?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[GFS2] Shrink gfs2_inode (7) - di_payload_format
[linux-2.6.git] / fs / gfs2 / inode.c
blobf6177fc683200d6faa959ba7789ab266591d683e
1 /*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
9
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/posix_acl.h>
16 #include <linux/sort.h>
17 #include <linux/gfs2_ondisk.h>
18 #include <linux/crc32.h>
19 #include <linux/lm_interface.h>
20 #include <linux/security.h>
21
22 #include "gfs2.h"
23 #include "incore.h"
24 #include "acl.h"
25 #include "bmap.h"
26 #include "dir.h"
27 #include "eattr.h"
28 #include "glock.h"
29 #include "glops.h"
30 #include "inode.h"
31 #include "log.h"
32 #include "meta_io.h"
33 #include "ops_address.h"
34 #include "ops_file.h"
35 #include "ops_inode.h"
36 #include "quota.h"
37 #include "rgrp.h"
38 #include "trans.h"
39 #include "util.h"
40
41 /**
42 * gfs2_inode_attr_in - Copy attributes from the dinode into the VFS inode
43 * @ip: The GFS2 inode (with embedded disk inode data)
44 * @inode: The Linux VFS inode
45 *
46 */
47
48 void gfs2_inode_attr_in(struct gfs2_inode *ip)
49 {
50 struct inode *inode = &ip->i_inode;
51 struct gfs2_dinode_host *di = &ip->i_di;
52
53 inode->i_ino = ip->i_num.no_addr;
54 i_size_write(inode, di->di_size);
55 inode->i_blocks = di->di_blocks <<
56 (GFS2_SB(inode)->sd_sb.sb_bsize_shift - GFS2_BASIC_BLOCK_SHIFT);
57
58 if (di->di_flags & GFS2_DIF_IMMUTABLE)
59 inode->i_flags |= S_IMMUTABLE;
60 else
61 inode->i_flags &= ~S_IMMUTABLE;
62
63 if (di->di_flags & GFS2_DIF_APPENDONLY)
64 inode->i_flags |= S_APPEND;
65 else
66 inode->i_flags &= ~S_APPEND;
67 }
68
69 static int iget_test(struct inode *inode, void *opaque)
70 {
71 struct gfs2_inode *ip = GFS2_I(inode);
72 struct gfs2_inum_host *inum = opaque;
73
74 if (ip && ip->i_num.no_addr == inum->no_addr)
75 return 1;
76
77 return 0;
78 }
79
80 static int iget_set(struct inode *inode, void *opaque)
81 {
82 struct gfs2_inode *ip = GFS2_I(inode);
83 struct gfs2_inum_host *inum = opaque;
84
85 ip->i_num = *inum;
86 return 0;
87 }
88
89 struct inode *gfs2_ilookup(struct super_block *sb, struct gfs2_inum_host *inum)
90 {
91 return ilookup5(sb, (unsigned long)inum->no_formal_ino,
92 iget_test, inum);
93 }
94
95 static struct inode *gfs2_iget(struct super_block *sb, struct gfs2_inum_host *inum)
96 {
97 return iget5_locked(sb, (unsigned long)inum->no_formal_ino,
98 iget_test, iget_set, inum);
99 }
100
101 /**
102 * gfs2_inode_lookup - Lookup an inode
103 * @sb: The super block
104 * @inum: The inode number
105 * @type: The type of the inode
106 *
107 * Returns: A VFS inode, or an error
108 */
109
110 struct inode *gfs2_inode_lookup(struct super_block *sb, struct gfs2_inum_host *inum, unsigned int type)
111 {
112 struct inode *inode = gfs2_iget(sb, inum);
113 struct gfs2_inode *ip = GFS2_I(inode);
114 struct gfs2_glock *io_gl;
115 int error;
116
117 if (!inode)
118 return ERR_PTR(-ENOBUFS);
119
120 if (inode->i_state & I_NEW) {
121 struct gfs2_sbd *sdp = GFS2_SB(inode);
122 umode_t mode = DT2IF(type);
123 inode->i_private = ip;
124 inode->i_mode = mode;
125
126 if (S_ISREG(mode)) {
127 inode->i_op = &gfs2_file_iops;
128 inode->i_fop = &gfs2_file_fops;
129 inode->i_mapping->a_ops = &gfs2_file_aops;
130 } else if (S_ISDIR(mode)) {
131 inode->i_op = &gfs2_dir_iops;
132 inode->i_fop = &gfs2_dir_fops;
133 } else if (S_ISLNK(mode)) {
