |
fe4f7940 AG |
608 | if (gfs2_withdrawn(sdp)) { |
609 | gfs2_destroy_threads(sdp); | |
f66af88e | 610 | gfs2_quota_cleanup(sdp); |
fe4f7940 | 611 | } |
5a61ae14 AG |
612 | WARN_ON(gfs2_withdrawing(sdp)); |
613 | ||
9e6e0a12 SW |
614 | /* At this point, we're through modifying the disk */ |
615 | ||
616 | /* Release stuff */ | |
617 | ||
b77b4a48 AG |
618 | gfs2_freeze_unlock(&sdp->sd_freeze_gh); |
619 | ||
9e6e0a12 | 620 | iput(sdp->sd_jindex); |
9e6e0a12 SW |
621 | iput(sdp->sd_statfs_inode); |
622 | iput(sdp->sd_rindex); | |
623 | iput(sdp->sd_quota_inode); | |
624 | ||
625 | gfs2_glock_put(sdp->sd_rename_gl); | |
24972557 | 626 | gfs2_glock_put(sdp->sd_freeze_gl); |
9e6e0a12 SW |
627 | |
628 | if (!sdp->sd_args.ar_spectator) { | |
601ef0d5 BP |
629 | if (gfs2_holder_initialized(&sdp->sd_journal_gh)) |
630 | gfs2_glock_dq_uninit(&sdp->sd_journal_gh); | |
631 | if (gfs2_holder_initialized(&sdp->sd_jinode_gh)) | |
632 | gfs2_glock_dq_uninit(&sdp->sd_jinode_gh); | |
70c11ba8 | 633 | brelse(sdp->sd_sc_bh); |
9e6e0a12 SW |
634 | gfs2_glock_dq_uninit(&sdp->sd_sc_gh); |
635 | gfs2_glock_dq_uninit(&sdp->sd_qc_gh); | |
97fd734b | 636 | free_local_statfs_inodes(sdp); |
9e6e0a12 SW |
637 | iput(sdp->sd_qc_inode); |
638 | } | |
639 | ||
640 | gfs2_glock_dq_uninit(&sdp->sd_live_gh); | |
641 | gfs2_clear_rgrpd(sdp); | |
642 | gfs2_jindex_free(sdp); | |
643 | /* Take apart glock structures and buffer lists */ | |
644 | gfs2_gl_hash_clear(sdp); | |
a9dd945c | 645 | truncate_inode_pages_final(&sdp->sd_aspace); |
b2fb7dab | 646 | gfs2_delete_debugfs_file(sdp); |
9e6e0a12 SW |
647 | /* Unmount the locking protocol */ |
648 | gfs2_lm_unmount(sdp); | |
649 | ||
650 | /* At this point, we're through participating in the lockspace */ | |
651 | gfs2_sys_fs_del(sdp); | |
c2a04b02 | 652 | free_sbd(sdp); |
9e6e0a12 SW |
653 | } |
654 | ||
9e6e0a12 SW |
655 | /** |
656 | * gfs2_sync_fs - sync the filesystem | |
657 | * @sb: the superblock | |
c551f66c | 658 | * @wait: true to wait for completion |
9e6e0a12 SW |
659 | * |
660 | * Flushes the log to disk. | |
661 | */ | |
662 | ||
663 | static int gfs2_sync_fs(struct super_block *sb, int wait) | |
664 | { | |
1027efaa | 665 | struct gfs2_sbd *sdp = sb->s_fs_info; |
a1177825 JK |
666 | |
667 | gfs2_quota_sync(sb, -1); | |
942b0cdd | 668 | if (wait) |
805c0907 BP |
669 | gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL | |
670 | GFS2_LFC_SYNC_FS); | |
942b0cdd | 671 | return sdp->sd_log_error; |
9e6e0a12 SW |
672 | } |
673 | ||
b77b4a48 | 674 | static int gfs2_freeze_locally(struct gfs2_sbd *sdp) |
2e60d768 | 675 | { |
2e60d768 | 676 | struct super_block *sb = sdp->sd_vfs; |
b77b4a48 | 677 | int error; |
2e60d768 | 678 | |
880b9577 | 679 | error = freeze_super(sb, FREEZE_HOLDER_USERSPACE); |
b77b4a48 | 680 | if (error) |
5432af15 | 681 | return error; |
b77b4a48 AG |
682 | |
683 | if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) { | |
684 | gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE | | |
685 | GFS2_LFC_FREEZE_GO_SYNC); | |
686 | if (gfs2_withdrawn(sdp)) { | |
880b9577 DW |
687 | error = thaw_super(sb, FREEZE_HOLDER_USERSPACE); |
688 | if (error) | |
689 | return error; | |
5432af15 | 690 | return -EIO; |
2e60d768 | 691 | } |
2e60d768 | 692 | } |
b77b4a48 | 693 | return 0; |
b77b4a48 AG |
694 | } |
695 | ||
696 | static int gfs2_do_thaw(struct gfs2_sbd *sdp) | |
697 | { | |
698 | struct super_block *sb = sdp->sd_vfs; | |
699 | int error; | |
700 | ||
6c7410f4 | 701 | error = gfs2_freeze_lock_shared(sdp); |
b77b4a48 AG |
702 | if (error) |
703 | goto fail; | |
880b9577 | 704 | error = thaw_super(sb, FREEZE_HOLDER_USERSPACE); |
b77b4a48 AG |
705 | if (!error) |
706 | return 0; | |
707 | ||
708 | fail: | |
709 | fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n", error); | |
710 | gfs2_assert_withdraw(sdp, 0); | |
711 | return error; | |
712 | } | |
713 | ||
714 | void gfs2_freeze_func(struct work_struct *work) | |
715 | { | |
716 | struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work); | |
717 | struct super_block *sb = sdp->sd_vfs; | |
718 | int error; | |
719 | ||
720 | mutex_lock(&sdp->sd_freeze_mutex); | |
721 | error = -EBUSY; | |
5432af15 | 722 | if (test_bit(SDF_FROZEN, &sdp->sd_flags)) |
b77b4a48 AG |
723 | goto freeze_failed; |
724 | ||
725 | error = gfs2_freeze_locally(sdp); | |
726 | if (error) | |
727 | goto freeze_failed; | |
728 | ||
729 | gfs2_freeze_unlock(&sdp->sd_freeze_gh); | |
5432af15 | 730 | set_bit(SDF_FROZEN, &sdp->sd_flags); |
b77b4a48 AG |
731 | |
732 | error = gfs2_do_thaw(sdp); | |
733 | if (error) | |
734 | goto out; | |
735 | ||
5432af15 | 736 | clear_bit(SDF_FROZEN, &sdp->sd_flags); |
b77b4a48 AG |
737 | goto out; |
738 | ||
