]>
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; |
b3b94faa DT |
335 | int error; |
336 | ||
b3b94faa DT |
337 | list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) { |
338 | lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL); | |
339 | if (!lfcc) { | |
340 | error = -ENOMEM; | |
341 | goto out; | |
342 | } | |
feaa7bba SW |
343 | ip = GFS2_I(jd->jd_inode); |
344 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh); | |
b3b94faa DT |
345 | if (error) { |
346 | kfree(lfcc); | |
347 | goto out; | |
348 | } | |
349 | list_add(&lfcc->list, &list); | |
350 | } | |
351 | ||
24972557 | 352 | error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE, |
ebdc416c AG |
353 | LM_FLAG_NOEXP | GL_NOPID, |
354 | &sdp->sd_freeze_gh); | |
52b1cdcb BP |
355 | if (error) |
356 | goto out; | |
b3b94faa DT |
357 | |
358 | list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) { | |
359 | error = gfs2_jdesc_check(jd); | |
360 | if (error) | |
361 | break; | |
f4686c26 | 362 | error = gfs2_find_jhead(jd, &lh, false); |
b3b94faa DT |
363 | if (error) |
364 | break; | |
365 | if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) { | |
366 | error = -EBUSY; | |
367 | break; | |
368 | } | |
369 | } | |
370 | ||
371 | if (error) | |
c77b52c0 | 372 | gfs2_freeze_unlock(&sdp->sd_freeze_gh); |
b3b94faa | 373 | |
a91ea69f | 374 | out: |
b3b94faa | 375 | while (!list_empty(&list)) { |
969183bc | 376 | lfcc = list_first_entry(&list, struct lfcc, list); |
b3b94faa DT |
377 | list_del(&lfcc->list); |
378 | gfs2_glock_dq_uninit(&lfcc->gh); | |
379 | kfree(lfcc); | |
380 | } | |
b3b94faa DT |
381 | return error; |
382 | } | |
383 | ||
9eed04cd SW |
384 | void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf) |
385 | { | |
7db35444 | 386 | const struct inode *inode = &ip->i_inode; |
9eed04cd SW |
387 | struct gfs2_dinode *str = buf; |
388 | ||
389 | str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC); | |
390 | str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI); | |
391 | str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI); | |
392 | str->di_num.no_addr = cpu_to_be64(ip->i_no_addr); | |
393 | str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino); | |
7db35444 AG |
394 | str->di_mode = cpu_to_be32(inode->i_mode); |
395 | str->di_uid = cpu_to_be32(i_uid_read(inode)); | |
396 | str->di_gid = cpu_to_be32(i_gid_read(inode)); | |
397 | str->di_nlink = cpu_to_be32(inode->i_nlink); | |
398 | str->di_size = cpu_to_be64(i_size_read(inode)); | |
399 | str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(inode)); | |
400 | str->di_atime = cpu_to_be64(inode->i_atime.tv_sec); | |
401 | str->di_mtime = cpu_to_be64(inode->i_mtime.tv_sec); | |
402 | str->di_ctime = cpu_to_be64(inode->i_ctime.tv_sec); | |
9eed04cd SW |
403 | |
404 | str->di_goal_meta = cpu_to_be64(ip->i_goal); | |
405 | str->di_goal_data = cpu_to_be64(ip->i_goal); | |
406 | str->di_generation = cpu_to_be64(ip->i_generation); | |
407 | ||
408 | str->di_flags = cpu_to_be32(ip->i_diskflags); | |
409 | str->di_height = cpu_to_be16(ip->i_height); | |
7db35444 | 410 | str->di_payload_format = cpu_to_be32(S_ISDIR(inode->i_mode) && |
9eed04cd SW |
411 | !(ip->i_diskflags & GFS2_DIF_EXHASH) ? |
412 | GFS2_FORMAT_DE : 0); | |
413 | str->di_depth = cpu_to_be16(ip->i_depth); | |
414 | str->di_entries = cpu_to_be32(ip->i_entries); | |
415 | ||
416 | str->di_eattr = cpu_to_be64(ip->i_eattr); | |
7db35444 AG |
417 | str->di_atime_nsec = cpu_to_be32(inode->i_atime.tv_nsec); |
418 | str->di_mtime_nsec = cpu_to_be32(inode->i_mtime.tv_nsec); | |
419 | str->di_ctime_nsec = cpu_to_be32(inode->i_ctime.tv_nsec); | |
9eed04cd | 420 | } |
9e6e0a12 SW |
421 | |
422 | /** | |
423 | * gfs2_write_inode - Make sure the inode is stable on the disk | |
424 | * @inode: The inode | |
1027efaa | 425 | * @wbc: The writeback control structure |
9e6e0a12 SW |
426 | * |
427 | * Returns: errno | |
428 | */ | |
429 | ||
a9185b41 | 430 | static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc) |
9e6e0a12 SW |
431 | { |
432 | struct gfs2_inode *ip = GFS2_I(inode); | |
433 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1027efaa | 434 | struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl); |
de1414a6 | 435 | struct backing_dev_info *bdi = inode_to_bdi(metamapping->host); |
ab9bbda0 | 436 | int ret = 0; |
adbc3ddf | 437 | bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip)); |
ab9bbda0 | 438 | |
adbc3ddf | 439 | if (flush_all) |
c1696fb8 | 440 | gfs2_log_flush(GFS2_SB(inode), ip->i_gl, |
805c0907 BP |
441 | GFS2_LOG_HEAD_FLUSH_NORMAL | |
442 | GFS2_LFC_WRITE_INODE); | |
a88a341a | 443 | if (bdi->wb.dirty_exceeded) |
4667a0ec | 444 | gfs2_ail1_flush(sdp, wbc); |
1d4ec642 SW |
445 | else |
446 | filemap_fdatawrite(metamapping); | |
adbc3ddf | 447 | if (flush_all) |
1027efaa SW |
448 | ret = filemap_fdatawait(metamapping); |
449 | if (ret) | |
450 | mark_inode_dirty_sync(inode); | |
957a7acd AD |
451 | else { |
452 | spin_lock(&inode->i_lock); | |
453 | if (!(inode->i_flags & I_DIRTY)) | |
454 | gfs2_ordered_del_inode(ip); | |
455 | spin_unlock(&inode->i_lock); | |
456 | } | |
9e6e0a12 SW |
457 | return ret; |
458 | } | |
459 | ||
ab9bbda0 SW |
460 | /** |
461 | * gfs2_dirty_inode - check for atime updates | |
462 | * @inode: The inode in question | |
463 | * @flags: The type of dirty | |
464 | * | |
465 | * Unfortunately it can be called under any combination of inode | |
af1abe11 | 466 | * glock and freeze glock, so we have to check carefully. |
ab9bbda0 SW |
467 | * |
468 | * At the moment this deals only with atime - it should be possible | |
469 | * to expand that role in future, once a review of the locking has | |
470 | * been carried out. | |
471 | */ | |
472 | ||
