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 | ||
35264909 | 70 | down_write(&sdp->sd_log_flush_lock); |
601ef0d5 | 71 | sdp->sd_jdesc = NULL; |
35264909 AG |
72 | up_write(&sdp->sd_log_flush_lock); |
73 | ||
fefc03bf | 74 | while (!list_empty(&list)) { |
969183bc | 75 | jd = list_first_entry(&list, struct gfs2_jdesc, jd_list); |
35264909 | 76 | BUG_ON(jd->jd_log_bio); |
b50f227b | 77 | gfs2_free_journal_extents(jd); |
fefc03bf SW |
78 | list_del(&jd->jd_list); |
79 | iput(jd->jd_inode); | |
601ef0d5 | 80 | jd->jd_inode = NULL; |
fefc03bf SW |
81 | kfree(jd); |
82 | } | |
83 | } | |
84 | ||
b3b94faa DT |
85 | static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid) |
86 | { | |
87 | struct gfs2_jdesc *jd; | |
b3b94faa DT |
88 | |
89 | list_for_each_entry(jd, head, jd_list) { | |
736b2f77 AG |
90 | if (jd->jd_jid == jid) |
91 | return jd; | |
b3b94faa | 92 | } |
736b2f77 | 93 | return NULL; |
b3b94faa DT |
94 | } |
95 | ||
96 | struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid) | |
97 | { | |
98 | struct gfs2_jdesc *jd; | |
99 | ||
100 | spin_lock(&sdp->sd_jindex_spin); | |
101 | jd = jdesc_find_i(&sdp->sd_jindex_list, jid); | |
102 | spin_unlock(&sdp->sd_jindex_spin); | |
103 | ||
104 | return jd; | |
105 | } | |
106 | ||
b3b94faa DT |
107 | int gfs2_jdesc_check(struct gfs2_jdesc *jd) |
108 | { | |
feaa7bba SW |
109 | struct gfs2_inode *ip = GFS2_I(jd->jd_inode); |
110 | struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); | |
a2e0f799 | 111 | u64 size = i_size_read(jd->jd_inode); |
b3b94faa | 112 | |
47a9a527 | 113 | if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30))) |
b3b94faa | 114 | return -EIO; |
b3b94faa | 115 | |
a2e0f799 SW |
116 | jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift; |
117 | ||
118 | if (gfs2_write_alloc_required(ip, 0, size)) { | |
b3b94faa | 119 | gfs2_consist_inode(ip); |
461cb419 | 120 | return -EIO; |
b3b94faa DT |
121 | } |
122 | ||
461cb419 | 123 | return 0; |
b3b94faa DT |
124 | } |
125 | ||
b3b94faa DT |
126 | /** |
127 | * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one | |
128 | * @sdp: the filesystem | |
129 | * | |
130 | * Returns: errno | |
131 | */ | |
132 | ||
133 | int gfs2_make_fs_rw(struct gfs2_sbd *sdp) | |
134 | { | |
feaa7bba | 135 | struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode); |
5c676f6d | 136 | struct gfs2_glock *j_gl = ip->i_gl; |
55167622 | 137 | struct gfs2_log_header_host head; |
b3b94faa DT |
138 | int error; |
139 | ||
1a14d3a6 | 140 | j_gl->gl_ops->go_inval(j_gl, DIO_METADATA); |
4d927b03 | 141 | if (gfs2_withdrawing_or_withdrawn(sdp)) |
a28dc123 | 142 | return -EIO; |
b3b94faa | 143 | |
f4686c26 | 144 | error = gfs2_find_jhead(sdp->sd_jdesc, &head, false); |
b66f723b AG |
145 | if (error) { |
146 | gfs2_consist(sdp); | |
a28dc123 | 147 | return error; |
b66f723b | 148 | } |
b3b94faa DT |
149 | |
150 | if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) { | |
151 | gfs2_consist(sdp); | |
a28dc123 | 152 | return -EIO; |
b3b94faa DT |
153 | } |
154 | ||
155 | /* Initialize some head of the log stuff */ | |
156 | sdp->sd_log_sequence = head.lh_sequence + 1; | |
157 | gfs2_log_pointers_init(sdp, head.lh_blkno); | |
158 | ||
b3b94faa | 159 | error = gfs2_quota_init(sdp); |
4d927b03 | 160 | if (!error && gfs2_withdrawing_or_withdrawn(sdp)) |
b66f723b AG |
161 | error = -EIO; |
162 | if (!error) | |
a28dc123 | 163 | set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags); |
b3b94faa DT |
164 | return error; |
165 | } | |
166 | ||
1946f70a | 167 | void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf) |
bb8d8a6f SW |
168 | { |
169 | const struct gfs2_statfs_change *str = buf; | |
170 | ||
171 | sc->sc_total = be64_to_cpu(str->sc_total); | |
172 | sc->sc_free = be64_to_cpu(str->sc_free); | |
173 | sc->sc_dinodes = be64_to_cpu(str->sc_dinodes); | |
174 | } | |
175 | ||
73092698 | 176 | void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf) |
bb8d8a6f SW |
177 | { |
178 | struct gfs2_statfs_change *str = buf; | |
179 | ||
180 | str->sc_total = cpu_to_be64(sc->sc_total); | |
181 | str->sc_free = cpu_to_be64(sc->sc_free); | |
182 | str->sc_dinodes = cpu_to_be64(sc->sc_dinodes); | |
183 | } | |
184 | ||
b3b94faa DT |
185 | int gfs2_statfs_init(struct gfs2_sbd *sdp) |
186 | { | |
feaa7bba | 187 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); |
bd209cc0 | 188 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; |
bd209cc0 | 189 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; |
70c11ba8 | 190 | struct buffer_head *m_bh; |
b3b94faa DT |
191 | struct gfs2_holder gh; |
192 | int error; | |
193 | ||
194 | error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE, | |
195 | &gh); | |
196 | if (error) | |
197 | return error; | |
198 | ||
199 | error = gfs2_meta_inode_buffer(m_ip, &m_bh); | |
200 | if (error) | |
201 | goto out; | |
202 | ||
203 | if (sdp->sd_args.ar_spectator) { | |
204 | spin_lock(&sdp->sd_statfs_spin); | |
205 | gfs2_statfs_change_in(m_sc, m_bh->b_data + | |
206 | sizeof(struct gfs2_dinode)); | |
207 | spin_unlock(&sdp->sd_statfs_spin); | |
208 | } else { | |
b3b94faa DT |
209 | spin_lock(&sdp->sd_statfs_spin); |
210 | gfs2_statfs_change_in(m_sc, m_bh->b_data + | |
211 | sizeof(struct gfs2_dinode)); | |
70c11ba8 | 212 | gfs2_statfs_change_in(l_sc, sdp->sd_sc_bh->b_data + |
b3b94faa DT |
213 | sizeof(struct gfs2_dinode)); |
214 | spin_unlock(&sdp->sd_statfs_spin); | |
215 | ||
b3b94faa DT |
216 | } |
217 | ||
b3b94faa | 218 | brelse(m_bh); |
a91ea69f | 219 | out: |
b3b94faa | 220 | gfs2_glock_dq_uninit(&gh); |
b3b94faa DT |
221 | return 0; |
222 | } | |
223 | ||
cd915493 SW |
224 | void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free, |
225 | s64 dinodes) | |
b3b94faa | 226 | { |
feaa7bba | 227 | struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode); |
bd209cc0 | 228 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; |
3d3c10f2 | 229 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; |
c14f5735 BM |
230 | s64 x, y; |
231 | int need_sync = 0; | |
b3b94faa | 232 | |
70c11ba8 | 233 | gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh); |
b3b94faa DT |
234 | |
235 | spin_lock(&sdp->sd_statfs_spin); | |
236 | l_sc->sc_total += total; | |
237 | l_sc->sc_free += free; | |
238 | l_sc->sc_dinodes += dinodes; | |
70c11ba8 BP |
239 | gfs2_statfs_change_out(l_sc, sdp->sd_sc_bh->b_data + |
240 | sizeof(struct gfs2_dinode)); | |
c14f5735 BM |
241 | if (sdp->sd_args.ar_statfs_percent) { |
242 | x = 100 * l_sc->sc_free; | |
243 | y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent; | |
244 | if (x >= y || x <= -y) | |
245 | need_sync = 1; | |
246 | } | |
b3b94faa DT |
247 | spin_unlock(&sdp->sd_statfs_spin); |
248 | ||
c14f5735 | 249 | if (need_sync) |
3d3c10f2 | 250 | gfs2_wake_up_statfs(sdp); |
b3b94faa DT |
251 | } |
252 | ||
70c11ba8 | 253 | void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh) |
1946f70a BM |
254 | { |
255 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); | |
256 | struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode); | |
257 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; | |