739 | freeze_failed: | |
740 | fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n", error); | |
741 | ||
742 | out: | |
743 | mutex_unlock(&sdp->sd_freeze_mutex); | |
2e60d768 | 744 | deactivate_super(sb); |
2e60d768 BM |
745 | } |
746 | ||
9e6e0a12 | 747 | /** |
097cca52 | 748 | * gfs2_freeze_super - prevent further writes to the filesystem |
9e6e0a12 SW |
749 | * @sb: the VFS structure for the filesystem |
750 | * | |
751 | */ | |
752 | ||
880b9577 | 753 | static int gfs2_freeze_super(struct super_block *sb, enum freeze_holder who) |
9e6e0a12 SW |
754 | { |
755 | struct gfs2_sbd *sdp = sb->s_fs_info; | |
ff132c5f | 756 | int error; |
9e6e0a12 | 757 | |
b77b4a48 AG |
758 | if (!mutex_trylock(&sdp->sd_freeze_mutex)) |
759 | return -EBUSY; | |
760 | error = -EBUSY; | |
5432af15 | 761 | if (test_bit(SDF_FROZEN, &sdp->sd_flags)) |
2e60d768 BM |
762 | goto out; |
763 | ||
9e6e0a12 | 764 | for (;;) { |
b77b4a48 AG |
765 | error = gfs2_freeze_locally(sdp); |
766 | if (error) { | |
767 | fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n", | |
768 | error); | |
60528afa BP |
769 | goto out; |
770 | } | |
771 | ||
52b1cdcb | 772 | error = gfs2_lock_fs_check_clean(sdp); |
9e6e0a12 | 773 | if (!error) |
b77b4a48 AG |
774 | break; /* success */ |
775 | ||
776 | error = gfs2_do_thaw(sdp); | |
777 | if (error) | |
778 | goto out; | |
9e6e0a12 | 779 | |
55317f5b | 780 | if (error == -EBUSY) |
9e6e0a12 | 781 | fs_err(sdp, "waiting for recovery before freeze\n"); |
52b1cdcb BP |
782 | else if (error == -EIO) { |
783 | fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due " | |
784 | "to recovery error.\n"); | |
785 | goto out; | |
786 | } else { | |
9e6e0a12 | 787 | fs_err(sdp, "error freezing FS: %d\n", error); |
52b1cdcb | 788 | } |
9e6e0a12 SW |
789 | fs_err(sdp, "retrying...\n"); |
790 | msleep(1000); | |
791 | } | |
b77b4a48 | 792 | |
2e60d768 | 793 | out: |
b77b4a48 AG |
794 | if (!error) { |
795 | set_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags); | |
5432af15 | 796 | set_bit(SDF_FROZEN, &sdp->sd_flags); |
b77b4a48 | 797 | } |
2e60d768 BM |
798 | mutex_unlock(&sdp->sd_freeze_mutex); |
799 | return error; | |
9e6e0a12 SW |
800 | } |
801 | ||
802 | /** | |
097cca52 | 803 | * gfs2_thaw_super - reallow writes to the filesystem |
9e6e0a12 SW |
804 | * @sb: the VFS structure for the filesystem |
805 | * | |
806 | */ | |
807 | ||
880b9577 | 808 | static int gfs2_thaw_super(struct super_block *sb, enum freeze_holder who) |
9e6e0a12 | 809 | { |
d564053f | 810 | struct gfs2_sbd *sdp = sb->s_fs_info; |
b77b4a48 | 811 | int error; |
d564053f | 812 | |
b77b4a48 AG |
813 | if (!mutex_trylock(&sdp->sd_freeze_mutex)) |
814 | return -EBUSY; | |
815 | error = -EINVAL; | |
816 | if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags)) | |
817 | goto out; | |
818 | ||
819 | gfs2_freeze_unlock(&sdp->sd_freeze_gh); | |
820 | ||
821 | error = gfs2_do_thaw(sdp); | |
822 | ||
823 | if (!error) { | |
824 | clear_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags); | |
5432af15 | 825 | clear_bit(SDF_FROZEN, &sdp->sd_flags); |
2e60d768 | 826 | } |
b77b4a48 AG |
827 | out: |
828 | mutex_unlock(&sdp->sd_freeze_mutex); | |
829 | return error; | |
830 | } | |
831 | ||
832 | void gfs2_thaw_freeze_initiator(struct super_block *sb) | |
833 | { | |
834 | struct gfs2_sbd *sdp = sb->s_fs_info; | |
835 | ||
836 | mutex_lock(&sdp->sd_freeze_mutex); | |
837 | if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags)) | |
838 | goto out; | |
2e60d768 | 839 | |
c77b52c0 | 840 | gfs2_freeze_unlock(&sdp->sd_freeze_gh); |
b77b4a48 AG |
841 | |
842 | out: | |
2e60d768 | 843 | mutex_unlock(&sdp->sd_freeze_mutex); |
9e6e0a12 SW |
844 | } |
845 | ||
846 | /** | |
c551f66c | 847 | * statfs_slow_fill - fill in the sg for a given RG |
9e6e0a12 SW |
848 | * @rgd: the RG |
849 | * @sc: the sc structure | |
850 | * | |
851 | * Returns: 0 on success, -ESTALE if the LVB is invalid | |
852 | */ | |
853 | ||
854 | static int statfs_slow_fill(struct gfs2_rgrpd *rgd, | |
855 | struct gfs2_statfs_change_host *sc) | |
856 | { | |
857 | gfs2_rgrp_verify(rgd); | |
858 | sc->sc_total += rgd->rd_data; | |
859 | sc->sc_free += rgd->rd_free; | |
860 | sc->sc_dinodes += rgd->rd_dinodes; | |
861 | return 0; | |
862 | } | |
863 | ||
864 | /** | |
865 | * gfs2_statfs_slow - Stat a filesystem using asynchronous locking | |
866 | * @sdp: the filesystem | |
867 | * @sc: the sc info that will be returned | |
868 | * | |
869 | * Any error (other than a signal) will cause this routine to fall back | |
870 | * to the synchronous version. | |
871 | * | |
872 | * FIXME: This really shouldn't busy wait like this. | |
873 | * | |
874 | * Returns: errno | |
875 | */ | |