473 | static void gfs2_dirty_inode(struct inode *inode, int flags) | |
474 | { | |
475 | struct gfs2_inode *ip = GFS2_I(inode); | |
476 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
477 | struct buffer_head *bh; | |
478 | struct gfs2_holder gh; | |
479 | int need_unlock = 0; | |
480 | int need_endtrans = 0; | |
481 | int ret; | |
482 | ||
38552ff6 AG |
483 | if (unlikely(!ip->i_gl)) { |
484 | /* This can only happen during incomplete inode creation. */ | |
485 | BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags)); | |
486 | return; | |
487 | } | |
488 | ||
eb43e660 | 489 | if (unlikely(gfs2_withdrawn(sdp))) |
0d1c7ae9 | 490 | return; |
ab9bbda0 SW |
491 | if (!gfs2_glock_is_locked_by_me(ip->i_gl)) { |
492 | ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); | |
493 | if (ret) { | |
494 | fs_err(sdp, "dirty_inode: glock %d\n", ret); | |
e28c02b9 | 495 | gfs2_dump_glock(NULL, ip->i_gl, true); |
ab9bbda0 SW |
496 | return; |
497 | } | |
498 | need_unlock = 1; | |
3d162688 BM |
499 | } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE)) |
500 | return; | |
ab9bbda0 SW |
501 | |
502 | if (current->journal_info == NULL) { | |
503 | ret = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
504 | if (ret) { | |
505 | fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret); | |
506 | goto out; | |
507 | } | |
508 | need_endtrans = 1; | |
509 | } | |
510 | ||
511 | ret = gfs2_meta_inode_buffer(ip, &bh); | |
512 | if (ret == 0) { | |
350a9b0a | 513 | gfs2_trans_add_meta(ip->i_gl, bh); |
ab9bbda0 SW |
514 | gfs2_dinode_out(ip, bh->b_data); |
515 | brelse(bh); | |
516 | } | |
517 | ||
518 | if (need_endtrans) | |
519 | gfs2_trans_end(sdp); | |
520 | out: | |
521 | if (need_unlock) | |
522 | gfs2_glock_dq_uninit(&gh); | |
523 | } | |
524 | ||
9e6e0a12 SW |
525 | /** |
526 | * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one | |
527 | * @sdp: the filesystem | |
528 | * | |
529 | * Returns: errno | |
530 | */ | |
531 | ||
eb602521 | 532 | void gfs2_make_fs_ro(struct gfs2_sbd *sdp) |
9e6e0a12 | 533 | { |
601ef0d5 BP |
534 | int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags); |
535 | ||
6b388abc AG |
536 | if (!test_bit(SDF_DEACTIVATING, &sdp->sd_flags)) |
537 | gfs2_flush_delete_work(sdp); | |
538 | ||
601ef0d5 BP |
539 | if (!log_write_allowed && current == sdp->sd_quotad_process) |
540 | fs_warn(sdp, "The quotad daemon is withdrawing.\n"); | |
541 | else if (sdp->sd_quotad_process) | |
5b3a9f34 BP |
542 | kthread_stop(sdp->sd_quotad_process); |
543 | sdp->sd_quotad_process = NULL; | |
601ef0d5 BP |
544 | |
545 | if (!log_write_allowed && current == sdp->sd_logd_process) | |
546 | fs_warn(sdp, "The logd daemon is withdrawing.\n"); | |
547 | else if (sdp->sd_logd_process) | |
5b3a9f34 BP |
548 | kthread_stop(sdp->sd_logd_process); |
549 | sdp->sd_logd_process = NULL; | |
8ad151c2 | 550 | |
601ef0d5 BP |
551 | if (log_write_allowed) { |
552 | gfs2_quota_sync(sdp->sd_vfs, 0); | |
553 | gfs2_statfs_sync(sdp->sd_vfs, 0); | |
9e6e0a12 | 554 | |
68ca088d BP |
555 | /* We do two log flushes here. The first one commits dirty inodes |
556 | * and rgrps to the journal, but queues up revokes to the ail list. | |
557 | * The second flush writes out and removes the revokes. | |
558 | * | |
559 | * The first must be done before the FLUSH_SHUTDOWN code | |
560 | * clears the LIVE flag, otherwise it will not be able to start | |
561 | * a transaction to write its revokes, and the error will cause | |
562 | * a withdraw of the file system. */ | |
563 | gfs2_log_flush(sdp, NULL, GFS2_LFC_MAKE_FS_RO); | |
601ef0d5 BP |
564 | gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN | |
565 | GFS2_LFC_MAKE_FS_RO); | |
f3708fb5 AG |
566 | wait_event_timeout(sdp->sd_log_waitq, |
567 | gfs2_log_is_empty(sdp), | |
568 | HZ * 5); | |
569 | gfs2_assert_warn(sdp, gfs2_log_is_empty(sdp)); | |
601ef0d5 | 570 | } else { |
f3708fb5 AG |
571 | wait_event_timeout(sdp->sd_log_waitq, |
572 | gfs2_log_is_empty(sdp), | |
601ef0d5 BP |
573 | HZ * 5); |
574 | } | |
9e6e0a12 SW |
575 | gfs2_quota_cleanup(sdp); |
576 | ||
601ef0d5 BP |
577 | if (!log_write_allowed) |
578 | sdp->sd_vfs->s_flags |= SB_RDONLY; | |
9e6e0a12 SW |
579 | } |
580 | ||
9e6e0a12 SW |
581 | /** |
582 | * gfs2_put_super - Unmount the filesystem | |
583 | * @sb: The VFS superblock | |
584 | * | |
585 | */ | |
586 | ||
587 | static void gfs2_put_super(struct super_block *sb) | |
588 | { | |
589 | struct gfs2_sbd *sdp = sb->s_fs_info; | |
9e6e0a12 SW |
590 | struct gfs2_jdesc *jd; |
591 | ||
9e6e0a12 SW |
592 | /* No more recovery requests */ |
593 | set_bit(SDF_NORECOVERY, &sdp->sd_flags); | |
594 | smp_mb(); | |
595 | ||
596 | /* Wait on outstanding recovery */ | |
597 | restart: | |
598 | spin_lock(&sdp->sd_jindex_spin); | |
599 | list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) { | |
600 | if (!test_bit(JDF_RECOVERY, &jd->jd_flags)) | |
601 | continue; | |
602 | spin_unlock(&sdp->sd_jindex_spin); | |
603 | wait_on_bit(&jd->jd_flags, JDF_RECOVERY, | |
74316201 | 604 | TASK_UNINTERRUPTIBLE); |
9e6e0a12 SW |
605 | goto restart; |
606 | } | |
607 | spin_unlock(&sdp->sd_jindex_spin); | |
608 | ||
bc98a42c | 609 | if (!sb_rdonly(sb)) { |
eb602521 | 610 | gfs2_make_fs_ro(sdp); |
9e6e0a12 | 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 | ||
618 | iput(sdp->sd_jindex); | |
9e6e0a12 SW |
619 | iput(sdp->sd_statfs_inode); |
620 | iput(sdp->sd_rindex); | |
621 | iput(sdp->sd_quota_inode); | |
622 | ||
623 | gfs2_glock_put(sdp->sd_rename_gl); | |
24972557 | 624 | gfs2_glock_put(sdp->sd_freeze_gl); |
9e6e0a12 SW |
625 | |
626 | if (!sdp->sd_args.ar_spectator) { | |
601ef0d5 BP |
627 | if (gfs2_holder_initialized(&sdp->sd_journal_gh)) |
628 | gfs2_glock_dq_uninit(&sdp->sd_journal_gh); | |