134 inode->i_op = &gfs2_symlink_iops;
135 } else {
136 inode->i_op = &gfs2_dev_iops;
137 }
138
139 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl);
140 if (unlikely(error))
141 goto fail;
142 ip->i_gl->gl_object = ip;
143
144 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_iopen_glops, CREATE, &io_gl);
145 if (unlikely(error))
146 goto fail_put;
147
148 ip->i_vn = ip->i_gl->gl_vn - 1;
149 error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh);
150 if (unlikely(error))
151 goto fail_iopen;
152
153 gfs2_glock_put(io_gl);
154 unlock_new_inode(inode);
155 }
156
157 return inode;
158 fail_iopen:
159 gfs2_glock_put(io_gl);
160 fail_put:
161 ip->i_gl->gl_object = NULL;
162 gfs2_glock_put(ip->i_gl);
163 fail:
164 iput(inode);
165 return ERR_PTR(error);
166 }
167
168 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
169 {
170 struct gfs2_dinode_host *di = &ip->i_di;
171 const struct gfs2_dinode *str = buf;
172
173 if (ip->i_num.no_addr != be64_to_cpu(str->di_num.no_addr)) {
174 if (gfs2_consist_inode(ip))
175 gfs2_dinode_print(ip);
176 return -EIO;
177 }
178 if (ip->i_num.no_formal_ino != be64_to_cpu(str->di_num.no_formal_ino))
179 return -ESTALE;
180
181 ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
182 ip->i_inode.i_rdev = 0;
183 switch (ip->i_inode.i_mode & S_IFMT) {
184 case S_IFBLK:
185 case S_IFCHR:
186 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
187 be32_to_cpu(str->di_minor));
188 break;
189 };
190
191 ip->i_inode.i_uid = be32_to_cpu(str->di_uid);
192 ip->i_inode.i_gid = be32_to_cpu(str->di_gid);
193 /*
194 * We will need to review setting the nlink count here in the
195 * light of the forthcoming ro bind mount work. This is a reminder
196 * to do that.
197 */
198 ip->i_inode.i_nlink = be32_to_cpu(str->di_nlink);
199 di->di_size = be64_to_cpu(str->di_size);
200 di->di_blocks = be64_to_cpu(str->di_blocks);
201 ip->i_inode.i_atime.tv_sec = be64_to_cpu(str->di_atime);
202 ip->i_inode.i_atime.tv_nsec = 0;
203 ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
204 ip->i_inode.i_mtime.tv_nsec = 0;
205 ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
206 ip->i_inode.i_ctime.tv_nsec = 0;
207
208 di->di_goal_meta = be64_to_cpu(str->di_goal_meta);
209 di->di_goal_data = be64_to_cpu(str->di_goal_data);
210 di->di_generation = be64_to_cpu(str->di_generation);
211
212 di->di_flags = be32_to_cpu(str->di_flags);
213 di->di_height = be16_to_cpu(str->di_height);
214
215 di->di_depth = be16_to_cpu(str->di_depth);
216 di->di_entries = be32_to_cpu(str->di_entries);
217
218 di->di_eattr = be64_to_cpu(str->di_eattr);
219 return 0;
220 }
221
222 /**
223 * gfs2_inode_refresh - Refresh the incore copy of the dinode
224 * @ip: The GFS2 inode
225 *
226 * Returns: errno
227 */
228
229 int gfs2_inode_refresh(struct gfs2_inode *ip)
230 {
231 struct buffer_head *dibh;
232 int error;
233
234 error = gfs2_meta_inode_buffer(ip, &dibh);
235 if (error)
236 return error;
237
238 if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), dibh, GFS2_METATYPE_DI)) {
239 brelse(dibh);
240 return -EIO;
241 }
242
243 error = gfs2_dinode_in(ip, dibh->b_data);
244 brelse(dibh);
245 ip->i_vn = ip->i_gl->gl_vn;
246
247 return error;
248 }
249
250 int gfs2_dinode_dealloc(struct gfs2_inode *ip)
251 {
252 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
253 struct gfs2_alloc *al;
254 struct gfs2_rgrpd *rgd;
255 int error;