876 | ||
877 | static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc) | |
878 | { | |
9e6e0a12 SW |
879 | struct gfs2_rgrpd *rgd_next; |
880 | struct gfs2_holder *gha, *gh; | |
881 | unsigned int slots = 64; | |
882 | unsigned int x; | |
883 | int done; | |
884 | int error = 0, err; | |
885 | ||
886 | memset(sc, 0, sizeof(struct gfs2_statfs_change_host)); | |
6da2ec56 | 887 | gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL); |
9e6e0a12 SW |
888 | if (!gha) |
889 | return -ENOMEM; | |
6df9f9a2 AG |
890 | for (x = 0; x < slots; x++) |
891 | gfs2_holder_mark_uninitialized(gha + x); | |
9e6e0a12 | 892 | |
9e6e0a12 SW |
893 | rgd_next = gfs2_rgrpd_get_first(sdp); |
894 | ||
895 | for (;;) { | |
896 | done = 1; | |
897 | ||
898 | for (x = 0; x < slots; x++) { | |
899 | gh = gha + x; | |
900 | ||
6df9f9a2 | 901 | if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) { |
9e6e0a12 SW |
902 | err = gfs2_glock_wait(gh); |
903 | if (err) { | |
904 | gfs2_holder_uninit(gh); | |
905 | error = err; | |
906 | } else { | |
6f6597ba AG |
907 | if (!error) { |
908 | struct gfs2_rgrpd *rgd = | |
909 | gfs2_glock2rgrp(gh->gh_gl); | |
910 | ||
911 | error = statfs_slow_fill(rgd, sc); | |
912 | } | |
9e6e0a12 SW |
913 | gfs2_glock_dq_uninit(gh); |
914 | } | |
915 | } | |
916 | ||
6df9f9a2 | 917 | if (gfs2_holder_initialized(gh)) |
9e6e0a12 SW |
918 | done = 0; |
919 | else if (rgd_next && !error) { | |
920 | error = gfs2_glock_nq_init(rgd_next->rd_gl, | |
921 | LM_ST_SHARED, | |
922 | GL_ASYNC, | |
923 | gh); | |
924 | rgd_next = gfs2_rgrpd_get_next(rgd_next); | |
925 | done = 0; | |
926 | } | |
927 | ||
928 | if (signal_pending(current)) | |
929 | error = -ERESTARTSYS; | |
930 | } | |
931 | ||
932 | if (done) | |
933 | break; | |
934 | ||
935 | yield(); | |
936 | } | |
937 | ||
9e6e0a12 SW |
938 | kfree(gha); |
939 | return error; | |
940 | } | |
941 | ||
942 | /** | |
943 | * gfs2_statfs_i - Do a statfs | |
944 | * @sdp: the filesystem | |
c551f66c | 945 | * @sc: the sc structure |
9e6e0a12 SW |
946 | * |
947 | * Returns: errno | |
948 | */ | |
949 | ||
950 | static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc) | |
951 | { | |
952 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; | |
953 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; | |
954 | ||
955 | spin_lock(&sdp->sd_statfs_spin); | |
956 | ||
957 | *sc = *m_sc; | |
958 | sc->sc_total += l_sc->sc_total; | |
959 | sc->sc_free += l_sc->sc_free; | |
960 | sc->sc_dinodes += l_sc->sc_dinodes; | |
961 | ||
962 | spin_unlock(&sdp->sd_statfs_spin); | |
963 | ||
964 | if (sc->sc_free < 0) | |
965 | sc->sc_free = 0; | |
966 | if (sc->sc_free > sc->sc_total) | |
967 | sc->sc_free = sc->sc_total; | |
968 | if (sc->sc_dinodes < 0) | |
969 | sc->sc_dinodes = 0; | |
970 | ||
971 | return 0; | |
972 | } | |
973 | ||
974 | /** | |
975 | * gfs2_statfs - Gather and return stats about the filesystem | |
c551f66c LJ |
976 | * @dentry: The name of the link |
977 | * @buf: The buffer | |
9e6e0a12 SW |
978 | * |
979 | * Returns: 0 on success or error code | |
980 | */ | |
981 | ||
982 | static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf) | |
983 | { | |
fc64005c | 984 | struct super_block *sb = dentry->d_sb; |
9e6e0a12 SW |
985 | struct gfs2_sbd *sdp = sb->s_fs_info; |
986 | struct gfs2_statfs_change_host sc; | |
987 | int error; | |
988 | ||
8339ee54 SW |
989 | error = gfs2_rindex_update(sdp); |
990 | if (error) | |
991 | return error; | |
992 | ||
9e6e0a12 SW |
993 | if (gfs2_tune_get(sdp, gt_statfs_slow)) |
994 | error = gfs2_statfs_slow(sdp, &sc); | |
995 | else | |
996 | error = gfs2_statfs_i(sdp, &sc); | |
997 | ||
998 | if (error) | |
999 | return error; | |
1000 | ||
1001 | buf->f_type = GFS2_MAGIC; | |
1002 | buf->f_bsize = sdp->sd_sb.sb_bsize; | |
1003 | buf->f_blocks = sc.sc_total; | |
1004 | buf->f_bfree = sc.sc_free; | |
1005 | buf->f_bavail = sc.sc_free; | |
1006 | buf->f_files = sc.sc_dinodes + sc.sc_free; | |
1007 | buf->f_ffree = sc.sc_free; | |
1008 | buf->f_namelen = GFS2_FNAMESIZE; | |
d6fc6c93 | 1009 | buf->f_fsid = uuid_to_fsid(sb->s_uuid.b); |
9e6e0a12 SW |
1010 | |
1011 | return 0; | |
1012 | } | |
1013 | ||
9e6e0a12 SW |
1014 | /** |
1015 | * gfs2_drop_inode - Drop an inode (test for remote unlink) | |
1016 | * @inode: The inode to drop | |
1017 | * | |
61b91cfd | 1018 | * If we've received a callback on an iopen lock then it's because a |
9e6e0a12 SW |
1019 | * remote node tried to deallocate the inode but failed due to this node |
1020 | * still having the inode open. Here we mark the link count zero | |
1021 | * since we know that it must have reached zero if the GLF_DEMOTE flag | |
1022 | * is set on the iopen glock. If we didn't do a disk read since the | |