629 | if (gfs2_holder_initialized(&sdp->sd_jinode_gh)) | |
630 | gfs2_glock_dq_uninit(&sdp->sd_jinode_gh); | |
70c11ba8 | 631 | brelse(sdp->sd_sc_bh); |
9e6e0a12 SW |
632 | gfs2_glock_dq_uninit(&sdp->sd_sc_gh); |
633 | gfs2_glock_dq_uninit(&sdp->sd_qc_gh); | |
97fd734b | 634 | free_local_statfs_inodes(sdp); |
9e6e0a12 SW |
635 | iput(sdp->sd_qc_inode); |
636 | } | |
637 | ||
638 | gfs2_glock_dq_uninit(&sdp->sd_live_gh); | |
639 | gfs2_clear_rgrpd(sdp); | |
640 | gfs2_jindex_free(sdp); | |
641 | /* Take apart glock structures and buffer lists */ | |
642 | gfs2_gl_hash_clear(sdp); | |
a9dd945c | 643 | truncate_inode_pages_final(&sdp->sd_aspace); |
b2fb7dab | 644 | gfs2_delete_debugfs_file(sdp); |
9e6e0a12 SW |
645 | /* Unmount the locking protocol */ |
646 | gfs2_lm_unmount(sdp); | |
647 | ||
648 | /* At this point, we're through participating in the lockspace */ | |
649 | gfs2_sys_fs_del(sdp); | |
c2a04b02 | 650 | free_sbd(sdp); |
9e6e0a12 SW |
651 | } |
652 | ||
9e6e0a12 SW |
653 | /** |
654 | * gfs2_sync_fs - sync the filesystem | |
655 | * @sb: the superblock | |
c551f66c | 656 | * @wait: true to wait for completion |
9e6e0a12 SW |
657 | * |
658 | * Flushes the log to disk. | |
659 | */ | |
660 | ||
661 | static int gfs2_sync_fs(struct super_block *sb, int wait) | |
662 | { | |
1027efaa | 663 | struct gfs2_sbd *sdp = sb->s_fs_info; |
a1177825 JK |
664 | |
665 | gfs2_quota_sync(sb, -1); | |
942b0cdd | 666 | if (wait) |
805c0907 BP |
667 | gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL | |
668 | GFS2_LFC_SYNC_FS); | |
942b0cdd | 669 | return sdp->sd_log_error; |
9e6e0a12 SW |
670 | } |
671 | ||
2e60d768 BM |
672 | void gfs2_freeze_func(struct work_struct *work) |
673 | { | |
674 | int error; | |
675 | struct gfs2_holder freeze_gh; | |
676 | struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work); | |
677 | struct super_block *sb = sdp->sd_vfs; | |
678 | ||
679 | atomic_inc(&sb->s_active); | |
c77b52c0 | 680 | error = gfs2_freeze_lock(sdp, &freeze_gh, 0); |
2e60d768 | 681 | if (error) { |
2e60d768 | 682 | gfs2_assert_withdraw(sdp, 0); |
8f918219 | 683 | } else { |
2e60d768 BM |
684 | atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN); |
685 | error = thaw_super(sb); | |
686 | if (error) { | |
f29e62ee BP |
687 | fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n", |
688 | error); | |
2e60d768 BM |
689 | gfs2_assert_withdraw(sdp, 0); |
690 | } | |
c77b52c0 | 691 | gfs2_freeze_unlock(&freeze_gh); |
2e60d768 BM |
692 | } |
693 | deactivate_super(sb); | |
8f918219 AD |
694 | clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags); |
695 | wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN); | |
2e60d768 BM |
696 | return; |
697 | } | |
698 | ||
9e6e0a12 SW |
699 | /** |
700 | * gfs2_freeze - prevent further writes to the filesystem | |
701 | * @sb: the VFS structure for the filesystem | |
702 | * | |
703 | */ | |
704 | ||
705 | static int gfs2_freeze(struct super_block *sb) | |
706 | { | |
707 | struct gfs2_sbd *sdp = sb->s_fs_info; | |
ff132c5f | 708 | int error; |
9e6e0a12 | 709 | |
2e60d768 | 710 | mutex_lock(&sdp->sd_freeze_mutex); |
ff132c5f BP |
711 | if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN) { |
712 | error = -EBUSY; | |
2e60d768 | 713 | goto out; |
ff132c5f | 714 | } |
2e60d768 | 715 | |
9e6e0a12 | 716 | for (;;) { |
60528afa BP |
717 | if (gfs2_withdrawn(sdp)) { |
718 | error = -EINVAL; | |
719 | goto out; | |
720 | } | |
721 | ||
52b1cdcb | 722 | error = gfs2_lock_fs_check_clean(sdp); |
9e6e0a12 SW |
723 | if (!error) |
724 | break; | |
725 | ||
55317f5b | 726 | if (error == -EBUSY) |
9e6e0a12 | 727 | fs_err(sdp, "waiting for recovery before freeze\n"); |
52b1cdcb BP |
728 | else if (error == -EIO) { |
729 | fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due " | |
730 | "to recovery error.\n"); | |
731 | goto out; | |
732 | } else { | |
9e6e0a12 | 733 | fs_err(sdp, "error freezing FS: %d\n", error); |
52b1cdcb | 734 | } |
9e6e0a12 SW |
735 | fs_err(sdp, "retrying...\n"); |
736 | msleep(1000); | |
737 | } | |
8f918219 | 738 | set_bit(SDF_FS_FROZEN, &sdp->sd_flags); |
2e60d768 BM |
739 | out: |
740 | mutex_unlock(&sdp->sd_freeze_mutex); | |
741 | return error; | |
9e6e0a12 SW |
742 | } |
743 | ||
744 | /** | |
745 | * gfs2_unfreeze - reallow writes to the filesystem | |
746 | * @sb: the VFS structure for the filesystem | |
747 | * | |
748 | */ | |
749 | ||
750 | static int gfs2_unfreeze(struct super_block *sb) | |
751 | { | |
d564053f SW |
752 | struct gfs2_sbd *sdp = sb->s_fs_info; |
753 | ||
2e60d768 | 754 | mutex_lock(&sdp->sd_freeze_mutex); |
ff132c5f | 755 | if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN || |
6df9f9a2 | 756 | !gfs2_holder_initialized(&sdp->sd_freeze_gh)) { |
2e60d768 | 757 | mutex_unlock(&sdp->sd_freeze_mutex); |
ff132c5f | 758 | return -EINVAL; |
2e60d768 BM |
759 | } |
760 | ||
c77b52c0 | 761 | gfs2_freeze_unlock(&sdp->sd_freeze_gh); |
2e60d768 | 762 | mutex_unlock(&sdp->sd_freeze_mutex); |
8f918219 | 763 | return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE); |
9e6e0a12 SW |
764 | } |
765 | ||
766 | /** | |
c551f66c | 767 | * statfs_slow_fill - fill in the sg for a given RG |
9e6e0a12 SW |
768 | * @rgd: the RG |
769 | * @sc: the sc structure | |
770 | * | |
771 | * Returns: 0 on success, -ESTALE if the LVB is invalid | |
772 | */ | |
773 | ||
774 | static int statfs_slow_fill(struct gfs2_rgrpd *rgd, | |
775 | struct gfs2_statfs_change_host *sc) | |
776 | { | |
777 | gfs2_rgrp_verify(rgd); | |
778 | sc->sc_total += rgd->rd_data; | |
779 | sc->sc_free += rgd->rd_free; | |
780 | sc->sc_dinodes += rgd->rd_dinodes; | |
781 | return 0; | |
782 | } | |
783 | ||
784 | /** | |
785 | * gfs2_statfs_slow - Stat a filesystem using asynchronous locking | |