258 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; | |
259 | ||
70c11ba8 | 260 | gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh); |
901c6c66 | 261 | gfs2_trans_add_meta(m_ip->i_gl, m_bh); |
1946f70a BM |
262 | |
263 | spin_lock(&sdp->sd_statfs_spin); | |
264 | m_sc->sc_total += l_sc->sc_total; | |
265 | m_sc->sc_free += l_sc->sc_free; | |
266 | m_sc->sc_dinodes += l_sc->sc_dinodes; | |
267 | memset(l_sc, 0, sizeof(struct gfs2_statfs_change)); | |
70c11ba8 | 268 | memset(sdp->sd_sc_bh->b_data + sizeof(struct gfs2_dinode), |
1946f70a | 269 | 0, sizeof(struct gfs2_statfs_change)); |
1946f70a | 270 | gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode)); |
901c6c66 | 271 | spin_unlock(&sdp->sd_statfs_spin); |
1946f70a BM |
272 | } |
273 | ||
8c42d637 | 274 | int gfs2_statfs_sync(struct super_block *sb, int type) |
b3b94faa | 275 | { |
8c42d637 | 276 | struct gfs2_sbd *sdp = sb->s_fs_info; |
feaa7bba | 277 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); |
bd209cc0 AV |
278 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; |
279 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; | |
b3b94faa | 280 | struct gfs2_holder gh; |
70c11ba8 | 281 | struct buffer_head *m_bh; |
b3b94faa DT |
282 | int error; |
283 | ||
284 | error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE, | |
285 | &gh); | |
286 | if (error) | |
2e60d768 | 287 | goto out; |
b3b94faa DT |
288 | |
289 | error = gfs2_meta_inode_buffer(m_ip, &m_bh); | |
290 | if (error) | |
2e60d768 | 291 | goto out_unlock; |
b3b94faa DT |
292 | |
293 | spin_lock(&sdp->sd_statfs_spin); | |
294 | gfs2_statfs_change_in(m_sc, m_bh->b_data + | |
907b9bce | 295 | sizeof(struct gfs2_dinode)); |
b3b94faa DT |
296 | if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) { |
297 | spin_unlock(&sdp->sd_statfs_spin); | |
298 | goto out_bh; | |
299 | } | |
300 | spin_unlock(&sdp->sd_statfs_spin); | |
301 | ||
b3b94faa DT |
302 | error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0); |
303 | if (error) | |
70c11ba8 | 304 | goto out_bh; |
b3b94faa | 305 | |
70c11ba8 | 306 | update_statfs(sdp, m_bh); |
3d3c10f2 | 307 | sdp->sd_statfs_force_sync = 0; |
b3b94faa DT |
308 | |
309 | gfs2_trans_end(sdp); | |
310 | ||
a91ea69f | 311 | out_bh: |
b3b94faa | 312 | brelse(m_bh); |
2e60d768 | 313 | out_unlock: |
b3b94faa | 314 | gfs2_glock_dq_uninit(&gh); |
2e60d768 | 315 | out: |
b3b94faa DT |
316 | return error; |
317 | } | |
318 | ||
b3b94faa DT |
319 | struct lfcc { |
320 | struct list_head list; | |
321 | struct gfs2_holder gh; | |
322 | }; | |
323 | ||
324 | /** | |
325 | * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all | |
326 | * journals are clean | |
327 | * @sdp: the file system | |
b3b94faa DT |
328 | * |
329 | * Returns: errno | |
330 | */ | |
331 | ||
52b1cdcb | 332 | static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp) |
b3b94faa | 333 | { |
5c676f6d | 334 | struct gfs2_inode *ip; |
b3b94faa DT |
335 | struct gfs2_jdesc *jd; |
336 | struct lfcc *lfcc; | |
337 | LIST_HEAD(list); | |
55167622 | 338 | struct gfs2_log_header_host lh; |
b77b4a48 AG |
339 | int error, error2; |
340 | ||
341 | /* | |
342 | * Grab all the journal glocks in SH mode. We are *probably* doing | |
343 | * that to prevent recovery. | |
344 | */ | |
b3b94faa | 345 | |
b3b94faa DT |
346 | list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) { |
347 | lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL); | |
348 | if (!lfcc) { | |
349 | error = -ENOMEM; | |
350 | goto out; | |
351 | } | |
feaa7bba SW |
352 | ip = GFS2_I(jd->jd_inode); |
353 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh); | |
b3b94faa DT |
354 | if (error) { |
355 | kfree(lfcc); | |
356 | goto out; | |
357 | } | |
358 | list_add(&lfcc->list, &list); | |
359 | } | |
360 | ||
b77b4a48 AG |
361 | gfs2_freeze_unlock(&sdp->sd_freeze_gh); |
362 | ||
24972557 | 363 | error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE, |
ebdc416c AG |
364 | LM_FLAG_NOEXP | GL_NOPID, |
365 | &sdp->sd_freeze_gh); | |
52b1cdcb | 366 | if (error) |
b77b4a48 | 367 | goto relock_shared; |
b3b94faa DT |
368 | |
369 | list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) { | |
370 | error = gfs2_jdesc_check(jd); | |
371 | if (error) | |
372 | break; | |
f4686c26 | 373 | error = gfs2_find_jhead(jd, &lh, false); |
b3b94faa DT |
374 | if (error) |
375 | break; | |
376 | if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) { | |
377 | error = -EBUSY; | |
378 | break; | |
379 | } | |
380 | } | |
381 | ||
b77b4a48 AG |
382 | if (!error) |
383 | goto out; /* success */ | |
384 | ||
385 | gfs2_freeze_unlock(&sdp->sd_freeze_gh); | |
386 | ||
387 | relock_shared: | |
6c7410f4 | 388 | error2 = gfs2_freeze_lock_shared(sdp); |
b77b4a48 | 389 | gfs2_assert_withdraw(sdp, !error2); |
b3b94faa | 390 | |
a91ea69f | 391 | out: |
b3b94faa | 392 | while (!list_empty(&list)) { |
969183bc | 393 | lfcc = list_first_entry(&list, struct lfcc, list); |
b3b94faa DT |
394 | list_del(&lfcc->list); |
395 | gfs2_glock_dq_uninit(&lfcc->gh); | |
396 | kfree(lfcc); | |
397 | } | |
b3b94faa DT |
398 | return error; |
399 | } | |
400 | ||
9eed04cd SW |
401 | void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf) |
402 | { | |
7db35444 | 403 | const struct inode *inode = &ip->i_inode; |
9eed04cd SW |
404 | struct gfs2_dinode *str = buf; |
405 | ||
406 | str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC); | |
407 | str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI); | |
408 | str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI); | |
409 | str->di_num.no_addr = cpu_to_be64(ip->i_no_addr); | |
410 | str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino); | |
7db35444 AG |
411 | str->di_mode = cpu_to_be32(inode->i_mode); |
412 | str->di_uid = cpu_to_be32(i_uid_read(inode)); | |
413 | str->di_gid = cpu_to_be32(i_gid_read(inode)); | |
414 | str->di_nlink = cpu_to_be32(inode->i_nlink); | |
415 | str->di_size = cpu_to_be64(i_size_read(inode)); | |
416 | str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(inode)); | |
580f721b JL |
417 | str->di_atime = cpu_to_be64(inode_get_atime_sec(inode)); |
418 | str->di_mtime = cpu_to_be64(inode_get_mtime_sec(inode)); | |
419 | str->di_ctime = cpu_to_be64(inode_get_ctime_sec(inode)); | |
9eed04cd SW |
420 | |
421 | str->di_goal_meta = cpu_to_be64(ip->i_goal); | |
422 | str->di_goal_data = cpu_to_be64(ip->i_goal); | |
423 | str->di_generation = cpu_to_be64(ip->i_generation); | |
424 | ||
425 | str->di_flags = cpu_to_be32(ip->i_diskflags); | |
426 | str->di_height = cpu_to_be16(ip->i_height); | |
7db35444 | 427 | str->di_payload_format = cpu_to_be32(S_ISDIR(inode->i_mode) && |
9eed04cd SW |
428 | !(ip->i_diskflags & GFS2_DIF_EXHASH) ? |
429 | GFS2_FORMAT_DE : 0); | |
430 | str->di_depth = cpu_to_be16(ip->i_depth); | |
431 | str->di_entries = cpu_to_be32(ip->i_entries); | |
432 | ||
433 | str->di_eattr = cpu_to_be64(ip->i_eattr); | |
580f721b JL |
434 | str->di_atime_nsec = cpu_to_be32(inode_get_atime_nsec(inode)); |
435 | str->di_mtime_nsec = cpu_to_be32(inode_get_mtime_nsec(inode)); | |
436 | str->di_ctime_nsec = cpu_to_be32(inode_get_ctime_nsec(inode)); | |
9eed04cd | 437 | } |