1023 | * remote node removed the final link then we might otherwise miss | |
1024 | * this event. This check ensures that this node will deallocate the | |
1025 | * inode's blocks, or alternatively pass the baton on to another | |
1026 | * node for later deallocation. | |
1027 | */ | |
1028 | ||
45321ac5 | 1029 | static int gfs2_drop_inode(struct inode *inode) |
9e6e0a12 SW |
1030 | { |
1031 | struct gfs2_inode *ip = GFS2_I(inode); | |
b88beb9a | 1032 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
9e6e0a12 | 1033 | |
38552ff6 | 1034 | if (inode->i_nlink && |
6df9f9a2 | 1035 | gfs2_holder_initialized(&ip->i_iopen_gh)) { |
9e6e0a12 | 1036 | struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl; |
6df9f9a2 | 1037 | if (test_bit(GLF_DEMOTE, &gl->gl_flags)) |
9e6e0a12 SW |
1038 | clear_nlink(inode); |
1039 | } | |
6a1c8f6d AG |
1040 | |
1041 | /* | |
1042 | * When under memory pressure when an inode's link count has dropped to | |
1043 | * zero, defer deleting the inode to the delete workqueue. This avoids | |
1044 | * calling into DLM under memory pressure, which can deadlock. | |
1045 | */ | |
1046 | if (!inode->i_nlink && | |
1047 | unlikely(current->flags & PF_MEMALLOC) && | |
1048 | gfs2_holder_initialized(&ip->i_iopen_gh)) { | |
1049 | struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl; | |
1050 | ||
1051 | gfs2_glock_hold(gl); | |
f0e56edc | 1052 | if (!gfs2_queue_try_to_evict(gl)) |
6a1c8f6d | 1053 | gfs2_glock_queue_put(gl); |
ba3ca2bc | 1054 | return 0; |
6a1c8f6d AG |
1055 | } |
1056 | ||
b88beb9a AG |
1057 | /* |
1058 | * No longer cache inodes when trying to evict them all. | |
1059 | */ | |
1060 | if (test_bit(SDF_EVICTING, &sdp->sd_flags)) | |
1061 | return 1; | |
1062 | ||
45321ac5 | 1063 | return generic_drop_inode(inode); |
9e6e0a12 SW |
1064 | } |
1065 | ||
1066 | static int is_ancestor(const struct dentry *d1, const struct dentry *d2) | |
1067 | { | |
1068 | do { | |
1069 | if (d1 == d2) | |
1070 | return 1; | |
1071 | d1 = d1->d_parent; | |
1072 | } while (!IS_ROOT(d1)); | |
1073 | return 0; | |
1074 | } | |
1075 | ||
1076 | /** | |
1077 | * gfs2_show_options - Show mount options for /proc/mounts | |
1078 | * @s: seq_file structure | |
34c80b1d | 1079 | * @root: root of this (sub)tree |
9e6e0a12 SW |
1080 | * |
1081 | * Returns: 0 on success or error code | |
1082 | */ | |
1083 | ||
34c80b1d | 1084 | static int gfs2_show_options(struct seq_file *s, struct dentry *root) |
9e6e0a12 | 1085 | { |
34c80b1d | 1086 | struct gfs2_sbd *sdp = root->d_sb->s_fs_info; |
9e6e0a12 | 1087 | struct gfs2_args *args = &sdp->sd_args; |
6fa0a72c TL |
1088 | unsigned int logd_secs, statfs_slow, statfs_quantum, quota_quantum; |
1089 | ||
1090 | spin_lock(&sdp->sd_tune.gt_spin); | |
1091 | logd_secs = sdp->sd_tune.gt_logd_secs; | |
1092 | quota_quantum = sdp->sd_tune.gt_quota_quantum; | |
1093 | statfs_quantum = sdp->sd_tune.gt_statfs_quantum; | |
1094 | statfs_slow = sdp->sd_tune.gt_statfs_slow; | |
1095 | spin_unlock(&sdp->sd_tune.gt_spin); | |
9e6e0a12 | 1096 | |
34c80b1d | 1097 | if (is_ancestor(root, sdp->sd_master_dir)) |
eaebdedc | 1098 | seq_puts(s, ",meta"); |
9e6e0a12 | 1099 | if (args->ar_lockproto[0]) |
a068acf2 | 1100 | seq_show_option(s, "lockproto", args->ar_lockproto); |
9e6e0a12 | 1101 | if (args->ar_locktable[0]) |
a068acf2 | 1102 | seq_show_option(s, "locktable", args->ar_locktable); |
9e6e0a12 | 1103 | if (args->ar_hostdata[0]) |
a068acf2 | 1104 | seq_show_option(s, "hostdata", args->ar_hostdata); |
9e6e0a12 | 1105 | if (args->ar_spectator) |
eaebdedc | 1106 | seq_puts(s, ",spectator"); |
9e6e0a12 | 1107 | if (args->ar_localflocks) |
eaebdedc | 1108 | seq_puts(s, ",localflocks"); |
9e6e0a12 | 1109 | if (args->ar_debug) |
eaebdedc | 1110 | seq_puts(s, ",debug"); |
9e6e0a12 | 1111 | if (args->ar_posix_acl) |
eaebdedc | 1112 | seq_puts(s, ",acl"); |
9e6e0a12 SW |
1113 | if (args->ar_quota != GFS2_QUOTA_DEFAULT) { |
1114 | char *state; | |
1115 | switch (args->ar_quota) { | |
1116 | case GFS2_QUOTA_OFF: | |
1117 | state = "off"; | |
1118 | break; | |
1119 | case GFS2_QUOTA_ACCOUNT: | |
1120 | state = "account"; | |
1121 | break; | |
1122 | case GFS2_QUOTA_ON: | |
1123 | state = "on"; | |
1124 | break; | |
eef46ab7 BP |
1125 | case GFS2_QUOTA_QUIET: |
1126 | state = "quiet"; | |
1127 | break; | |
9e6e0a12 SW |
1128 | default: |
1129 | state = "unknown"; | |
1130 | break; | |
1131 | } | |
1132 | seq_printf(s, ",quota=%s", state); | |
1133 | } | |
1134 | if (args->ar_suiddir) | |
eaebdedc | 1135 | seq_puts(s, ",suiddir"); |
9e6e0a12 SW |
1136 | if (args->ar_data != GFS2_DATA_DEFAULT) { |
1137 | char *state; | |
1138 | switch (args->ar_data) { | |