256
257 if (ip->i_di.di_blocks != 1) {
258 if (gfs2_consist_inode(ip))
259 gfs2_dinode_print(ip);
260 return -EIO;
261 }
262
263 al = gfs2_alloc_get(ip);
264
265 error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
266 if (error)
267 goto out;
268
269 error = gfs2_rindex_hold(sdp, &al->al_ri_gh);
270 if (error)
271 goto out_qs;
272
273 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr);
274 if (!rgd) {
275 gfs2_consist_inode(ip);
276 error = -EIO;
277 goto out_rindex_relse;
278 }
279
280 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0,
281 &al->al_rgd_gh);
282 if (error)
283 goto out_rindex_relse;
284
285 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA, 1);
286 if (error)
287 goto out_rg_gunlock;
288
289 gfs2_trans_add_gl(ip->i_gl);
290
291 gfs2_free_di(rgd, ip);
292
293 gfs2_trans_end(sdp);
294 clear_bit(GLF_STICKY, &ip->i_gl->gl_flags);
295
296 out_rg_gunlock:
297 gfs2_glock_dq_uninit(&al->al_rgd_gh);
298 out_rindex_relse:
299 gfs2_glock_dq_uninit(&al->al_ri_gh);
300 out_qs:
301 gfs2_quota_unhold(ip);
302 out:
303 gfs2_alloc_put(ip);
304 return error;
305 }
306
307 /**
308 * gfs2_change_nlink - Change nlink count on inode
309 * @ip: The GFS2 inode
310 * @diff: The change in the nlink count required
311 *
312 * Returns: errno
313 */
314
315 int gfs2_change_nlink(struct gfs2_inode *ip, int diff)
316 {
317 struct gfs2_sbd *sdp = ip->i_inode.i_sb->s_fs_info;
318 struct buffer_head *dibh;
319 u32 nlink;
320 int error;
321
322 BUG_ON(diff != 1 && diff != -1);
323 nlink = ip->i_inode.i_nlink + diff;
324
325 /* If we are reducing the nlink count, but the new value ends up being
326 bigger than the old one, we must have underflowed. */
327 if (diff < 0 && nlink > ip->i_inode.i_nlink) {
328 if (gfs2_consist_inode(ip))
329 gfs2_dinode_print(ip);
330 return -EIO;
331 }
332
333 error = gfs2_meta_inode_buffer(ip, &dibh);
334 if (error)
335 return error;
336
337 if (diff > 0)
338 inc_nlink(&ip->i_inode);
339 else
340 drop_nlink(&ip->i_inode);
341
342 ip->i_inode.i_ctime.tv_sec = get_seconds();
343
344 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
345 gfs2_dinode_out(ip, dibh->b_data);
346 brelse(dibh);
347 mark_inode_dirty(&ip->i_inode);
348
349 if (ip->i_inode.i_nlink == 0) {
350 struct gfs2_rgrpd *rgd;
351 struct gfs2_holder ri_gh, rg_gh;
352
353 error = gfs2_rindex_hold(sdp, &ri_gh);
354 if (error)
355 goto out;
356 error = -EIO;
357 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr);
358 if (!rgd)
359 goto out_norgrp;
360 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &rg_gh);
361 if (error)
362 goto out_norgrp;
363
364 gfs2_unlink_di(&ip->i_inode); /* mark inode unlinked */
365 gfs2_glock_dq_uninit(&rg_gh);
366 out_norgrp:
367 gfs2_glock_dq_uninit(&ri_gh);
368 }
369 out:
370 return error;
371 }
372
373 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name)
374 {
375 struct qstr qstr;
376 gfs2_str2qstr(&qstr, name);
377 return gfs2_lookupi(dip, &qstr, 1, NULL);
378 }
379
380
381 /**
382 * gfs2_lookupi - Look up a filename in a directory and return its inode
383 * @d_gh: An initialized holder for the directory glock
384 * @name: The name of the inode to look for
385 * @is_root: If 1, ignore the caller's permissions
386 * @i_gh: An uninitialized holder for the new inode glock
387 *
388 * There will always be a vnode (Linux VFS inode) for the d_gh inode unless
389 * @is_root is true.