9e6e0a12 SW |
438 | |
439 | /** | |
440 | * gfs2_write_inode - Make sure the inode is stable on the disk | |
441 | * @inode: The inode | |
1027efaa | 442 | * @wbc: The writeback control structure |
9e6e0a12 SW |
443 | * |
444 | * Returns: errno | |
445 | */ | |
446 | ||
a9185b41 | 447 | static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc) |
9e6e0a12 SW |
448 | { |
449 | struct gfs2_inode *ip = GFS2_I(inode); | |
450 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1027efaa | 451 | struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl); |
de1414a6 | 452 | struct backing_dev_info *bdi = inode_to_bdi(metamapping->host); |
ab9bbda0 | 453 | int ret = 0; |
adbc3ddf | 454 | bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip)); |
ab9bbda0 | 455 | |
adbc3ddf | 456 | if (flush_all) |
c1696fb8 | 457 | gfs2_log_flush(GFS2_SB(inode), ip->i_gl, |
805c0907 BP |
458 | GFS2_LOG_HEAD_FLUSH_NORMAL | |
459 | GFS2_LFC_WRITE_INODE); | |
a88a341a | 460 | if (bdi->wb.dirty_exceeded) |
4667a0ec | 461 | gfs2_ail1_flush(sdp, wbc); |
1d4ec642 SW |
462 | else |
463 | filemap_fdatawrite(metamapping); | |
adbc3ddf | 464 | if (flush_all) |
1027efaa SW |
465 | ret = filemap_fdatawait(metamapping); |
466 | if (ret) | |
467 | mark_inode_dirty_sync(inode); | |
957a7acd AD |
468 | else { |
469 | spin_lock(&inode->i_lock); | |
470 | if (!(inode->i_flags & I_DIRTY)) | |
471 | gfs2_ordered_del_inode(ip); | |
472 | spin_unlock(&inode->i_lock); | |
473 | } | |
9e6e0a12 SW |
474 | return ret; |
475 | } | |
476 | ||
ab9bbda0 SW |
477 | /** |
478 | * gfs2_dirty_inode - check for atime updates | |
479 | * @inode: The inode in question | |
480 | * @flags: The type of dirty | |
481 | * | |
482 | * Unfortunately it can be called under any combination of inode | |
af1abe11 | 483 | * glock and freeze glock, so we have to check carefully. |
ab9bbda0 SW |
484 | * |
485 | * At the moment this deals only with atime - it should be possible | |
486 | * to expand that role in future, once a review of the locking has | |
487 | * been carried out. | |
488 | */ | |
489 | ||
490 | static void gfs2_dirty_inode(struct inode *inode, int flags) | |
491 | { | |
492 | struct gfs2_inode *ip = GFS2_I(inode); | |
493 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
494 | struct buffer_head *bh; | |
495 | struct gfs2_holder gh; | |
496 | int need_unlock = 0; | |
497 | int need_endtrans = 0; | |
498 | int ret; | |
499 | ||
38552ff6 AG |
500 | if (unlikely(!ip->i_gl)) { |
501 | /* This can only happen during incomplete inode creation. */ | |
502 | BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags)); | |
503 | return; | |
504 | } | |
505 | ||
4d927b03 | 506 | if (gfs2_withdrawing_or_withdrawn(sdp)) |
0d1c7ae9 | 507 | return; |
ab9bbda0 SW |
508 | if (!gfs2_glock_is_locked_by_me(ip->i_gl)) { |
509 | ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); | |
510 | if (ret) { | |
511 | fs_err(sdp, "dirty_inode: glock %d\n", ret); | |
e28c02b9 | 512 | gfs2_dump_glock(NULL, ip->i_gl, true); |
ab9bbda0 SW |
513 | return; |
514 | } | |
515 | need_unlock = 1; | |
3d162688 BM |
516 | } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE)) |
517 | return; | |
ab9bbda0 SW |
518 | |
519 | if (current->journal_info == NULL) { | |
520 | ret = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
521 | if (ret) { | |
522 | fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret); | |
523 | goto out; | |
524 | } | |
525 | need_endtrans = 1; | |
526 | } | |
527 | ||
528 | ret = gfs2_meta_inode_buffer(ip, &bh); | |
529 | if (ret == 0) { | |
350a9b0a | 530 | gfs2_trans_add_meta(ip->i_gl, bh); |
ab9bbda0 SW |
531 | gfs2_dinode_out(ip, bh->b_data); |
532 | brelse(bh); | |
533 | } | |
534 | ||
535 | if (need_endtrans) | |
536 | gfs2_trans_end(sdp); | |
537 | out: | |
538 | if (need_unlock) | |
539 | gfs2_glock_dq_uninit(&gh); | |
540 | } | |
541 | ||
9e6e0a12 SW |
542 | /** |
543 | * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one | |
544 | * @sdp: the filesystem | |
545 | * | |
546 | * Returns: errno | |
547 | */ | |
548 | ||
eb602521 | 549 | void gfs2_make_fs_ro(struct gfs2_sbd *sdp) |
9e6e0a12 | 550 | { |
601ef0d5 BP |
551 | int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags); |
552 | ||
e7beb8b6 | 553 | if (!test_bit(SDF_KILL, &sdp->sd_flags)) |
6b388abc AG |
554 | gfs2_flush_delete_work(sdp); |
555 | ||
fe4f7940 | 556 | gfs2_destroy_threads(sdp); |
8ad151c2 | 557 | |
601ef0d5 BP |
558 | if (log_write_allowed) { |
559 | gfs2_quota_sync(sdp->sd_vfs, 0); | |
560 | gfs2_statfs_sync(sdp->sd_vfs, 0); | |
9e6e0a12 | 561 | |
68ca088d BP |
562 | /* We do two log flushes here. The first one commits dirty inodes |
563 | * and rgrps to the journal, but queues up revokes to the ail list. | |
564 | * The second flush writes out and removes the revokes. | |
565 | * | |
566 | * The first must be done before the FLUSH_SHUTDOWN code | |
567 | * clears the LIVE flag, otherwise it will not be able to start | |
568 | * a transaction to write its revokes, and the error will cause | |
569 | * a withdraw of the file system. */ | |
570 | gfs2_log_flush(sdp, NULL, GFS2_LFC_MAKE_FS_RO); | |
601ef0d5 BP |
571 | gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN | |
572 | GFS2_LFC_MAKE_FS_RO); | |
f3708fb5 AG |
573 | wait_event_timeout(sdp->sd_log_waitq, |
574 | gfs2_log_is_empty(sdp), | |
575 | HZ * 5); | |
576 | gfs2_assert_warn(sdp, gfs2_log_is_empty(sdp)); | |
601ef0d5 | 577 | } |
9e6e0a12 | 578 | gfs2_quota_cleanup(sdp); |
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 | ||
bdcb8aa4 | 609 | if (!sb_rdonly(sb)) |
eb602521 | 610 | gfs2_make_fs_ro(sdp); |
bdcb8aa4 | 611 | else { |
4d927b03 | 612 | if (gfs2_withdrawing_or_withdrawn(sdp)) |
bdcb8aa4 JD |
613 | gfs2_destroy_threads(sdp); |
614 | ||
f66af88e | 615 | gfs2_quota_cleanup(sdp); |
fe4f7940 | 616 | } |
bdcb8aa4 | 617 | |
5a61ae14 AG |
618 | WARN_ON(gfs2_withdrawing(sdp)); |
619 | ||
9e6e0a12 SW |
620 | /* At this point, we're through modifying the disk */ |
621 | ||
622 | /* Release stuff */ | |
623 | ||
b77b4a48 AG |
624 | gfs2_freeze_unlock(&sdp->sd_freeze_gh); |
625 | ||
9e6e0a12 | 626 | iput(sdp->sd_jindex); |
9e6e0a12 SW |
627 | iput(sdp->sd_statfs_inode); |
628 | iput(sdp->sd_rindex); | |
629 | iput(sdp->sd_quota_inode); | |
630 | ||
631 | gfs2_glock_put(sdp->sd_rename_gl); | |
24972557 | 632 | gfs2_glock_put(sdp->sd_freeze_gl); |
9e6e0a12 SW |
633 | |
634 | if (!sdp->sd_args.ar_spectator) { | |
601ef0d5 BP |
635 | if (gfs2_holder_initialized(&sdp->sd_journal_gh)) |
636 | gfs2_glock_dq_uninit(&sdp->sd_journal_gh); | |
637 | if (gfs2_holder_initialized(&sdp->sd_jinode_gh)) | |
638 | gfs2_glock_dq_uninit(&sdp->sd_jinode_gh); | |
70c11ba8 | 639 | brelse(sdp->sd_sc_bh); |
9e6e0a12 SW |
640 | gfs2_glock_dq_uninit(&sdp->sd_sc_gh); |
641 | gfs2_glock_dq_uninit(&sdp->sd_qc_gh); | |
97fd734b | 642 | free_local_statfs_inodes(sdp); |
9e6e0a12 SW |
643 | iput(sdp->sd_qc_inode); |
644 | } | |
645 | ||
646 | gfs2_glock_dq_uninit(&sdp->sd_live_gh); | |
647 | gfs2_clear_rgrpd(sdp); | |
648 | gfs2_jindex_free(sdp); | |
649 | /* Take apart glock structures and buffer lists */ | |
650 | gfs2_gl_hash_clear(sdp); | |