1139 | case GFS2_DATA_WRITEBACK: | |
1140 | state = "writeback"; | |
1141 | break; | |
1142 | case GFS2_DATA_ORDERED: | |
1143 | state = "ordered"; | |
1144 | break; | |
1145 | default: | |
1146 | state = "unknown"; | |
1147 | break; | |
1148 | } | |
1149 | seq_printf(s, ",data=%s", state); | |
1150 | } | |
1151 | if (args->ar_discard) | |
eaebdedc | 1152 | seq_puts(s, ",discard"); |
6fa0a72c TL |
1153 | if (logd_secs != 30) |
1154 | seq_printf(s, ",commit=%d", logd_secs); | |
1155 | if (statfs_quantum != 30) | |
1156 | seq_printf(s, ",statfs_quantum=%d", statfs_quantum); | |
1157 | else if (statfs_slow) | |
2b9731e8 | 1158 | seq_puts(s, ",statfs_quantum=0"); |
6fa0a72c TL |
1159 | if (quota_quantum != 60) |
1160 | seq_printf(s, ",quota_quantum=%d", quota_quantum); | |
3d3c10f2 BM |
1161 | if (args->ar_statfs_percent) |
1162 | seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent); | |
d34843d0 BP |
1163 | if (args->ar_errors != GFS2_ERRORS_DEFAULT) { |
1164 | const char *state; | |
1165 | ||
1166 | switch (args->ar_errors) { | |
1167 | case GFS2_ERRORS_WITHDRAW: | |
1168 | state = "withdraw"; | |
1169 | break; | |
1170 | case GFS2_ERRORS_PANIC: | |
1171 | state = "panic"; | |
1172 | break; | |
1173 | default: | |
1174 | state = "unknown"; | |
1175 | break; | |
1176 | } | |
1177 | seq_printf(s, ",errors=%s", state); | |
1178 | } | |
cdcfde62 | 1179 | if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) |
eaebdedc | 1180 | seq_puts(s, ",nobarrier"); |
913a71d2 | 1181 | if (test_bit(SDF_DEMOTE, &sdp->sd_flags)) |
eaebdedc | 1182 | seq_puts(s, ",demote_interface_used"); |
90306c41 | 1183 | if (args->ar_rgrplvb) |
eaebdedc | 1184 | seq_puts(s, ",rgrplvb"); |
471f3db2 BM |
1185 | if (args->ar_loccookie) |
1186 | seq_puts(s, ",loccookie"); | |
9e6e0a12 SW |
1187 | return 0; |
1188 | } | |
1189 | ||
f42ab085 SW |
1190 | static void gfs2_final_release_pages(struct gfs2_inode *ip) |
1191 | { | |
1192 | struct inode *inode = &ip->i_inode; | |
1193 | struct gfs2_glock *gl = ip->i_gl; | |
1194 | ||
38552ff6 AG |
1195 | if (unlikely(!gl)) { |
1196 | /* This can only happen during incomplete inode creation. */ | |
1197 | BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags)); | |
1198 | return; | |
1199 | } | |
1200 | ||
1201 | truncate_inode_pages(gfs2_glock2aspace(gl), 0); | |
f42ab085 SW |
1202 | truncate_inode_pages(&inode->i_data, 0); |
1203 | ||
638803d4 | 1204 | if (atomic_read(&gl->gl_revokes) == 0) { |
f42ab085 SW |
1205 | clear_bit(GLF_LFLUSH, &gl->gl_flags); |
1206 | clear_bit(GLF_DIRTY, &gl->gl_flags); | |
1207 | } | |
1208 | } | |
1209 | ||
1210 | static int gfs2_dinode_dealloc(struct gfs2_inode *ip) | |
1211 | { | |
1212 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
f42ab085 | 1213 | struct gfs2_rgrpd *rgd; |
564e12b1 | 1214 | struct gfs2_holder gh; |
f42ab085 SW |
1215 | int error; |
1216 | ||
1217 | if (gfs2_get_inode_blocks(&ip->i_inode) != 1) { | |
94fb763b | 1218 | gfs2_consist_inode(ip); |
f42ab085 SW |
1219 | return -EIO; |
1220 | } | |
1221 | ||
9b620429 | 1222 | gfs2_rindex_update(sdp); |
f42ab085 | 1223 | |
f4108a60 | 1224 | error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE); |
f42ab085 | 1225 | if (error) |
5407e242 | 1226 | return error; |
f42ab085 | 1227 | |
66fc061b | 1228 | rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1); |
f42ab085 SW |
1229 | if (!rgd) { |
1230 | gfs2_consist_inode(ip); | |
1231 | error = -EIO; | |
8339ee54 | 1232 | goto out_qs; |
f42ab085 SW |
1233 | } |
1234 | ||
4fc7ec31 BP |
1235 | error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, |
1236 | LM_FLAG_NODE_SCOPE, &gh); | |
f42ab085 | 1237 | if (error) |
8339ee54 | 1238 | goto out_qs; |
f42ab085 | 1239 | |
4667a0ec SW |
1240 | error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA, |
1241 | sdp->sd_jdesc->jd_blocks); | |
f42ab085 SW |
1242 | if (error) |
1243 | goto out_rg_gunlock; | |
1244 | ||
1245 | gfs2_free_di(rgd, ip); | |
1246 | ||
1247 | gfs2_final_release_pages(ip); | |
1248 | ||
1249 | gfs2_trans_end(sdp); | |
1250 | ||
1251 | out_rg_gunlock: | |
564e12b1 | 1252 | gfs2_glock_dq_uninit(&gh); |
f42ab085 SW |
1253 | out_qs: |
1254 | gfs2_quota_unhold(ip); | |
f42ab085 SW |
1255 | return error; |
1256 | } | |
1257 | ||
71c1b213 AG |
1258 | /** |
1259 | * gfs2_glock_put_eventually | |
1260 | * @gl: The glock to put | |
1261 | * | |
1262 | * When under memory pressure, trigger a deferred glock put to make sure we | |
1263 | * won't call into DLM and deadlock. Otherwise, put the glock directly. | |
1264 | */ | |
1265 | ||
1266 | static void gfs2_glock_put_eventually(struct gfs2_glock *gl) | |
1267 | { | |
1268 | if (current->flags & PF_MEMALLOC) | |
1269 | gfs2_glock_queue_put(gl); | |