786 | * @sdp: the filesystem | |
787 | * @sc: the sc info that will be returned | |
788 | * | |
789 | * Any error (other than a signal) will cause this routine to fall back | |
790 | * to the synchronous version. | |
791 | * | |
792 | * FIXME: This really shouldn't busy wait like this. | |
793 | * | |
794 | * Returns: errno | |
795 | */ | |
796 | ||
797 | static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc) | |
798 | { | |
9e6e0a12 SW |
799 | struct gfs2_rgrpd *rgd_next; |
800 | struct gfs2_holder *gha, *gh; | |
801 | unsigned int slots = 64; | |
802 | unsigned int x; | |
803 | int done; | |
804 | int error = 0, err; | |
805 | ||
806 | memset(sc, 0, sizeof(struct gfs2_statfs_change_host)); | |
6da2ec56 | 807 | gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL); |
9e6e0a12 SW |
808 | if (!gha) |
809 | return -ENOMEM; | |
6df9f9a2 AG |
810 | for (x = 0; x < slots; x++) |
811 | gfs2_holder_mark_uninitialized(gha + x); | |
9e6e0a12 | 812 | |
9e6e0a12 SW |
813 | rgd_next = gfs2_rgrpd_get_first(sdp); |
814 | ||
815 | for (;;) { | |
816 | done = 1; | |
817 | ||
818 | for (x = 0; x < slots; x++) { | |
819 | gh = gha + x; | |
820 | ||
6df9f9a2 | 821 | if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) { |
9e6e0a12 SW |
822 | err = gfs2_glock_wait(gh); |
823 | if (err) { | |
824 | gfs2_holder_uninit(gh); | |
825 | error = err; | |
826 | } else { | |
6f6597ba AG |
827 | if (!error) { |
828 | struct gfs2_rgrpd *rgd = | |
829 | gfs2_glock2rgrp(gh->gh_gl); | |
830 | ||
831 | error = statfs_slow_fill(rgd, sc); | |
832 | } | |
9e6e0a12 SW |
833 | gfs2_glock_dq_uninit(gh); |
834 | } | |
835 | } | |
836 | ||
6df9f9a2 | 837 | if (gfs2_holder_initialized(gh)) |
9e6e0a12 SW |
838 | done = 0; |
839 | else if (rgd_next && !error) { | |
840 | error = gfs2_glock_nq_init(rgd_next->rd_gl, | |
841 | LM_ST_SHARED, | |
842 | GL_ASYNC, | |
843 | gh); | |
844 | rgd_next = gfs2_rgrpd_get_next(rgd_next); | |
845 | done = 0; | |
846 | } | |
847 | ||
848 | if (signal_pending(current)) | |
849 | error = -ERESTARTSYS; | |
850 | } | |
851 | ||
852 | if (done) | |
853 | break; | |
854 | ||
855 | yield(); | |
856 | } | |
857 | ||
9e6e0a12 SW |
858 | kfree(gha); |
859 | return error; | |
860 | } | |
861 | ||
862 | /** | |
863 | * gfs2_statfs_i - Do a statfs | |
864 | * @sdp: the filesystem | |
c551f66c | 865 | * @sc: the sc structure |
9e6e0a12 SW |
866 | * |
867 | * Returns: errno | |
868 | */ | |
869 | ||
870 | static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc) | |
871 | { | |
872 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; | |
873 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; | |
874 | ||
875 | spin_lock(&sdp->sd_statfs_spin); | |
876 | ||
877 | *sc = *m_sc; | |
878 | sc->sc_total += l_sc->sc_total; | |
879 | sc->sc_free += l_sc->sc_free; | |
880 | sc->sc_dinodes += l_sc->sc_dinodes; | |
881 | ||
882 | spin_unlock(&sdp->sd_statfs_spin); | |
883 | ||
884 | if (sc->sc_free < 0) | |
885 | sc->sc_free = 0; | |
886 | if (sc->sc_free > sc->sc_total) | |
887 | sc->sc_free = sc->sc_total; | |
888 | if (sc->sc_dinodes < 0) | |
889 | sc->sc_dinodes = 0; | |
890 | ||
891 | return 0; | |
892 | } | |
893 | ||
894 | /** | |
895 | * gfs2_statfs - Gather and return stats about the filesystem | |
c551f66c LJ |
896 | * @dentry: The name of the link |
897 | * @buf: The buffer | |
9e6e0a12 SW |
898 | * |
899 | * Returns: 0 on success or error code | |
900 | */ | |
901 | ||
902 | static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf) | |
903 | { | |
fc64005c | 904 | struct super_block *sb = dentry->d_sb; |
9e6e0a12 SW |
905 | struct gfs2_sbd *sdp = sb->s_fs_info; |
906 | struct gfs2_statfs_change_host sc; | |
907 | int error; | |
908 | ||
8339ee54 SW |
909 | error = gfs2_rindex_update(sdp); |
910 | if (error) | |
911 | return error; | |
912 | ||
9e6e0a12 SW |
913 | if (gfs2_tune_get(sdp, gt_statfs_slow)) |
914 | error = gfs2_statfs_slow(sdp, &sc); | |
915 | else | |
916 | error = gfs2_statfs_i(sdp, &sc); | |
917 | ||
918 | if (error) | |
919 | return error; | |
920 | ||
921 | buf->f_type = GFS2_MAGIC; | |
922 | buf->f_bsize = sdp->sd_sb.sb_bsize; | |
923 | buf->f_blocks = sc.sc_total; | |
924 | buf->f_bfree = sc.sc_free; | |
925 | buf->f_bavail = sc.sc_free; | |
926 | buf->f_files = sc.sc_dinodes + sc.sc_free; | |
927 | buf->f_ffree = sc.sc_free; | |
928 | buf->f_namelen = GFS2_FNAMESIZE; | |
929 | ||
930 | return 0; | |
931 | } | |
932 | ||
9e6e0a12 SW |
933 | /** |
934 | * gfs2_drop_inode - Drop an inode (test for remote unlink) | |
935 | * @inode: The inode to drop | |
936 | * | |
61b91cfd | 937 | * If we've received a callback on an iopen lock then it's because a |
9e6e0a12 SW |
938 | * remote node tried to deallocate the inode but failed due to this node |
939 | * still having the inode open. Here we mark the link count zero | |
940 | * since we know that it must have reached zero if the GLF_DEMOTE flag | |
941 | * is set on the iopen glock. If we didn't do a disk read since the | |
942 | * remote node removed the final link then we might otherwise miss | |
943 | * this event. This check ensures that this node will deallocate the | |
944 | * inode's blocks, or alternatively pass the baton on to another | |
945 | * node for later deallocation. | |
946 | */ | |
947 | ||
45321ac5 | 948 | static int gfs2_drop_inode(struct inode *inode) |
9e6e0a12 SW |
949 | { |
950 | struct gfs2_inode *ip = GFS2_I(inode); | |
b88beb9a | 951 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
9e6e0a12 | 952 | |
38552ff6 | 953 | if (inode->i_nlink && |
6df9f9a2 | 954 | gfs2_holder_initialized(&ip->i_iopen_gh)) { |
9e6e0a12 | 955 | struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl; |
6df9f9a2 | 956 | if (test_bit(GLF_DEMOTE, &gl->gl_flags)) |