a9dd945c | 651 | truncate_inode_pages_final(&sdp->sd_aspace); |
b2fb7dab | 652 | gfs2_delete_debugfs_file(sdp); |
9e6e0a12 SW |
653 | /* Unmount the locking protocol */ |
654 | gfs2_lm_unmount(sdp); | |
655 | ||
656 | /* At this point, we're through participating in the lockspace */ | |
657 | gfs2_sys_fs_del(sdp); | |
c2a04b02 | 658 | free_sbd(sdp); |
9e6e0a12 SW |
659 | } |
660 | ||
9e6e0a12 SW |
661 | /** |
662 | * gfs2_sync_fs - sync the filesystem | |
663 | * @sb: the superblock | |
c551f66c | 664 | * @wait: true to wait for completion |
9e6e0a12 SW |
665 | * |
666 | * Flushes the log to disk. | |
667 | */ | |
668 | ||
669 | static int gfs2_sync_fs(struct super_block *sb, int wait) | |
670 | { | |
1027efaa | 671 | struct gfs2_sbd *sdp = sb->s_fs_info; |
a1177825 JK |
672 | |
673 | gfs2_quota_sync(sb, -1); | |
942b0cdd | 674 | if (wait) |
805c0907 BP |
675 | gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL | |
676 | GFS2_LFC_SYNC_FS); | |
942b0cdd | 677 | return sdp->sd_log_error; |
9e6e0a12 SW |
678 | } |
679 | ||
b77b4a48 AG |
680 | static int gfs2_do_thaw(struct gfs2_sbd *sdp) |
681 | { | |
682 | struct super_block *sb = sdp->sd_vfs; | |
683 | int error; | |
684 | ||
6c7410f4 | 685 | error = gfs2_freeze_lock_shared(sdp); |
b77b4a48 AG |
686 | if (error) |
687 | goto fail; | |
880b9577 | 688 | error = thaw_super(sb, FREEZE_HOLDER_USERSPACE); |
b77b4a48 AG |
689 | if (!error) |
690 | return 0; | |
691 | ||
692 | fail: | |
693 | fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n", error); | |
694 | gfs2_assert_withdraw(sdp, 0); | |
695 | return error; | |
696 | } | |
697 | ||
698 | void gfs2_freeze_func(struct work_struct *work) | |
699 | { | |
700 | struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work); | |
701 | struct super_block *sb = sdp->sd_vfs; | |
702 | int error; | |
703 | ||
704 | mutex_lock(&sdp->sd_freeze_mutex); | |
705 | error = -EBUSY; | |
5432af15 | 706 | if (test_bit(SDF_FROZEN, &sdp->sd_flags)) |
b77b4a48 AG |
707 | goto freeze_failed; |
708 | ||
e345b87b | 709 | error = freeze_super(sb, FREEZE_HOLDER_USERSPACE); |
b77b4a48 AG |
710 | if (error) |
711 | goto freeze_failed; | |
712 | ||
713 | gfs2_freeze_unlock(&sdp->sd_freeze_gh); | |
5432af15 | 714 | set_bit(SDF_FROZEN, &sdp->sd_flags); |
b77b4a48 AG |
715 | |
716 | error = gfs2_do_thaw(sdp); | |
717 | if (error) | |
718 | goto out; | |
719 | ||
5432af15 | 720 | clear_bit(SDF_FROZEN, &sdp->sd_flags); |
b77b4a48 AG |
721 | goto out; |
722 | ||
723 | freeze_failed: | |
724 | fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n", error); | |
725 | ||
726 | out: | |
727 | mutex_unlock(&sdp->sd_freeze_mutex); | |
2e60d768 | 728 | deactivate_super(sb); |
2e60d768 BM |
729 | } |
730 | ||
9e6e0a12 | 731 | /** |
097cca52 | 732 | * gfs2_freeze_super - prevent further writes to the filesystem |
9e6e0a12 SW |
733 | * @sb: the VFS structure for the filesystem |
734 | * | |
735 | */ | |
736 | ||
880b9577 | 737 | static int gfs2_freeze_super(struct super_block *sb, enum freeze_holder who) |
9e6e0a12 SW |
738 | { |
739 | struct gfs2_sbd *sdp = sb->s_fs_info; | |
ff132c5f | 740 | int error; |
9e6e0a12 | 741 | |
b77b4a48 AG |
742 | if (!mutex_trylock(&sdp->sd_freeze_mutex)) |
743 | return -EBUSY; | |
5a7a9646 AG |
744 | if (test_bit(SDF_FROZEN, &sdp->sd_flags)) { |
745 | mutex_unlock(&sdp->sd_freeze_mutex); | |
746 | return -EBUSY; | |
747 | } | |
2e60d768 | 748 | |
9e6e0a12 | 749 | for (;;) { |
e345b87b | 750 | error = freeze_super(sb, FREEZE_HOLDER_USERSPACE); |
b77b4a48 AG |
751 | if (error) { |
752 | fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n", | |
753 | error); | |
60528afa BP |
754 | goto out; |
755 | } | |
756 | ||
52b1cdcb | 757 | error = gfs2_lock_fs_check_clean(sdp); |
5a7a9646 AG |
758 | if (!error) { |
759 | set_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags); | |
760 | set_bit(SDF_FROZEN, &sdp->sd_flags); | |
761 | break; | |
762 | } | |
b77b4a48 AG |
763 | |
764 | error = gfs2_do_thaw(sdp); | |
765 | if (error) | |
766 | goto out; | |
9e6e0a12 | 767 | |
55317f5b | 768 | if (error == -EBUSY) |
9e6e0a12 | 769 | fs_err(sdp, "waiting for recovery before freeze\n"); |
52b1cdcb BP |
770 | else if (error == -EIO) { |
771 | fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due " | |
772 | "to recovery error.\n"); | |
773 | goto out; | |
774 | } else { | |
9e6e0a12 | 775 | fs_err(sdp, "error freezing FS: %d\n", error); |
52b1cdcb | 776 | } |
9e6e0a12 SW |
777 | fs_err(sdp, "retrying...\n"); |
778 | msleep(1000); | |
779 | } | |
b77b4a48 | 780 | |
2e60d768 BM |
781 | out: |
782 | mutex_unlock(&sdp->sd_freeze_mutex); | |
783 | return error; | |
9e6e0a12 SW |
784 | } |
785 | ||
e345b87b AG |
786 | static int gfs2_freeze_fs(struct super_block *sb) |
787 | { | |
788 | struct gfs2_sbd *sdp = sb->s_fs_info; | |
789 | ||
790 | if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) { | |
791 | gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE | | |
792 | GFS2_LFC_FREEZE_GO_SYNC); | |
793 | if (gfs2_withdrawing_or_withdrawn(sdp)) | |
794 | return -EIO; | |
795 | } | |
796 | return 0; | |
797 | } | |
798 | ||
9e6e0a12 | 799 | /** |
097cca52 | 800 | * gfs2_thaw_super - reallow writes to the filesystem |
9e6e0a12 SW |
801 | * @sb: the VFS structure for the filesystem |
802 | * | |
803 | */ | |
804 | ||
880b9577 | 805 | static int gfs2_thaw_super(struct super_block *sb, enum freeze_holder who) |
9e6e0a12 | 806 | { |
d564053f | 807 | struct gfs2_sbd *sdp = sb->s_fs_info; |
b77b4a48 | 808 | int error; |
d564053f | 809 | |
b77b4a48 AG |
810 | if (!mutex_trylock(&sdp->sd_freeze_mutex)) |
811 | return -EBUSY; | |
5a7a9646 AG |
812 | if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags)) { |
813 | mutex_unlock(&sdp->sd_freeze_mutex); | |
814 | return -EINVAL; | |
815 | } | |
b77b4a48 | 816 | |
4e58543e | 817 | atomic_inc(&sb->s_active); |
b77b4a48 AG |
818 | gfs2_freeze_unlock(&sdp->sd_freeze_gh); |
819 | ||
820 | error = gfs2_do_thaw(sdp); | |
821 | ||
822 | if (!error) { | |
823 | clear_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags); | |
5432af15 | 824 | clear_bit(SDF_FROZEN, &sdp->sd_flags); |
2e60d768 | 825 | } |
b77b4a48 | 826 | mutex_unlock(&sdp->sd_freeze_mutex); |
4e58543e | 827 | deactivate_super(sb); |
b77b4a48 AG |
828 | return error; |
829 | } | |
830 | ||
831 | void gfs2_thaw_freeze_initiator(struct super_block *sb) | |
832 | { | |
833 | struct gfs2_sbd *sdp = sb->s_fs_info; | |
834 | ||
835 | mutex_lock(&sdp->sd_freeze_mutex); | |
836 | if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags)) | |
837 | goto out; | |
2e60d768 | 838 | |
c77b52c0 | 839 | gfs2_freeze_unlock(&sdp->sd_freeze_gh); |
b77b4a48 AG |
840 | |
841 | out: | |
2e60d768 | 842 | mutex_unlock(&sdp->sd_freeze_mutex); |
9e6e0a12 SW |
843 | } |
844 | ||
845 | /** | |
c551f66c | 846 | * statfs_slow_fill - fill in the sg for a given RG |
9e6e0a12 SW |
847 | * @rgd: the RG |
848 | * @sc: the sc structure | |
849 | * | |
850 | * Returns: 0 on success, -ESTALE if the LVB is invalid | |
851 | */ | |
852 | ||
853 | static int statfs_slow_fill(struct gfs2_rgrpd *rgd, | |
854 | struct gfs2_statfs_change_host *sc) | |
855 | { | |