1270 | else | |
1271 | gfs2_glock_put(gl); | |
1272 | } | |
1273 | ||
9e73330f AG |
1274 | static bool gfs2_upgrade_iopen_glock(struct inode *inode) |
1275 | { | |
1276 | struct gfs2_inode *ip = GFS2_I(inode); | |
1277 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1278 | struct gfs2_holder *gh = &ip->i_iopen_gh; | |
1279 | long timeout = 5 * HZ; | |
1280 | int error; | |
1281 | ||
1282 | gh->gh_flags |= GL_NOCACHE; | |
1283 | gfs2_glock_dq_wait(gh); | |
1284 | ||
1285 | /* | |
2d143955 AG |
1286 | * If there are no other lock holders, we will immediately get |
1287 | * exclusive access to the iopen glock here. | |
1288 | * | |
9e73330f | 1289 | * Otherwise, the other nodes holding the lock will be notified about |
2d143955 AG |
1290 | * our locking request. If they do not have the inode open, they are |
1291 | * expected to evict the cached inode and release the lock, allowing us | |
1292 | * to proceed. | |
1293 | * | |
1294 | * Otherwise, if they cannot evict the inode, they are expected to poke | |
1295 | * the inode glock (note: not the iopen glock). We will notice that | |
1296 | * and stop waiting for the iopen glock immediately. The other node(s) | |
1297 | * are then expected to take care of deleting the inode when they no | |
1298 | * longer use it. | |
1299 | * | |
1300 | * As a last resort, if another node keeps holding the iopen glock | |
1301 | * without showing any activity on the inode glock, we will eventually | |
1302 | * time out and fail the iopen glock upgrade. | |
9e73330f AG |
1303 | */ |
1304 | ||
9e73330f AG |
1305 | gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh); |
1306 | error = gfs2_glock_nq(gh); | |
1307 | if (error) | |
1308 | return false; | |
1309 | ||
1310 | timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait, | |
9e8990de AG |
1311 | !test_bit(HIF_WAIT, &gh->gh_iflags) || |
1312 | test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags), | |
9e73330f AG |
1313 | timeout); |
1314 | if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) { | |
1315 | gfs2_glock_dq(gh); | |
1316 | return false; | |
1317 | } | |
53d69132 | 1318 | return gfs2_glock_holder_ready(gh) == 0; |
9e73330f AG |
1319 | } |
1320 | ||
53dbc27e BP |
1321 | /** |
1322 | * evict_should_delete - determine whether the inode is eligible for deletion | |
1323 | * @inode: The inode to evict | |
c551f66c | 1324 | * @gh: The glock holder structure |
53dbc27e BP |
1325 | * |
1326 | * This function determines whether the evicted inode is eligible to be deleted | |
1327 | * and locks the inode glock. | |
1328 | * | |
1329 | * Returns: the fate of the dinode | |
1330 | */ | |
1331 | static enum dinode_demise evict_should_delete(struct inode *inode, | |
1332 | struct gfs2_holder *gh) | |
1333 | { | |
1334 | struct gfs2_inode *ip = GFS2_I(inode); | |
1335 | struct super_block *sb = inode->i_sb; | |
1336 | struct gfs2_sbd *sdp = sb->s_fs_info; | |
1337 | int ret; | |
1338 | ||
38552ff6 | 1339 | if (unlikely(test_bit(GIF_ALLOC_FAILED, &ip->i_flags))) |
53dbc27e | 1340 | goto should_delete; |
53dbc27e BP |
1341 | |
1342 | if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags)) | |
1343 | return SHOULD_DEFER_EVICTION; | |
1344 | ||
1345 | /* Deletes should never happen under memory pressure anymore. */ | |
1346 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC)) | |
1347 | return SHOULD_DEFER_EVICTION; | |
1348 | ||
1349 | /* Must not read inode block until block type has been verified */ | |
1350 | ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh); | |
1351 | if (unlikely(ret)) { | |
1352 | glock_clear_object(ip->i_iopen_gh.gh_gl, ip); | |
1353 | ip->i_iopen_gh.gh_flags |= GL_NOCACHE; | |
1354 | gfs2_glock_dq_uninit(&ip->i_iopen_gh); | |
1355 | return SHOULD_DEFER_EVICTION; | |
1356 | } | |
1357 | ||
1358 | if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino)) | |
1359 | return SHOULD_NOT_DELETE_DINODE; | |
1360 | ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED); | |
1361 | if (ret) | |
1362 | return SHOULD_NOT_DELETE_DINODE; | |
1363 | ||
8d567162 AG |
1364 | ret = gfs2_instantiate(gh); |
1365 | if (ret) | |
1366 | return SHOULD_NOT_DELETE_DINODE; | |
53dbc27e BP |
1367 | |
1368 | /* | |
1369 | * The inode may have been recreated in the meantime. | |
1370 | */ | |
1371 | if (inode->i_nlink) | |
1372 | return SHOULD_NOT_DELETE_DINODE; | |
1373 | ||
1374 | should_delete: | |
1375 | if (gfs2_holder_initialized(&ip->i_iopen_gh) && | |
1376 | test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) { | |
1377 | if (!gfs2_upgrade_iopen_glock(inode)) { | |
1378 | gfs2_holder_uninit(&ip->i_iopen_gh); | |
1379 | return SHOULD_NOT_DELETE_DINODE; | |
1380 | } | |