9e6e0a12 SW |
957 | clear_nlink(inode); |
958 | } | |
6a1c8f6d AG |
959 | |
960 | /* | |
961 | * When under memory pressure when an inode's link count has dropped to | |
962 | * zero, defer deleting the inode to the delete workqueue. This avoids | |
963 | * calling into DLM under memory pressure, which can deadlock. | |
964 | */ | |
965 | if (!inode->i_nlink && | |
966 | unlikely(current->flags & PF_MEMALLOC) && | |
967 | gfs2_holder_initialized(&ip->i_iopen_gh)) { | |
968 | struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl; | |
969 | ||
970 | gfs2_glock_hold(gl); | |
f0e56edc | 971 | if (!gfs2_queue_try_to_evict(gl)) |
6a1c8f6d | 972 | gfs2_glock_queue_put(gl); |
ba3ca2bc | 973 | return 0; |
6a1c8f6d AG |
974 | } |
975 | ||
b88beb9a AG |
976 | /* |
977 | * No longer cache inodes when trying to evict them all. | |
978 | */ | |
979 | if (test_bit(SDF_EVICTING, &sdp->sd_flags)) | |
980 | return 1; | |
981 | ||
45321ac5 | 982 | return generic_drop_inode(inode); |
9e6e0a12 SW |
983 | } |
984 | ||
985 | static int is_ancestor(const struct dentry *d1, const struct dentry *d2) | |
986 | { | |
987 | do { | |
988 | if (d1 == d2) | |
989 | return 1; | |
990 | d1 = d1->d_parent; | |
991 | } while (!IS_ROOT(d1)); | |
992 | return 0; | |
993 | } | |
994 | ||
995 | /** | |
996 | * gfs2_show_options - Show mount options for /proc/mounts | |
997 | * @s: seq_file structure | |
34c80b1d | 998 | * @root: root of this (sub)tree |
9e6e0a12 SW |
999 | * |
1000 | * Returns: 0 on success or error code | |
1001 | */ | |
1002 | ||
34c80b1d | 1003 | static int gfs2_show_options(struct seq_file *s, struct dentry *root) |
9e6e0a12 | 1004 | { |
34c80b1d | 1005 | struct gfs2_sbd *sdp = root->d_sb->s_fs_info; |
9e6e0a12 | 1006 | struct gfs2_args *args = &sdp->sd_args; |
6fa0a72c TL |
1007 | unsigned int logd_secs, statfs_slow, statfs_quantum, quota_quantum; |
1008 | ||
1009 | spin_lock(&sdp->sd_tune.gt_spin); | |
1010 | logd_secs = sdp->sd_tune.gt_logd_secs; | |
1011 | quota_quantum = sdp->sd_tune.gt_quota_quantum; | |
1012 | statfs_quantum = sdp->sd_tune.gt_statfs_quantum; | |
1013 | statfs_slow = sdp->sd_tune.gt_statfs_slow; | |
1014 | spin_unlock(&sdp->sd_tune.gt_spin); | |
9e6e0a12 | 1015 | |
34c80b1d | 1016 | if (is_ancestor(root, sdp->sd_master_dir)) |
eaebdedc | 1017 | seq_puts(s, ",meta"); |
9e6e0a12 | 1018 | if (args->ar_lockproto[0]) |
a068acf2 | 1019 | seq_show_option(s, "lockproto", args->ar_lockproto); |
9e6e0a12 | 1020 | if (args->ar_locktable[0]) |
a068acf2 | 1021 | seq_show_option(s, "locktable", args->ar_locktable); |
9e6e0a12 | 1022 | if (args->ar_hostdata[0]) |
a068acf2 | 1023 | seq_show_option(s, "hostdata", args->ar_hostdata); |
9e6e0a12 | 1024 | if (args->ar_spectator) |
eaebdedc | 1025 | seq_puts(s, ",spectator"); |
9e6e0a12 | 1026 | if (args->ar_localflocks) |
eaebdedc | 1027 | seq_puts(s, ",localflocks"); |
9e6e0a12 | 1028 | if (args->ar_debug) |
eaebdedc | 1029 | seq_puts(s, ",debug"); |
9e6e0a12 | 1030 | if (args->ar_posix_acl) |
eaebdedc | 1031 | seq_puts(s, ",acl"); |
9e6e0a12 SW |
1032 | if (args->ar_quota != GFS2_QUOTA_DEFAULT) { |
1033 | char *state; | |
1034 | switch (args->ar_quota) { | |
1035 | case GFS2_QUOTA_OFF: | |
1036 | state = "off"; | |
1037 | break; | |
1038 | case GFS2_QUOTA_ACCOUNT: | |
1039 | state = "account"; | |
1040 | break; | |
1041 | case GFS2_QUOTA_ON: | |
1042 | state = "on"; | |
1043 | break; | |
1044 | default: | |
1045 | state = "unknown"; | |
1046 | break; | |
1047 | } | |
1048 | seq_printf(s, ",quota=%s", state); | |
1049 | } | |
1050 | if (args->ar_suiddir) | |
eaebdedc | 1051 | seq_puts(s, ",suiddir"); |
9e6e0a12 SW |
1052 | if (args->ar_data != GFS2_DATA_DEFAULT) { |
1053 | char *state; | |
1054 | switch (args->ar_data) { | |
1055 | case GFS2_DATA_WRITEBACK: | |
1056 | state = "writeback"; | |
1057 | break; | |
1058 | case GFS2_DATA_ORDERED: | |
1059 | state = "ordered"; | |
1060 | break; | |
1061 | default: | |
1062 | state = "unknown"; | |
1063 | break; | |
1064 | } | |
1065 | seq_printf(s, ",data=%s", state); | |
1066 | } | |
1067 | if (args->ar_discard) | |
eaebdedc | 1068 | seq_puts(s, ",discard"); |
6fa0a72c TL |
1069 | if (logd_secs != 30) |
1070 | seq_printf(s, ",commit=%d", logd_secs); | |
1071 | if (statfs_quantum != 30) | |
1072 | seq_printf(s, ",statfs_quantum=%d", statfs_quantum); | |
1073 | else if (statfs_slow) | |
2b9731e8 | 1074 | seq_puts(s, ",statfs_quantum=0"); |
6fa0a72c TL |
1075 | if (quota_quantum != 60) |
1076 | seq_printf(s, ",quota_quantum=%d", quota_quantum); | |
3d3c10f2 BM |
1077 | if (args->ar_statfs_percent) |
1078 | seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent); | |
d34843d0 BP |
1079 | if (args->ar_errors != GFS2_ERRORS_DEFAULT) { |
1080 | const char *state; | |
1081 | ||
1082 | switch (args->ar_errors) { | |
1083 | case GFS2_ERRORS_WITHDRAW: | |
1084 | state = "withdraw"; | |
1085 | break; | |
1086 | case GFS2_ERRORS_PANIC: | |
1087 | state = "panic"; | |
1088 | break; | |
1089 | default: | |
1090 | state = "unknown"; | |
1091 | break; | |
1092 | } | |
1093 | seq_printf(s, ",errors=%s", state); | |
1094 | } | |
cdcfde62 | 1095 | if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) |
eaebdedc | 1096 | seq_puts(s, ",nobarrier"); |
913a71d2 | 1097 | if (test_bit(SDF_DEMOTE, &sdp->sd_flags)) |
eaebdedc | 1098 | seq_puts(s, ",demote_interface_used"); |
90306c41 | 1099 | if (args->ar_rgrplvb) |
eaebdedc | 1100 | seq_puts(s, ",rgrplvb"); |
471f3db2 BM |
1101 | if (args->ar_loccookie) |
1102 | seq_puts(s, ",loccookie"); | |
9e6e0a12 SW |
1103 | return 0; |
1104 | } | |
1105 | ||
f42ab085 SW |
1106 | static void gfs2_final_release_pages(struct gfs2_inode *ip) |
1107 | { | |
1108 | struct inode *inode = &ip->i_inode; | |