856 | gfs2_rgrp_verify(rgd); | |
857 | sc->sc_total += rgd->rd_data; | |
858 | sc->sc_free += rgd->rd_free; | |
859 | sc->sc_dinodes += rgd->rd_dinodes; | |
860 | return 0; | |
861 | } | |
862 | ||
863 | /** | |
864 | * gfs2_statfs_slow - Stat a filesystem using asynchronous locking | |
865 | * @sdp: the filesystem | |
866 | * @sc: the sc info that will be returned | |
867 | * | |
868 | * Any error (other than a signal) will cause this routine to fall back | |
869 | * to the synchronous version. | |
870 | * | |
871 | * FIXME: This really shouldn't busy wait like this. | |
872 | * | |
873 | * Returns: errno | |
874 | */ | |
875 | ||
876 | static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc) | |
877 | { | |
9e6e0a12 SW |
878 | struct gfs2_rgrpd *rgd_next; |
879 | struct gfs2_holder *gha, *gh; | |
880 | unsigned int slots = 64; | |
881 | unsigned int x; | |
882 | int done; | |
883 | int error = 0, err; | |
884 | ||
885 | memset(sc, 0, sizeof(struct gfs2_statfs_change_host)); | |
6da2ec56 | 886 | gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL); |
9e6e0a12 SW |
887 | if (!gha) |
888 | return -ENOMEM; | |
6df9f9a2 AG |
889 | for (x = 0; x < slots; x++) |
890 | gfs2_holder_mark_uninitialized(gha + x); | |
9e6e0a12 | 891 | |
9e6e0a12 SW |
892 | rgd_next = gfs2_rgrpd_get_first(sdp); |
893 | ||
894 | for (;;) { | |
895 | done = 1; | |
896 | ||
897 | for (x = 0; x < slots; x++) { | |
898 | gh = gha + x; | |
899 | ||
6df9f9a2 | 900 | if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) { |
9e6e0a12 SW |
901 | err = gfs2_glock_wait(gh); |
902 | if (err) { | |
903 | gfs2_holder_uninit(gh); | |
904 | error = err; | |
905 | } else { | |
6f6597ba AG |
906 | if (!error) { |
907 | struct gfs2_rgrpd *rgd = | |
908 | gfs2_glock2rgrp(gh->gh_gl); | |
909 | ||
910 | error = statfs_slow_fill(rgd, sc); | |
911 | } | |
9e6e0a12 SW |
912 | gfs2_glock_dq_uninit(gh); |
913 | } | |
914 | } | |
915 | ||
6df9f9a2 | 916 | if (gfs2_holder_initialized(gh)) |
9e6e0a12 SW |
917 | done = 0; |
918 | else if (rgd_next && !error) { | |
919 | error = gfs2_glock_nq_init(rgd_next->rd_gl, | |
920 | LM_ST_SHARED, | |
921 | GL_ASYNC, | |
922 | gh); | |
923 | rgd_next = gfs2_rgrpd_get_next(rgd_next); | |
924 | done = 0; | |
925 | } | |
926 | ||
927 | if (signal_pending(current)) | |
928 | error = -ERESTARTSYS; | |
929 | } | |
930 | ||
931 | if (done) | |
932 | break; | |
933 | ||
934 | yield(); | |
935 | } | |
936 | ||
9e6e0a12 SW |
937 | kfree(gha); |
938 | return error; | |
939 | } | |
940 | ||
941 | /** | |
942 | * gfs2_statfs_i - Do a statfs | |
943 | * @sdp: the filesystem | |
c551f66c | 944 | * @sc: the sc structure |
9e6e0a12 SW |
945 | * |
946 | * Returns: errno | |
947 | */ | |
948 | ||
949 | static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc) | |
950 | { | |
951 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; | |
952 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; | |
953 | ||
954 | spin_lock(&sdp->sd_statfs_spin); | |
955 | ||
956 | *sc = *m_sc; | |
957 | sc->sc_total += l_sc->sc_total; | |
958 | sc->sc_free += l_sc->sc_free; | |
959 | sc->sc_dinodes += l_sc->sc_dinodes; | |
960 | ||
961 | spin_unlock(&sdp->sd_statfs_spin); | |
962 | ||
963 | if (sc->sc_free < 0) | |
964 | sc->sc_free = 0; | |
965 | if (sc->sc_free > sc->sc_total) | |
966 | sc->sc_free = sc->sc_total; | |
967 | if (sc->sc_dinodes < 0) | |
968 | sc->sc_dinodes = 0; | |
969 | ||
970 | return 0; | |
971 | } | |
972 | ||
973 | /** | |
974 | * gfs2_statfs - Gather and return stats about the filesystem | |
c551f66c LJ |
975 | * @dentry: The name of the link |
976 | * @buf: The buffer | |
9e6e0a12 SW |
977 | * |
978 | * Returns: 0 on success or error code | |
979 | */ | |
980 | ||
981 | static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf) | |
982 | { | |
fc64005c | 983 | struct super_block *sb = dentry->d_sb; |
9e6e0a12 SW |
984 | struct gfs2_sbd *sdp = sb->s_fs_info; |
985 | struct gfs2_statfs_change_host sc; | |
986 | int error; | |
987 | ||
8339ee54 SW |
988 | error = gfs2_rindex_update(sdp); |
989 | if (error) | |
990 | return error; | |
991 | ||
9e6e0a12 SW |
992 | if (gfs2_tune_get(sdp, gt_statfs_slow)) |
993 | error = gfs2_statfs_slow(sdp, &sc); | |
994 | else | |
995 | error = gfs2_statfs_i(sdp, &sc); | |
996 | ||
997 | if (error) | |
998 | return error; | |
999 | ||
1000 | buf->f_type = GFS2_MAGIC; | |
1001 | buf->f_bsize = sdp->sd_sb.sb_bsize; | |
1002 | buf->f_blocks = sc.sc_total; | |
1003 | buf->f_bfree = sc.sc_free; | |
1004 | buf->f_bavail = sc.sc_free; | |
1005 | buf->f_files = sc.sc_dinodes + sc.sc_free; | |
1006 | buf->f_ffree = sc.sc_free; | |
1007 | buf->f_namelen = GFS2_FNAMESIZE; | |
d6fc6c93 | 1008 | buf->f_fsid = uuid_to_fsid(sb->s_uuid.b); |
9e6e0a12 SW |
1009 | |
1010 | return 0; | |
1011 | } | |
1012 | ||
9e6e0a12 SW |
1013 | /** |
1014 | * gfs2_drop_inode - Drop an inode (test for remote unlink) | |
1015 | * @inode: The inode to drop | |
1016 | * | |
61b91cfd | 1017 | * If we've received a callback on an iopen lock then it's because a |
9e6e0a12 SW |
1018 | * remote node tried to deallocate the inode but failed due to this node |
1019 | * still having the inode open. Here we mark the link count zero | |
1020 | * since we know that it must have reached zero if the GLF_DEMOTE flag | |
1021 | * is set on the iopen glock. If we didn't do a disk read since the | |
1022 | * remote node removed the final link then we might otherwise miss | |
1023 | * this event. This check ensures that this node will deallocate the | |
1024 | * inode's blocks, or alternatively pass the baton on to another | |
1025 | * node for later deallocation. | |
1026 | */ | |
1027 | ||
45321ac5 | 1028 | static int gfs2_drop_inode(struct inode *inode) |
9e6e0a12 SW |
1029 | { |
1030 | struct gfs2_inode *ip = GFS2_I(inode); | |
b88beb9a | 1031 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
9e6e0a12 | 1032 | |
38552ff6 | 1033 | if (inode->i_nlink && |
6df9f9a2 | 1034 | gfs2_holder_initialized(&ip->i_iopen_gh)) { |
9e6e0a12 | 1035 | struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl; |
6df9f9a2 | 1036 | if (test_bit(GLF_DEMOTE, &gl->gl_flags)) |
9e6e0a12 SW |
1037 | clear_nlink(inode); |
1038 | } | |
6a1c8f6d AG |
1039 | |
1040 | /* | |
1041 | * When under memory pressure when an inode's link count has dropped to | |
1042 | * zero, defer deleting the inode to the delete workqueue. This avoids | |
1043 | * calling into DLM under memory pressure, which can deadlock. | |
1044 | */ | |
1045 | if (!inode->i_nlink && | |
1046 | unlikely(current->flags & PF_MEMALLOC) && | |
1047 | gfs2_holder_initialized(&ip->i_iopen_gh)) { | |
1048 | struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl; | |
1049 | ||
1050 | gfs2_glock_hold(gl); | |
f0e56edc | 1051 | if (!gfs2_queue_try_to_evict(gl)) |
ee2be7d7 | 1052 | gfs2_glock_put_async(gl); |
ba3ca2bc | 1053 | return 0; |
6a1c8f6d AG |
1054 | } |
1055 | ||
b88beb9a AG |
1056 | /* |
1057 | * No longer cache inodes when trying to evict them all. | |
1058 | */ | |
1059 | if (test_bit(SDF_EVICTING, &sdp->sd_flags)) | |
1060 | return 1; | |
1061 | ||
45321ac5 | 1062 | return generic_drop_inode(inode); |
9e6e0a12 SW |
1063 | } |