1381 | } | |
1382 | return SHOULD_DELETE_DINODE; | |
1383 | } | |
1384 | ||
6e7e9a50 BP |
1385 | /** |
1386 | * evict_unlinked_inode - delete the pieces of an unlinked evicted inode | |
1387 | * @inode: The inode to evict | |
1388 | */ | |
1389 | static int evict_unlinked_inode(struct inode *inode) | |
1390 | { | |
1391 | struct gfs2_inode *ip = GFS2_I(inode); | |
1392 | int ret; | |
1393 | ||
1394 | if (S_ISDIR(inode->i_mode) && | |
1395 | (ip->i_diskflags & GFS2_DIF_EXHASH)) { | |
1396 | ret = gfs2_dir_exhash_dealloc(ip); | |
1397 | if (ret) | |
1398 | goto out; | |
1399 | } | |
1400 | ||
1401 | if (ip->i_eattr) { | |
1402 | ret = gfs2_ea_dealloc(ip); | |
1403 | if (ret) | |
1404 | goto out; | |
1405 | } | |
1406 | ||
1407 | if (!gfs2_is_stuffed(ip)) { | |
1408 | ret = gfs2_file_dealloc(ip); | |
1409 | if (ret) | |
1410 | goto out; | |
1411 | } | |
1412 | ||
764665c6 AG |
1413 | /* |
1414 | * As soon as we clear the bitmap for the dinode, gfs2_create_inode() | |
1415 | * can get called to recreate it, or even gfs2_inode_lookup() if the | |
1416 | * inode was recreated on another node in the meantime. | |
1417 | * | |
1418 | * However, inserting the new inode into the inode hash table will not | |
1419 | * succeed until the old inode is removed, and that only happens after | |
1420 | * ->evict_inode() returns. The new inode is attached to its inode and | |
1421 | * iopen glocks after inserting it into the inode hash table, so at | |
1422 | * that point we can be sure that both glocks are unused. | |
1423 | */ | |
1424 | ||
6e7e9a50 | 1425 | ret = gfs2_dinode_dealloc(ip); |
f9da18cd BP |
1426 | if (!ret && ip->i_gl) |
1427 | gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino); | |
1428 | ||
6e7e9a50 BP |
1429 | out: |
1430 | return ret; | |
1431 | } | |
1432 | ||
d90be6ab BP |
1433 | /* |
1434 | * evict_linked_inode - evict an inode whose dinode has not been unlinked | |
1435 | * @inode: The inode to evict | |
1436 | */ | |
1437 | static int evict_linked_inode(struct inode *inode) | |
1438 | { | |
1439 | struct super_block *sb = inode->i_sb; | |
1440 | struct gfs2_sbd *sdp = sb->s_fs_info; | |
1441 | struct gfs2_inode *ip = GFS2_I(inode); | |
1442 | struct address_space *metamapping; | |
1443 | int ret; | |
1444 | ||
1445 | gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL | | |
1446 | GFS2_LFC_EVICT_INODE); | |
1447 | metamapping = gfs2_glock2aspace(ip->i_gl); | |
1448 | if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) { | |
1449 | filemap_fdatawrite(metamapping); | |
1450 | filemap_fdatawait(metamapping); | |
1451 | } | |
1452 | write_inode_now(inode, 1); | |
1453 | gfs2_ail_flush(ip->i_gl, 0); | |
1454 | ||
1455 | ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks); | |
1456 | if (ret) | |
1457 | return ret; | |
1458 | ||
1459 | /* Needs to be done before glock release & also in a transaction */ | |
1460 | truncate_inode_pages(&inode->i_data, 0); | |
1461 | truncate_inode_pages(metamapping, 0); | |
1462 | gfs2_trans_end(sdp); | |
1463 | return 0; | |
1464 | } | |
1465 | ||
380f7c65 SW |
1466 | /** |
1467 | * gfs2_evict_inode - Remove an inode from cache | |
1468 | * @inode: The inode to evict | |
1469 | * | |
1470 | * There are three cases to consider: | |
1471 | * 1. i_nlink == 0, we are final opener (and must deallocate) | |
1472 | * 2. i_nlink == 0, we are not the final opener (and cannot deallocate) | |
1473 | * 3. i_nlink > 0 | |
1474 | * | |
1475 | * If the fs is read only, then we have to treat all cases as per #3 | |
1476 | * since we are unable to do any deallocation. The inode will be | |
1477 | * deallocated by the next read/write node to attempt an allocation | |
1478 | * in the same resource group | |
1479 | * | |
9e6e0a12 SW |
1480 | * We have to (at the moment) hold the inodes main lock to cover |
1481 | * the gap between unlocking the shared lock on the iopen lock and | |
1482 | * taking the exclusive lock. I'd rather do a shared -> exclusive | |
1483 | * conversion on the iopen lock, but we can change that later. This | |
1484 | * is safe, just less efficient. | |
1485 | */ | |
1486 | ||
d5c1515c | 1487 | static void gfs2_evict_inode(struct inode *inode) |
9e6e0a12 | 1488 | { |
001e8e8d SW |
1489 | struct super_block *sb = inode->i_sb; |
1490 | struct gfs2_sbd *sdp = sb->s_fs_info; | |
9e6e0a12 SW |
1491 | struct gfs2_inode *ip = GFS2_I(inode); |
1492 | struct gfs2_holder gh; | |
23d828fc | 1493 | int ret; |
9e6e0a12 | 1494 | |
38552ff6 | 1495 | if (inode->i_nlink || sb_rdonly(sb) || !ip->i_no_addr) |
d5c1515c AV |
1496 | goto out; |
1497 | ||
504a10d9 BP |
1498 | /* |
1499 | * In case of an incomplete mount, gfs2_evict_inode() may be called for | |