1109 | struct gfs2_glock *gl = ip->i_gl; | |
1110 | ||
38552ff6 AG |
1111 | if (unlikely(!gl)) { |
1112 | /* This can only happen during incomplete inode creation. */ | |
1113 | BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags)); | |
1114 | return; | |
1115 | } | |
1116 | ||
1117 | truncate_inode_pages(gfs2_glock2aspace(gl), 0); | |
f42ab085 SW |
1118 | truncate_inode_pages(&inode->i_data, 0); |
1119 | ||
638803d4 | 1120 | if (atomic_read(&gl->gl_revokes) == 0) { |
f42ab085 SW |
1121 | clear_bit(GLF_LFLUSH, &gl->gl_flags); |
1122 | clear_bit(GLF_DIRTY, &gl->gl_flags); | |
1123 | } | |
1124 | } | |
1125 | ||
1126 | static int gfs2_dinode_dealloc(struct gfs2_inode *ip) | |
1127 | { | |
1128 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
f42ab085 | 1129 | struct gfs2_rgrpd *rgd; |
564e12b1 | 1130 | struct gfs2_holder gh; |
f42ab085 SW |
1131 | int error; |
1132 | ||
1133 | if (gfs2_get_inode_blocks(&ip->i_inode) != 1) { | |
94fb763b | 1134 | gfs2_consist_inode(ip); |
f42ab085 SW |
1135 | return -EIO; |
1136 | } | |
1137 | ||
9b620429 | 1138 | gfs2_rindex_update(sdp); |
f42ab085 | 1139 | |
f4108a60 | 1140 | error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE); |
f42ab085 | 1141 | if (error) |
5407e242 | 1142 | return error; |
f42ab085 | 1143 | |
66fc061b | 1144 | rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1); |
f42ab085 SW |
1145 | if (!rgd) { |
1146 | gfs2_consist_inode(ip); | |
1147 | error = -EIO; | |
8339ee54 | 1148 | goto out_qs; |
f42ab085 SW |
1149 | } |
1150 | ||
4fc7ec31 BP |
1151 | error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, |
1152 | LM_FLAG_NODE_SCOPE, &gh); | |
f42ab085 | 1153 | if (error) |
8339ee54 | 1154 | goto out_qs; |
f42ab085 | 1155 | |
4667a0ec SW |
1156 | error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA, |
1157 | sdp->sd_jdesc->jd_blocks); | |
f42ab085 SW |
1158 | if (error) |
1159 | goto out_rg_gunlock; | |
1160 | ||
1161 | gfs2_free_di(rgd, ip); | |
1162 | ||
1163 | gfs2_final_release_pages(ip); | |
1164 | ||
1165 | gfs2_trans_end(sdp); | |
1166 | ||
1167 | out_rg_gunlock: | |
564e12b1 | 1168 | gfs2_glock_dq_uninit(&gh); |
f42ab085 SW |
1169 | out_qs: |
1170 | gfs2_quota_unhold(ip); | |
f42ab085 SW |
1171 | return error; |
1172 | } | |
1173 | ||
71c1b213 AG |
1174 | /** |
1175 | * gfs2_glock_put_eventually | |
1176 | * @gl: The glock to put | |
1177 | * | |
1178 | * When under memory pressure, trigger a deferred glock put to make sure we | |
1179 | * won't call into DLM and deadlock. Otherwise, put the glock directly. | |
1180 | */ | |
1181 | ||
1182 | static void gfs2_glock_put_eventually(struct gfs2_glock *gl) | |
1183 | { | |
1184 | if (current->flags & PF_MEMALLOC) | |
1185 | gfs2_glock_queue_put(gl); | |
1186 | else | |
1187 | gfs2_glock_put(gl); | |
1188 | } | |
1189 | ||
9e73330f AG |
1190 | static bool gfs2_upgrade_iopen_glock(struct inode *inode) |
1191 | { | |
1192 | struct gfs2_inode *ip = GFS2_I(inode); | |
1193 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1194 | struct gfs2_holder *gh = &ip->i_iopen_gh; | |
1195 | long timeout = 5 * HZ; | |
1196 | int error; | |
1197 | ||
1198 | gh->gh_flags |= GL_NOCACHE; | |
1199 | gfs2_glock_dq_wait(gh); | |
1200 | ||
1201 | /* | |
2d143955 AG |
1202 | * If there are no other lock holders, we will immediately get |
1203 | * exclusive access to the iopen glock here. | |
1204 | * | |
9e73330f | 1205 | * Otherwise, the other nodes holding the lock will be notified about |
2d143955 AG |
1206 | * our locking request. If they do not have the inode open, they are |
1207 | * expected to evict the cached inode and release the lock, allowing us | |
1208 | * to proceed. | |
1209 | * | |
1210 | * Otherwise, if they cannot evict the inode, they are expected to poke | |
1211 | * the inode glock (note: not the iopen glock). We will notice that | |
1212 | * and stop waiting for the iopen glock immediately. The other node(s) | |
1213 | * are then expected to take care of deleting the inode when they no | |
1214 | * longer use it. | |
1215 | * | |
1216 | * As a last resort, if another node keeps holding the iopen glock | |
1217 | * without showing any activity on the inode glock, we will eventually | |
1218 | * time out and fail the iopen glock upgrade. | |
9e73330f AG |
1219 | * |
1220 | * Note that we're passing the LM_FLAG_TRY_1CB flag to the first | |
1221 | * locking request as an optimization to notify lock holders as soon as | |
1222 | * possible. Without that flag, they'd be notified implicitly by the | |
1223 | * second locking request. | |
1224 | */ | |
1225 | ||
1226 | gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, gh); | |
1227 | error = gfs2_glock_nq(gh); | |
1228 | if (error != GLR_TRYFAILED) | |
1229 | return !error; | |
1230 | ||
1231 | gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh); | |
1232 | error = gfs2_glock_nq(gh); | |
1233 | if (error) | |
1234 | return false; | |
1235 | ||
1236 | timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait, | |
9e8990de AG |
1237 | !test_bit(HIF_WAIT, &gh->gh_iflags) || |
1238 | test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags), | |
9e73330f AG |
1239 | timeout); |
1240 | if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) { | |
1241 | gfs2_glock_dq(gh); | |
1242 | return false; | |
1243 | } | |
53d69132 | 1244 | return gfs2_glock_holder_ready(gh) == 0; |
9e73330f AG |
1245 | } |
1246 | ||
53dbc27e BP |
1247 | /** |
1248 | * evict_should_delete - determine whether the inode is eligible for deletion | |
1249 | * @inode: The inode to evict | |
c551f66c | 1250 | * @gh: The glock holder structure |
53dbc27e BP |
1251 | * |
1252 | * This function determines whether the evicted inode is eligible to be deleted | |