1064 | ||
9e6e0a12 SW |
1065 | /** |
1066 | * gfs2_show_options - Show mount options for /proc/mounts | |
1067 | * @s: seq_file structure | |
34c80b1d | 1068 | * @root: root of this (sub)tree |
9e6e0a12 SW |
1069 | * |
1070 | * Returns: 0 on success or error code | |
1071 | */ | |
1072 | ||
34c80b1d | 1073 | static int gfs2_show_options(struct seq_file *s, struct dentry *root) |
9e6e0a12 | 1074 | { |
34c80b1d | 1075 | struct gfs2_sbd *sdp = root->d_sb->s_fs_info; |
9e6e0a12 | 1076 | struct gfs2_args *args = &sdp->sd_args; |
6fa0a72c TL |
1077 | unsigned int logd_secs, statfs_slow, statfs_quantum, quota_quantum; |
1078 | ||
1079 | spin_lock(&sdp->sd_tune.gt_spin); | |
1080 | logd_secs = sdp->sd_tune.gt_logd_secs; | |
1081 | quota_quantum = sdp->sd_tune.gt_quota_quantum; | |
1082 | statfs_quantum = sdp->sd_tune.gt_statfs_quantum; | |
1083 | statfs_slow = sdp->sd_tune.gt_statfs_slow; | |
1084 | spin_unlock(&sdp->sd_tune.gt_spin); | |
9e6e0a12 | 1085 | |
34d63b81 | 1086 | if (is_subdir(root, sdp->sd_master_dir)) |
eaebdedc | 1087 | seq_puts(s, ",meta"); |
9e6e0a12 | 1088 | if (args->ar_lockproto[0]) |
a068acf2 | 1089 | seq_show_option(s, "lockproto", args->ar_lockproto); |
9e6e0a12 | 1090 | if (args->ar_locktable[0]) |
a068acf2 | 1091 | seq_show_option(s, "locktable", args->ar_locktable); |
9e6e0a12 | 1092 | if (args->ar_hostdata[0]) |
a068acf2 | 1093 | seq_show_option(s, "hostdata", args->ar_hostdata); |
9e6e0a12 | 1094 | if (args->ar_spectator) |
eaebdedc | 1095 | seq_puts(s, ",spectator"); |
9e6e0a12 | 1096 | if (args->ar_localflocks) |
eaebdedc | 1097 | seq_puts(s, ",localflocks"); |
9e6e0a12 | 1098 | if (args->ar_debug) |
eaebdedc | 1099 | seq_puts(s, ",debug"); |
9e6e0a12 | 1100 | if (args->ar_posix_acl) |
eaebdedc | 1101 | seq_puts(s, ",acl"); |
9e6e0a12 SW |
1102 | if (args->ar_quota != GFS2_QUOTA_DEFAULT) { |
1103 | char *state; | |
1104 | switch (args->ar_quota) { | |
1105 | case GFS2_QUOTA_OFF: | |
1106 | state = "off"; | |
1107 | break; | |
1108 | case GFS2_QUOTA_ACCOUNT: | |
1109 | state = "account"; | |
1110 | break; | |
1111 | case GFS2_QUOTA_ON: | |
1112 | state = "on"; | |
1113 | break; | |
eef46ab7 BP |
1114 | case GFS2_QUOTA_QUIET: |
1115 | state = "quiet"; | |
1116 | break; | |
9e6e0a12 SW |
1117 | default: |
1118 | state = "unknown"; | |
1119 | break; | |
1120 | } | |
1121 | seq_printf(s, ",quota=%s", state); | |
1122 | } | |
1123 | if (args->ar_suiddir) | |
eaebdedc | 1124 | seq_puts(s, ",suiddir"); |
9e6e0a12 SW |
1125 | if (args->ar_data != GFS2_DATA_DEFAULT) { |
1126 | char *state; | |
1127 | switch (args->ar_data) { | |
1128 | case GFS2_DATA_WRITEBACK: | |
1129 | state = "writeback"; | |
1130 | break; | |
1131 | case GFS2_DATA_ORDERED: | |
1132 | state = "ordered"; | |
1133 | break; | |
1134 | default: | |
1135 | state = "unknown"; | |
1136 | break; | |
1137 | } | |
1138 | seq_printf(s, ",data=%s", state); | |
1139 | } | |
1140 | if (args->ar_discard) | |
eaebdedc | 1141 | seq_puts(s, ",discard"); |
6fa0a72c TL |
1142 | if (logd_secs != 30) |
1143 | seq_printf(s, ",commit=%d", logd_secs); | |
1144 | if (statfs_quantum != 30) | |
1145 | seq_printf(s, ",statfs_quantum=%d", statfs_quantum); | |
1146 | else if (statfs_slow) | |
2b9731e8 | 1147 | seq_puts(s, ",statfs_quantum=0"); |
6fa0a72c TL |
1148 | if (quota_quantum != 60) |
1149 | seq_printf(s, ",quota_quantum=%d", quota_quantum); | |
3d3c10f2 BM |
1150 | if (args->ar_statfs_percent) |
1151 | seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent); | |
d34843d0 BP |
1152 | if (args->ar_errors != GFS2_ERRORS_DEFAULT) { |
1153 | const char *state; | |
1154 | ||
1155 | switch (args->ar_errors) { | |
1156 | case GFS2_ERRORS_WITHDRAW: | |
1157 | state = "withdraw"; | |
1158 | break; | |
1159 | case GFS2_ERRORS_PANIC: | |
1160 | state = "panic"; | |
1161 | break; | |
1162 | default: | |
1163 | state = "unknown"; | |
1164 | break; | |
1165 | } | |
1166 | seq_printf(s, ",errors=%s", state); | |
1167 | } | |
cdcfde62 | 1168 | if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) |
eaebdedc | 1169 | seq_puts(s, ",nobarrier"); |
913a71d2 | 1170 | if (test_bit(SDF_DEMOTE, &sdp->sd_flags)) |
eaebdedc | 1171 | seq_puts(s, ",demote_interface_used"); |
90306c41 | 1172 | if (args->ar_rgrplvb) |
eaebdedc | 1173 | seq_puts(s, ",rgrplvb"); |
471f3db2 BM |
1174 | if (args->ar_loccookie) |
1175 | seq_puts(s, ",loccookie"); | |
9e6e0a12 SW |
1176 | return 0; |
1177 | } | |
1178 | ||
f42ab085 SW |
1179 | static void gfs2_final_release_pages(struct gfs2_inode *ip) |
1180 | { | |
1181 | struct inode *inode = &ip->i_inode; | |
1182 | struct gfs2_glock *gl = ip->i_gl; | |
1183 | ||
38552ff6 AG |
1184 | if (unlikely(!gl)) { |
1185 | /* This can only happen during incomplete inode creation. */ | |
1186 | BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags)); | |
1187 | return; | |
1188 | } | |
1189 | ||
1190 | truncate_inode_pages(gfs2_glock2aspace(gl), 0); | |
f42ab085 SW |
1191 | truncate_inode_pages(&inode->i_data, 0); |
1192 | ||
638803d4 | 1193 | if (atomic_read(&gl->gl_revokes) == 0) { |
f42ab085 SW |
1194 | clear_bit(GLF_LFLUSH, &gl->gl_flags); |
1195 | clear_bit(GLF_DIRTY, &gl->gl_flags); | |
1196 | } | |
1197 | } | |
1198 | ||
1199 | static int gfs2_dinode_dealloc(struct gfs2_inode *ip) | |
1200 | { | |
1201 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
f42ab085 | 1202 | struct gfs2_rgrpd *rgd; |
564e12b1 | 1203 | struct gfs2_holder gh; |
f42ab085 SW |
1204 | int error; |
1205 | ||
1206 | if (gfs2_get_inode_blocks(&ip->i_inode) != 1) { | |
94fb763b | 1207 | gfs2_consist_inode(ip); |
f42ab085 SW |
1208 | return -EIO; |
1209 | } | |
1210 | ||
9b620429 | 1211 | gfs2_rindex_update(sdp); |
f42ab085 | 1212 | |
f4108a60 | 1213 | error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE); |
f42ab085 | 1214 | if (error) |
5407e242 | 1215 | return error; |
f42ab085 | 1216 | |
66fc061b | 1217 | rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1); |
f42ab085 SW |
1218 | if (!rgd) { |
1219 | gfs2_consist_inode(ip); | |
1220 | error = -EIO; | |
8339ee54 | 1221 | goto out_qs; |
f42ab085 SW |
1222 | } |
1223 | ||
4fc7ec31 BP |
1224 | error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, |
1225 | LM_FLAG_NODE_SCOPE, &gh); | |
f42ab085 | 1226 | if (error) |
8339ee54 | 1227 | goto out_qs; |
f42ab085 | 1228 | |
4667a0ec SW |
1229 | error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA, |
1230 | sdp->sd_jdesc->jd_blocks); | |
f42ab085 SW |
1231 | if (error) |
1232 | goto out_rg_gunlock; | |
1233 | ||
1234 | gfs2_free_di(rgd, ip); | |
1235 | ||
1236 | gfs2_final_release_pages(ip); | |
1237 | ||
1238 | gfs2_trans_end(sdp); | |
1239 | ||
1240 | out_rg_gunlock: | |
564e12b1 | 1241 | gfs2_glock_dq_uninit(&gh); |
f42ab085 SW |
1242 | out_qs: |
1243 | gfs2_quota_unhold(ip); | |
f42ab085 SW |
1244 | return error; |
1245 | } | |
1246 | ||
71c1b213 AG |
1247 | /** |
1248 | * gfs2_glock_put_eventually | |
1249 | * @gl: The glock to put | |
1250 | * | |
1251 | * When under memory pressure, trigger a deferred glock put to make sure we | |
1252 | * won't call into DLM and deadlock. Otherwise, put the glock directly. | |
1253 | */ | |
1254 | ||