1500 | * system files without having an active journal to write to. In that | |
1501 | * case, skip the filesystem evict. | |
1502 | */ | |
1503 | if (!sdp->sd_jdesc) | |
1504 | goto out; | |
1505 | ||
53dbc27e BP |
1506 | gfs2_holder_mark_uninitialized(&gh); |
1507 | ret = evict_should_delete(inode, &gh); | |
1508 | if (ret == SHOULD_DEFER_EVICTION) | |
8c7b9262 | 1509 | goto out; |
0a0d9f55 BP |
1510 | if (ret == SHOULD_DELETE_DINODE) |
1511 | ret = evict_unlinked_inode(inode); | |
1512 | else | |
1513 | ret = evict_linked_inode(inode); | |
9e6e0a12 | 1514 | |
a097dc7e BP |
1515 | if (gfs2_rs_active(&ip->i_res)) |
1516 | gfs2_rs_deltree(&ip->i_res); | |
8e2e0047 | 1517 | |
9ffa1888 | 1518 | if (gfs2_holder_initialized(&gh)) |
e0b62e21 | 1519 | gfs2_glock_dq_uninit(&gh); |
23d828fc BP |
1520 | if (ret && ret != GLR_TRYFAILED && ret != -EROFS) |
1521 | fs_warn(sdp, "gfs2_evict_inode: %d\n", ret); | |
9e6e0a12 | 1522 | out: |
91b0abe3 | 1523 | truncate_inode_pages_final(&inode->i_data); |
2fba46a0 BP |
1524 | if (ip->i_qadata) |
1525 | gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0); | |
7336905a | 1526 | gfs2_rs_deltree(&ip->i_res); |
45138990 | 1527 | gfs2_ordered_del_inode(ip); |
dbd5768f | 1528 | clear_inode(inode); |
17d539f0 | 1529 | gfs2_dir_hash_inval(ip); |
6df9f9a2 | 1530 | if (gfs2_holder_initialized(&ip->i_iopen_gh)) { |
71c1b213 AG |
1531 | struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl; |
1532 | ||
1533 | glock_clear_object(gl, ip); | |
71c1b213 | 1534 | gfs2_glock_hold(gl); |
fe1bff65 AG |
1535 | ip->i_iopen_gh.gh_flags |= GL_NOCACHE; |
1536 | gfs2_glock_dq_uninit(&ip->i_iopen_gh); | |
71c1b213 | 1537 | gfs2_glock_put_eventually(gl); |
d5c1515c | 1538 | } |
49462e2b BP |
1539 | if (ip->i_gl) { |
1540 | glock_clear_object(ip->i_gl, ip); | |
1541 | wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE); | |
1542 | gfs2_glock_add_to_lru(ip->i_gl); | |
1543 | gfs2_glock_put_eventually(ip->i_gl); | |
0abd1557 | 1544 | rcu_assign_pointer(ip->i_gl, NULL); |
49462e2b | 1545 | } |
9e6e0a12 SW |
1546 | } |
1547 | ||
1548 | static struct inode *gfs2_alloc_inode(struct super_block *sb) | |
1549 | { | |
1550 | struct gfs2_inode *ip; | |
1551 | ||
fd60b288 | 1552 | ip = alloc_inode_sb(sb, gfs2_inode_cachep, GFP_KERNEL); |
d4031259 AG |
1553 | if (!ip) |
1554 | return NULL; | |
38552ff6 | 1555 | ip->i_no_addr = 0; |
d4031259 AG |
1556 | ip->i_flags = 0; |
1557 | ip->i_gl = NULL; | |
40e7e86e | 1558 | gfs2_holder_mark_uninitialized(&ip->i_iopen_gh); |
d4031259 AG |
1559 | memset(&ip->i_res, 0, sizeof(ip->i_res)); |
1560 | RB_CLEAR_NODE(&ip->i_res.rs_node); | |
1561 | ip->i_rahead = 0; | |
9e6e0a12 SW |
1562 | return &ip->i_inode; |
1563 | } | |
1564 | ||
784494e1 | 1565 | static void gfs2_free_inode(struct inode *inode) |
9e6e0a12 | 1566 | { |
784494e1 | 1567 | kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode)); |
fa0d7e3d NP |
1568 | } |
1569 | ||
0b2355fe | 1570 | void free_local_statfs_inodes(struct gfs2_sbd *sdp) |
97fd734b AD |
1571 | { |
1572 | struct local_statfs_inode *lsi, *safe; | |
1573 | ||
1574 | /* Run through the statfs inodes list to iput and free memory */ | |
1575 | list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) { | |
1576 | if (lsi->si_jid == sdp->sd_jdesc->jd_jid) | |
1577 | sdp->sd_sc_inode = NULL; /* belongs to this node */ | |
1578 | if (lsi->si_sc_inode) | |
1579 | iput(lsi->si_sc_inode); | |
1580 | list_del(&lsi->si_list); | |
1581 | kfree(lsi); | |
1582 | } | |
1583 | } | |
1584 | ||
0b2355fe AG |
1585 | struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp, |
1586 | unsigned int index) | |
97fd734b AD |
1587 | { |
1588 | struct local_statfs_inode *lsi; | |
1589 | ||
1590 | /* Return the local (per node) statfs inode in the | |
1591 | * sdp->sd_sc_inodes_list corresponding to the 'index'. */ | |
1592 | list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) { | |
1593 | if (lsi->si_jid == index) | |
1594 | return lsi->si_sc_inode; | |
1595 | } | |
1596 | return NULL; | |
1597 | } | |
1598 | ||
9e6e0a12 SW |
1599 | const struct super_operations gfs2_super_ops = { |
1600 | .alloc_inode = gfs2_alloc_inode, | |
784494e1 | 1601 | .free_inode = gfs2_free_inode, |
9e6e0a12 | 1602 | .write_inode = gfs2_write_inode, |
ab9bbda0 | 1603 | .dirty_inode = gfs2_dirty_inode, |
d5c1515c | 1604 | .evict_inode = gfs2_evict_inode, |
9e6e0a12 | 1605 | .put_super = gfs2_put_super, |
9e6e0a12 | 1606 | .sync_fs = gfs2_sync_fs, |
097cca52 AG |
1607 | .freeze_super = gfs2_freeze_super, |
1608 | .thaw_super = gfs2_thaw_super, | |
9e6e0a12 | 1609 | .statfs = gfs2_statfs, |
9e6e0a12 SW |
1610 | .drop_inode = gfs2_drop_inode, |
1611 | .show_options = gfs2_show_options, | |
1612 | }; | |
1613 |