1253 | * and locks the inode glock. | |
1254 | * | |
1255 | * Returns: the fate of the dinode | |
1256 | */ | |
1257 | static enum dinode_demise evict_should_delete(struct inode *inode, | |
1258 | struct gfs2_holder *gh) | |
1259 | { | |
1260 | struct gfs2_inode *ip = GFS2_I(inode); | |
1261 | struct super_block *sb = inode->i_sb; | |
1262 | struct gfs2_sbd *sdp = sb->s_fs_info; | |
1263 | int ret; | |
1264 | ||
38552ff6 | 1265 | if (unlikely(test_bit(GIF_ALLOC_FAILED, &ip->i_flags))) |
53dbc27e | 1266 | goto should_delete; |
53dbc27e BP |
1267 | |
1268 | if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags)) | |
1269 | return SHOULD_DEFER_EVICTION; | |
1270 | ||
1271 | /* Deletes should never happen under memory pressure anymore. */ | |
1272 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC)) | |
1273 | return SHOULD_DEFER_EVICTION; | |
1274 | ||
1275 | /* Must not read inode block until block type has been verified */ | |
1276 | ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh); | |
1277 | if (unlikely(ret)) { | |
1278 | glock_clear_object(ip->i_iopen_gh.gh_gl, ip); | |
1279 | ip->i_iopen_gh.gh_flags |= GL_NOCACHE; | |
1280 | gfs2_glock_dq_uninit(&ip->i_iopen_gh); | |
1281 | return SHOULD_DEFER_EVICTION; | |
1282 | } | |
1283 | ||
1284 | if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino)) | |
1285 | return SHOULD_NOT_DELETE_DINODE; | |
1286 | ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED); | |
1287 | if (ret) | |
1288 | return SHOULD_NOT_DELETE_DINODE; | |
1289 | ||
8d567162 AG |
1290 | ret = gfs2_instantiate(gh); |
1291 | if (ret) | |
1292 | return SHOULD_NOT_DELETE_DINODE; | |
53dbc27e BP |
1293 | |
1294 | /* | |
1295 | * The inode may have been recreated in the meantime. | |
1296 | */ | |
1297 | if (inode->i_nlink) | |
1298 | return SHOULD_NOT_DELETE_DINODE; | |
1299 | ||
1300 | should_delete: | |
1301 | if (gfs2_holder_initialized(&ip->i_iopen_gh) && | |
1302 | test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) { | |
1303 | if (!gfs2_upgrade_iopen_glock(inode)) { | |
1304 | gfs2_holder_uninit(&ip->i_iopen_gh); | |
1305 | return SHOULD_NOT_DELETE_DINODE; | |
1306 | } | |
1307 | } | |
1308 | return SHOULD_DELETE_DINODE; | |
1309 | } | |
1310 | ||
6e7e9a50 BP |
1311 | /** |
1312 | * evict_unlinked_inode - delete the pieces of an unlinked evicted inode | |
1313 | * @inode: The inode to evict | |
1314 | */ | |
1315 | static int evict_unlinked_inode(struct inode *inode) | |
1316 | { | |
1317 | struct gfs2_inode *ip = GFS2_I(inode); | |
1318 | int ret; | |
1319 | ||
1320 | if (S_ISDIR(inode->i_mode) && | |
1321 | (ip->i_diskflags & GFS2_DIF_EXHASH)) { | |
1322 | ret = gfs2_dir_exhash_dealloc(ip); | |
1323 | if (ret) | |
1324 | goto out; | |
1325 | } | |
1326 | ||
1327 | if (ip->i_eattr) { | |
1328 | ret = gfs2_ea_dealloc(ip); | |
1329 | if (ret) | |
1330 | goto out; | |
1331 | } | |
1332 | ||
1333 | if (!gfs2_is_stuffed(ip)) { | |
1334 | ret = gfs2_file_dealloc(ip); | |
1335 | if (ret) | |
1336 | goto out; | |
1337 | } | |
1338 | ||
764665c6 AG |
1339 | /* |
1340 | * As soon as we clear the bitmap for the dinode, gfs2_create_inode() | |
1341 | * can get called to recreate it, or even gfs2_inode_lookup() if the | |
1342 | * inode was recreated on another node in the meantime. | |
1343 | * | |
1344 | * However, inserting the new inode into the inode hash table will not | |
1345 | * succeed until the old inode is removed, and that only happens after | |
1346 | * ->evict_inode() returns. The new inode is attached to its inode and | |
1347 | * iopen glocks after inserting it into the inode hash table, so at | |
1348 | * that point we can be sure that both glocks are unused. | |
1349 | */ | |
1350 | ||
6e7e9a50 | 1351 | ret = gfs2_dinode_dealloc(ip); |
f9da18cd BP |
1352 | if (!ret && ip->i_gl) |
1353 | gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino); | |
1354 | ||
6e7e9a50 BP |
1355 | out: |
1356 | return ret; | |
1357 | } | |
1358 | ||
d90be6ab BP |
1359 | /* |
1360 | * evict_linked_inode - evict an inode whose dinode has not been unlinked | |
1361 | * @inode: The inode to evict | |
1362 | */ | |
1363 | static int evict_linked_inode(struct inode *inode) | |
1364 | { | |
1365 | struct super_block *sb = inode->i_sb; | |
1366 | struct gfs2_sbd *sdp = sb->s_fs_info; | |
1367 | struct gfs2_inode *ip = GFS2_I(inode); | |
1368 | struct address_space *metamapping; | |
1369 | int ret; | |
1370 | ||
1371 | gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL | | |
1372 | GFS2_LFC_EVICT_INODE); | |
1373 | metamapping = gfs2_glock2aspace(ip->i_gl); | |
1374 | if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) { | |
1375 | filemap_fdatawrite(metamapping); | |
1376 | filemap_fdatawait(metamapping); | |
1377 | } | |
1378 | write_inode_now(inode, 1); | |
1379 | gfs2_ail_flush(ip->i_gl, 0); | |
1380 | ||
1381 | ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks); | |
1382 | if (ret) | |
1383 | return ret; | |
1384 | ||
1385 | /* Needs to be done before glock release & also in a transaction */ | |
1386 | truncate_inode_pages(&inode->i_data, 0); | |
1387 | truncate_inode_pages(metamapping, 0); | |
1388 | gfs2_trans_end(sdp); | |
1389 | return 0; | |
1390 | } | |
1391 | ||
380f7c65 SW |
1392 | /** |
1393 | * gfs2_evict_inode - Remove an inode from cache | |
1394 | * @inode: The inode to evict | |
1395 | * | |
1396 | * There are three cases to consider: | |
1397 | * 1. i_nlink == 0, we are final opener (and must deallocate) | |
1398 | * 2. i_nlink == 0, we are not the final opener (and cannot deallocate) | |
1399 | * 3. i_nlink > 0 | |
1400 | * | |
1401 | * If the fs is read only, then we have to treat all cases as per #3 | |
1402 | * since we are unable to do any deallocation. The inode will be | |