1255 | static void gfs2_glock_put_eventually(struct gfs2_glock *gl) | |
1256 | { | |
1257 | if (current->flags & PF_MEMALLOC) | |
ee2be7d7 | 1258 | gfs2_glock_put_async(gl); |
71c1b213 AG |
1259 | else |
1260 | gfs2_glock_put(gl); | |
1261 | } | |
1262 | ||
9e73330f AG |
1263 | static bool gfs2_upgrade_iopen_glock(struct inode *inode) |
1264 | { | |
1265 | struct gfs2_inode *ip = GFS2_I(inode); | |
1266 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1267 | struct gfs2_holder *gh = &ip->i_iopen_gh; | |
1268 | long timeout = 5 * HZ; | |
1269 | int error; | |
1270 | ||
1271 | gh->gh_flags |= GL_NOCACHE; | |
1272 | gfs2_glock_dq_wait(gh); | |
1273 | ||
1274 | /* | |
2d143955 AG |
1275 | * If there are no other lock holders, we will immediately get |
1276 | * exclusive access to the iopen glock here. | |
1277 | * | |
9e73330f | 1278 | * Otherwise, the other nodes holding the lock will be notified about |
2d143955 AG |
1279 | * our locking request. If they do not have the inode open, they are |
1280 | * expected to evict the cached inode and release the lock, allowing us | |
1281 | * to proceed. | |
1282 | * | |
1283 | * Otherwise, if they cannot evict the inode, they are expected to poke | |
1284 | * the inode glock (note: not the iopen glock). We will notice that | |
1285 | * and stop waiting for the iopen glock immediately. The other node(s) | |
1286 | * are then expected to take care of deleting the inode when they no | |
1287 | * longer use it. | |
1288 | * | |
1289 | * As a last resort, if another node keeps holding the iopen glock | |
1290 | * without showing any activity on the inode glock, we will eventually | |
1291 | * time out and fail the iopen glock upgrade. | |
9e73330f AG |
1292 | */ |
1293 | ||
9e73330f AG |
1294 | gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh); |
1295 | error = gfs2_glock_nq(gh); | |
1296 | if (error) | |
1297 | return false; | |
1298 | ||
1299 | timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait, | |
9e8990de AG |
1300 | !test_bit(HIF_WAIT, &gh->gh_iflags) || |
1301 | test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags), | |
9e73330f AG |
1302 | timeout); |
1303 | if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) { | |
1304 | gfs2_glock_dq(gh); | |
1305 | return false; | |
1306 | } | |
53d69132 | 1307 | return gfs2_glock_holder_ready(gh) == 0; |
9e73330f AG |
1308 | } |
1309 | ||
53dbc27e BP |
1310 | /** |
1311 | * evict_should_delete - determine whether the inode is eligible for deletion | |
1312 | * @inode: The inode to evict | |
c551f66c | 1313 | * @gh: The glock holder structure |
53dbc27e BP |
1314 | * |
1315 | * This function determines whether the evicted inode is eligible to be deleted | |
1316 | * and locks the inode glock. | |
1317 | * | |
1318 | * Returns: the fate of the dinode | |
1319 | */ | |
1320 | static enum dinode_demise evict_should_delete(struct inode *inode, | |
1321 | struct gfs2_holder *gh) | |
1322 | { | |
1323 | struct gfs2_inode *ip = GFS2_I(inode); | |
1324 | struct super_block *sb = inode->i_sb; | |
1325 | struct gfs2_sbd *sdp = sb->s_fs_info; | |
1326 | int ret; | |
1327 | ||
38552ff6 | 1328 | if (unlikely(test_bit(GIF_ALLOC_FAILED, &ip->i_flags))) |
53dbc27e | 1329 | goto should_delete; |
53dbc27e BP |
1330 | |
1331 | if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags)) | |
1332 | return SHOULD_DEFER_EVICTION; | |
1333 | ||
1334 | /* Deletes should never happen under memory pressure anymore. */ | |
1335 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC)) | |
1336 | return SHOULD_DEFER_EVICTION; | |
1337 | ||
1338 | /* Must not read inode block until block type has been verified */ | |
1339 | ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh); | |
1340 | if (unlikely(ret)) { | |
1341 | glock_clear_object(ip->i_iopen_gh.gh_gl, ip); | |
1342 | ip->i_iopen_gh.gh_flags |= GL_NOCACHE; | |
1343 | gfs2_glock_dq_uninit(&ip->i_iopen_gh); | |
1344 | return SHOULD_DEFER_EVICTION; | |
1345 | } | |
1346 | ||
1347 | if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino)) | |
1348 | return SHOULD_NOT_DELETE_DINODE; | |
1349 | ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED); | |
1350 | if (ret) | |
1351 | return SHOULD_NOT_DELETE_DINODE; | |
1352 | ||
8d567162 AG |
1353 | ret = gfs2_instantiate(gh); |
1354 | if (ret) | |
1355 | return SHOULD_NOT_DELETE_DINODE; | |
53dbc27e BP |
1356 | |
1357 | /* | |
1358 | * The inode may have been recreated in the meantime. | |
1359 | */ | |
1360 | if (inode->i_nlink) | |
1361 | return SHOULD_NOT_DELETE_DINODE; | |
1362 | ||
1363 | should_delete: | |
1364 | if (gfs2_holder_initialized(&ip->i_iopen_gh) && | |
1365 | test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) { | |
1366 | if (!gfs2_upgrade_iopen_glock(inode)) { | |
1367 | gfs2_holder_uninit(&ip->i_iopen_gh); | |
1368 | return SHOULD_NOT_DELETE_DINODE; | |
1369 | } | |
1370 | } | |
1371 | return SHOULD_DELETE_DINODE; | |
1372 | } | |
1373 | ||
6e7e9a50 BP |
1374 | /** |
1375 | * evict_unlinked_inode - delete the pieces of an unlinked evicted inode | |
1376 | * @inode: The inode to evict | |
1377 | */ | |
1378 | static int evict_unlinked_inode(struct inode *inode) | |
1379 | { | |
1380 | struct gfs2_inode *ip = GFS2_I(inode); | |
1381 | int ret; | |
1382 | ||
1383 | if (S_ISDIR(inode->i_mode) && | |
1384 | (ip->i_diskflags & GFS2_DIF_EXHASH)) { | |
1385 | ret = gfs2_dir_exhash_dealloc(ip); | |
1386 | if (ret) | |
1387 | goto out; | |
1388 | } | |
1389 | ||
1390 | if (ip->i_eattr) { | |
1391 | ret = gfs2_ea_dealloc(ip); | |
1392 | if (ret) | |
1393 | goto out; | |
1394 | } | |
1395 | ||
1396 | if (!gfs2_is_stuffed(ip)) { | |
1397 | ret = gfs2_file_dealloc(ip); | |
1398 | if (ret) | |
1399 | goto out; | |
1400 | } | |
1401 | ||
764665c6 AG |
1402 | /* |
1403 | * As soon as we clear the bitmap for the dinode, gfs2_create_inode() | |
1404 | * can get called to recreate it, or even gfs2_inode_lookup() if the | |
1405 | * inode was recreated on another node in the meantime. | |
1406 | * | |
1407 | * However, inserting the new inode into the inode hash table will not | |
1408 | * succeed until the old inode is removed, and that only happens after | |
1409 | * ->evict_inode() returns. The new inode is attached to its inode and | |
1410 | * iopen glocks after inserting it into the inode hash table, so at | |
1411 | * that point we can be sure that both glocks are unused. | |
1412 | */ | |
1413 | ||
6e7e9a50 | 1414 | ret = gfs2_dinode_dealloc(ip); |
f9da18cd BP |
1415 | if (!ret && ip->i_gl) |
1416 | gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino); | |
1417 | ||
6e7e9a50 BP |
1418 | out: |
1419 | return ret; | |
1420 | } | |
1421 | ||
d90be6ab BP |
1422 | /* |
1423 | * evict_linked_inode - evict an inode whose dinode has not been unlinked | |
1424 | * @inode: The inode to evict | |
1425 | */ | |
1426 | static int evict_linked_inode(struct inode *inode) | |
1427 | { | |
1428 | struct super_block *sb = inode->i_sb; | |
1429 | struct gfs2_sbd *sdp = sb->s_fs_info; | |
1430 | struct gfs2_inode *ip = GFS2_I(inode); | |
1431 | struct address_space *metamapping; | |
1432 | int ret; | |
1433 | ||
1434 | gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL | | |
1435 | GFS2_LFC_EVICT_INODE); | |
1436 | metamapping = gfs2_glock2aspace(ip->i_gl); | |
1437 | if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) { | |
1438 | filemap_fdatawrite(metamapping); | |
1439 | filemap_fdatawait(metamapping); | |
1440 | } | |
1441 | write_inode_now(inode, 1); | |
1442 | gfs2_ail_flush(ip->i_gl, 0); | |
1443 | ||
1444 | ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks); | |
1445 | if (ret) | |
1446 | return ret; | |
1447 | ||
1448 | /* Needs to be done before glock release & also in a transaction */ | |
1449 | truncate_inode_pages(&inode->i_data, 0); | |
1450 | truncate_inode_pages(metamapping, 0); | |
1451 | gfs2_trans_end(sdp); | |
1452 | return 0; | |
1453 | } | |
1454 | ||
380f7c65 SW |
1455 | /** |
1456 | * gfs2_evict_inode - Remove an inode from cache | |
1457 | * @inode: The inode to evict | |
1458 | * | |
1459 | * There are three cases to consider: | |
1460 | * 1. i_nlink == 0, we are final opener (and must deallocate) | |
1461 | * 2. i_nlink == 0, we are not the final opener (and cannot deallocate) | |
1462 | * 3. i_nlink > 0 | |
1463 | * | |
1464 | * If the fs is read only, then we have to treat all cases as per #3 | |
1465 | * since we are unable to do any deallocation. The inode will be | |
1466 | * deallocated by the next read/write node to attempt an allocation | |
1467 | * in the same resource group | |
1468 | * | |
9e6e0a12 SW |
1469 | * We have to (at the moment) hold the inodes main lock to cover |
1470 | * the gap between unlocking the shared lock on the iopen lock and | |
1471 | * taking the exclusive lock. I'd rather do a shared -> exclusive | |
1472 | * conversion on the iopen lock, but we can change that later. This | |
1473 | * is safe, just less efficient. | |
1474 | */ | |
1475 | ||
d5c1515c | 1476 | static void gfs2_evict_inode(struct inode *inode) |
9e6e0a12 | 1477 | { |
001e8e8d SW |
1478 | struct super_block *sb = inode->i_sb; |
1479 | struct gfs2_sbd *sdp = sb->s_fs_info; | |
9e6e0a12 SW |
1480 | struct gfs2_inode *ip = GFS2_I(inode); |
1481 | struct gfs2_holder gh; | |
23d828fc | 1482 | int ret; |
9e6e0a12 | 1483 | |
38552ff6 | 1484 | if (inode->i_nlink || sb_rdonly(sb) || !ip->i_no_addr) |
d5c1515c AV |
1485 | goto out; |
1486 | ||
504a10d9 BP |
1487 | /* |
1488 | * In case of an incomplete mount, gfs2_evict_inode() may be called for | |
1489 | * system files without having an active journal to write to. In that | |
1490 | * case, skip the filesystem evict. | |
1491 | */ | |
1492 | if (!sdp->sd_jdesc) | |
1493 | goto out; | |
1494 | ||
53dbc27e BP |
1495 | gfs2_holder_mark_uninitialized(&gh); |
1496 | ret = evict_should_delete(inode, &gh); | |
1497 | if (ret == SHOULD_DEFER_EVICTION) | |
8c7b9262 | 1498 | goto out; |
0a0d9f55 BP |
1499 | if (ret == SHOULD_DELETE_DINODE) |
1500 | ret = evict_unlinked_inode(inode); | |
1501 | else | |
1502 | ret = evict_linked_inode(inode); | |
9e6e0a12 | 1503 | |
a097dc7e BP |
1504 | if (gfs2_rs_active(&ip->i_res)) |
1505 | gfs2_rs_deltree(&ip->i_res); | |
8e2e0047 | 1506 | |
9ffa1888 | 1507 | if (gfs2_holder_initialized(&gh)) |
e0b62e21 | 1508 | gfs2_glock_dq_uninit(&gh); |
23d828fc BP |
1509 | if (ret && ret != GLR_TRYFAILED && ret != -EROFS) |
1510 | fs_warn(sdp, "gfs2_evict_inode: %d\n", ret); | |
9e6e0a12 | 1511 | out: |
91b0abe3 | 1512 | truncate_inode_pages_final(&inode->i_data); |
2fba46a0 BP |
1513 | if (ip->i_qadata) |
1514 | gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0); | |
7336905a | 1515 | gfs2_rs_deltree(&ip->i_res); |
45138990 | 1516 | gfs2_ordered_del_inode(ip); |
dbd5768f | 1517 | clear_inode(inode); |
17d539f0 | 1518 | gfs2_dir_hash_inval(ip); |
6df9f9a2 | 1519 | if (gfs2_holder_initialized(&ip->i_iopen_gh)) { |
71c1b213 AG |
1520 | struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl; |
1521 | ||
1522 | glock_clear_object(gl, ip); | |
71c1b213 | 1523 | gfs2_glock_hold(gl); |
fe1bff65 AG |
1524 | ip->i_iopen_gh.gh_flags |= GL_NOCACHE; |
1525 | gfs2_glock_dq_uninit(&ip->i_iopen_gh); | |
71c1b213 | 1526 | gfs2_glock_put_eventually(gl); |
d5c1515c | 1527 | } |
49462e2b BP |
1528 | if (ip->i_gl) { |
1529 | glock_clear_object(ip->i_gl, ip); | |
1530 | wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE); | |
1531 | gfs2_glock_add_to_lru(ip->i_gl); | |
1532 | gfs2_glock_put_eventually(ip->i_gl); | |
0abd1557 | 1533 | rcu_assign_pointer(ip->i_gl, NULL); |
49462e2b | 1534 | } |
9e6e0a12 SW |
1535 | } |
1536 | ||
1537 | static struct inode *gfs2_alloc_inode(struct super_block *sb) | |
1538 | { | |
1539 | struct gfs2_inode *ip; | |
1540 | ||
fd60b288 | 1541 | ip = alloc_inode_sb(sb, gfs2_inode_cachep, GFP_KERNEL); |
d4031259 AG |
1542 | if (!ip) |
1543 | return NULL; | |
38552ff6 | 1544 | ip->i_no_addr = 0; |
d4031259 AG |
1545 | ip->i_flags = 0; |
1546 | ip->i_gl = NULL; | |
40e7e86e | 1547 | gfs2_holder_mark_uninitialized(&ip->i_iopen_gh); |
d4031259 AG |
1548 | memset(&ip->i_res, 0, sizeof(ip->i_res)); |
1549 | RB_CLEAR_NODE(&ip->i_res.rs_node); | |
1550 | ip->i_rahead = 0; | |
9e6e0a12 SW |
1551 | return &ip->i_inode; |
1552 | } | |
1553 | ||
784494e1 | 1554 | static void gfs2_free_inode(struct inode *inode) |
9e6e0a12 | 1555 | { |
784494e1 | 1556 | kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode)); |
fa0d7e3d NP |
1557 | } |
1558 | ||
0b2355fe | 1559 | void free_local_statfs_inodes(struct gfs2_sbd *sdp) |
97fd734b AD |
1560 | { |
1561 | struct local_statfs_inode *lsi, *safe; | |
1562 | ||
1563 | /* Run through the statfs inodes list to iput and free memory */ | |
1564 | list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) { | |
1565 | if (lsi->si_jid == sdp->sd_jdesc->jd_jid) | |
1566 | sdp->sd_sc_inode = NULL; /* belongs to this node */ | |
1567 | if (lsi->si_sc_inode) | |
1568 | iput(lsi->si_sc_inode); | |
1569 | list_del(&lsi->si_list); | |
1570 | kfree(lsi); | |
1571 | } | |
1572 | } | |
1573 | ||
0b2355fe AG |
1574 | struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp, |
1575 | unsigned int index) | |
97fd734b AD |
1576 | { |
1577 | struct local_statfs_inode *lsi; | |
1578 | ||
1579 | /* Return the local (per node) statfs inode in the | |
1580 | * sdp->sd_sc_inodes_list corresponding to the 'index'. */ | |
1581 | list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) { | |
1582 | if (lsi->si_jid == index) | |
1583 | return lsi->si_sc_inode; | |
1584 | } | |
1585 | return NULL; | |
1586 | } | |
1587 | ||
9e6e0a12 SW |
1588 | const struct super_operations gfs2_super_ops = { |
1589 | .alloc_inode = gfs2_alloc_inode, | |
784494e1 | 1590 | .free_inode = gfs2_free_inode, |
9e6e0a12 | 1591 | .write_inode = gfs2_write_inode, |
ab9bbda0 | 1592 | .dirty_inode = gfs2_dirty_inode, |
d5c1515c | 1593 | .evict_inode = gfs2_evict_inode, |
9e6e0a12 | 1594 | .put_super = gfs2_put_super, |
9e6e0a12 | 1595 | .sync_fs = gfs2_sync_fs, |
097cca52 | 1596 | .freeze_super = gfs2_freeze_super, |
e345b87b | 1597 | .freeze_fs = gfs2_freeze_fs, |
097cca52 | 1598 | .thaw_super = gfs2_thaw_super, |
9e6e0a12 | 1599 | .statfs = gfs2_statfs, |
9e6e0a12 SW |
1600 | .drop_inode = gfs2_drop_inode, |
1601 | .show_options = gfs2_show_options, | |
1602 | }; | |
1603 |