1403 | * deallocated by the next read/write node to attempt an allocation | |
1404 | * in the same resource group | |
1405 | * | |
9e6e0a12 SW |
1406 | * We have to (at the moment) hold the inodes main lock to cover |
1407 | * the gap between unlocking the shared lock on the iopen lock and | |
1408 | * taking the exclusive lock. I'd rather do a shared -> exclusive | |
1409 | * conversion on the iopen lock, but we can change that later. This | |
1410 | * is safe, just less efficient. | |
1411 | */ | |
1412 | ||
d5c1515c | 1413 | static void gfs2_evict_inode(struct inode *inode) |
9e6e0a12 | 1414 | { |
001e8e8d SW |
1415 | struct super_block *sb = inode->i_sb; |
1416 | struct gfs2_sbd *sdp = sb->s_fs_info; | |
9e6e0a12 SW |
1417 | struct gfs2_inode *ip = GFS2_I(inode); |
1418 | struct gfs2_holder gh; | |
23d828fc | 1419 | int ret; |
9e6e0a12 | 1420 | |
38552ff6 | 1421 | if (inode->i_nlink || sb_rdonly(sb) || !ip->i_no_addr) |
d5c1515c AV |
1422 | goto out; |
1423 | ||
504a10d9 BP |
1424 | /* |
1425 | * In case of an incomplete mount, gfs2_evict_inode() may be called for | |
1426 | * system files without having an active journal to write to. In that | |
1427 | * case, skip the filesystem evict. | |
1428 | */ | |
1429 | if (!sdp->sd_jdesc) | |
1430 | goto out; | |
1431 | ||
53dbc27e BP |
1432 | gfs2_holder_mark_uninitialized(&gh); |
1433 | ret = evict_should_delete(inode, &gh); | |
1434 | if (ret == SHOULD_DEFER_EVICTION) | |
8c7b9262 | 1435 | goto out; |
0a0d9f55 BP |
1436 | if (ret == SHOULD_DELETE_DINODE) |
1437 | ret = evict_unlinked_inode(inode); | |
1438 | else | |
1439 | ret = evict_linked_inode(inode); | |
9e6e0a12 | 1440 | |
a097dc7e BP |
1441 | if (gfs2_rs_active(&ip->i_res)) |
1442 | gfs2_rs_deltree(&ip->i_res); | |
8e2e0047 | 1443 | |
9ffa1888 | 1444 | if (gfs2_holder_initialized(&gh)) |
e0b62e21 | 1445 | gfs2_glock_dq_uninit(&gh); |
23d828fc BP |
1446 | if (ret && ret != GLR_TRYFAILED && ret != -EROFS) |
1447 | fs_warn(sdp, "gfs2_evict_inode: %d\n", ret); | |
9e6e0a12 | 1448 | out: |
91b0abe3 | 1449 | truncate_inode_pages_final(&inode->i_data); |
2fba46a0 BP |
1450 | if (ip->i_qadata) |
1451 | gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0); | |
7336905a | 1452 | gfs2_rs_deltree(&ip->i_res); |
45138990 | 1453 | gfs2_ordered_del_inode(ip); |
dbd5768f | 1454 | clear_inode(inode); |
17d539f0 | 1455 | gfs2_dir_hash_inval(ip); |
6df9f9a2 | 1456 | if (gfs2_holder_initialized(&ip->i_iopen_gh)) { |
71c1b213 AG |
1457 | struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl; |
1458 | ||
1459 | glock_clear_object(gl, ip); | |
71c1b213 | 1460 | gfs2_glock_hold(gl); |
fe1bff65 AG |
1461 | ip->i_iopen_gh.gh_flags |= GL_NOCACHE; |
1462 | gfs2_glock_dq_uninit(&ip->i_iopen_gh); | |
71c1b213 | 1463 | gfs2_glock_put_eventually(gl); |
d5c1515c | 1464 | } |
49462e2b BP |
1465 | if (ip->i_gl) { |
1466 | glock_clear_object(ip->i_gl, ip); | |
1467 | wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE); | |
1468 | gfs2_glock_add_to_lru(ip->i_gl); | |
1469 | gfs2_glock_put_eventually(ip->i_gl); | |
1470 | ip->i_gl = NULL; | |
1471 | } | |
9e6e0a12 SW |
1472 | } |
1473 | ||
1474 | static struct inode *gfs2_alloc_inode(struct super_block *sb) | |
1475 | { | |
1476 | struct gfs2_inode *ip; | |
1477 | ||
fd60b288 | 1478 | ip = alloc_inode_sb(sb, gfs2_inode_cachep, GFP_KERNEL); |
d4031259 AG |
1479 | if (!ip) |
1480 | return NULL; | |
38552ff6 | 1481 | ip->i_no_addr = 0; |
d4031259 AG |
1482 | ip->i_flags = 0; |
1483 | ip->i_gl = NULL; | |
40e7e86e | 1484 | gfs2_holder_mark_uninitialized(&ip->i_iopen_gh); |
d4031259 AG |
1485 | memset(&ip->i_res, 0, sizeof(ip->i_res)); |
1486 | RB_CLEAR_NODE(&ip->i_res.rs_node); | |
1487 | ip->i_rahead = 0; | |
9e6e0a12 SW |
1488 | return &ip->i_inode; |
1489 | } | |
1490 | ||
784494e1 | 1491 | static void gfs2_free_inode(struct inode *inode) |
9e6e0a12 | 1492 | { |
784494e1 | 1493 | kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode)); |
fa0d7e3d NP |
1494 | } |
1495 | ||
97fd734b AD |
1496 | extern void free_local_statfs_inodes(struct gfs2_sbd *sdp) |
1497 | { | |
1498 | struct local_statfs_inode *lsi, *safe; | |
1499 | ||
1500 | /* Run through the statfs inodes list to iput and free memory */ | |
1501 | list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) { | |
1502 | if (lsi->si_jid == sdp->sd_jdesc->jd_jid) | |
1503 | sdp->sd_sc_inode = NULL; /* belongs to this node */ | |
1504 | if (lsi->si_sc_inode) | |
1505 | iput(lsi->si_sc_inode); | |
1506 | list_del(&lsi->si_list); | |
1507 | kfree(lsi); | |
1508 | } | |
1509 | } | |
1510 | ||
1511 | extern struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp, | |
1512 | unsigned int index) | |
1513 | { | |
1514 | struct local_statfs_inode *lsi; | |
1515 | ||
1516 | /* Return the local (per node) statfs inode in the | |
1517 | * sdp->sd_sc_inodes_list corresponding to the 'index'. */ | |
1518 | list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) { | |
1519 | if (lsi->si_jid == index) | |
1520 | return lsi->si_sc_inode; | |
1521 | } | |
1522 | return NULL; | |
1523 | } | |
1524 | ||
9e6e0a12 SW |
1525 | const struct super_operations gfs2_super_ops = { |
1526 | .alloc_inode = gfs2_alloc_inode, | |
784494e1 | 1527 | .free_inode = gfs2_free_inode, |
9e6e0a12 | 1528 | .write_inode = gfs2_write_inode, |
ab9bbda0 | 1529 | .dirty_inode = gfs2_dirty_inode, |
d5c1515c | 1530 | .evict_inode = gfs2_evict_inode, |
9e6e0a12 | 1531 | .put_super = gfs2_put_super, |
9e6e0a12 | 1532 | .sync_fs = gfs2_sync_fs, |
2e60d768 BM |
1533 | .freeze_super = gfs2_freeze, |
1534 | .thaw_super = gfs2_unfreeze, | |
9e6e0a12 | 1535 | .statfs = gfs2_statfs, |
9e6e0a12 SW |
1536 | .drop_inode = gfs2_drop_inode, |
1537 | .show_options = gfs2_show_options, | |
1538 | }; | |
1539 |