390 *
391 * Returns: errno
392 */
393
394 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name,
395 int is_root, struct nameidata *nd)
396 {
397 struct super_block *sb = dir->i_sb;
398 struct gfs2_inode *dip = GFS2_I(dir);
399 struct gfs2_holder d_gh;
400 struct gfs2_inum_host inum;
401 unsigned int type;
402 int error = 0;
403 struct inode *inode = NULL;
404
405 if (!name->len || name->len > GFS2_FNAMESIZE)
406 return ERR_PTR(-ENAMETOOLONG);
407
408 if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) ||
409 (name->len == 2 && memcmp(name->name, "..", 2) == 0 &&
410 dir == sb->s_root->d_inode)) {
411 igrab(dir);
412 return dir;
413 }
414
415 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
416 if (error)
417 return ERR_PTR(error);
418
419 if (!is_root) {
420 error = permission(dir, MAY_EXEC, NULL);
421 if (error)
422 goto out;
423 }
424
425 error = gfs2_dir_search(dir, name, &inum, &type);
426 if (error)
427 goto out;
428
429 inode = gfs2_inode_lookup(sb, &inum, type);
430
431 out:
432 gfs2_glock_dq_uninit(&d_gh);
433 if (error == -ENOENT)
434 return NULL;
435 return inode;
436 }
437
438 static int pick_formal_ino_1(struct gfs2_sbd *sdp, u64 *formal_ino)
439 {
440 struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode);
441 struct buffer_head *bh;
442 struct gfs2_inum_range_host ir;
443 int error;
444
445 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
446 if (error)
447 return error;
448 mutex_lock(&sdp->sd_inum_mutex);
449
450 error = gfs2_meta_inode_buffer(ip, &bh);
451 if (error) {
452 mutex_unlock(&sdp->sd_inum_mutex);
453 gfs2_trans_end(sdp);
454 return error;
455 }
456
457 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
458
459 if (ir.ir_length) {
460 *formal_ino = ir.ir_start++;
461 ir.ir_length--;
462 gfs2_trans_add_bh(ip->i_gl, bh, 1);
463 gfs2_inum_range_out(&ir,
464 bh->b_data + sizeof(struct gfs2_dinode));
465 brelse(bh);
466 mutex_unlock(&sdp->sd_inum_mutex);
467 gfs2_trans_end(sdp);
468 return 0;
469 }
470
471 brelse(bh);
472
473 mutex_unlock(&sdp->sd_inum_mutex);
474 gfs2_trans_end(sdp);
475
476 return 1;
477 }
478
479 static int pick_formal_ino_2(struct gfs2_sbd *sdp, u64 *formal_ino)
480 {
481 struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode);
482 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_inum_inode);
483 struct gfs2_holder gh;
484 struct buffer_head *bh;
485 struct gfs2_inum_range_host ir;
486 int error;
487
488 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
489 if (error)
490 return error;
491
492 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
493 if (error)
494 goto out;
495 mutex_lock(&sdp->sd_inum_mutex);
496
497 error = gfs2_meta_inode_buffer(ip, &bh);
498 if (error)
499 goto out_end_trans;
500
501 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
502
503 if (!ir.ir_length) {
504 struct buffer_head *m_bh;
505 u64 x, y;
506 __be64 z;
507
508 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
509 if (error)
510 goto out_brelse;
511
512 z = *(__be64 *)(m_bh->b_data + sizeof(struct gfs2_dinode));
513 x = y = be64_to_cpu(z);
514 ir.ir_start = x;
515 ir.ir_length = GFS2_INUM_QUANTUM;
516 x += GFS2_INUM_QUANTUM;
517 if (x < y)
518 gfs2_consist_inode(m_ip);
519 z = cpu_to_be64(x);
520 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
521 *(__be64 *)(m_bh->b_data + sizeof(struct gfs2_dinode)) = z;
522
523 brelse(m_bh);
524 }
525
526 *formal_ino = ir.ir_start++;
527 ir.ir_length--;
528
529 gfs2_trans_add_bh(ip->i_gl, bh, 1);
530 gfs2_inum_range_out(&ir, bh->b_data + sizeof(struct gfs2_dinode));
531
532 out_brelse:
533 brelse(bh);
534 out_end_trans:
535 mutex_unlock(&sdp->sd_inum_mutex);
536 gfs2_trans_end(sdp);
537 out:
538 gfs2_glock_dq_uninit(&gh);
539 return error;
540 }
541
542 static int pick_formal_ino(struct gfs2_sbd *sdp, u64 *inum)
543 {
544 int error;
545
546 error = pick_formal_ino_1(sdp, inum);
547 if (error <= 0)
548 return error;
549
550 error = pick_formal_ino_2(sdp, inum);
551
552 return error;
553 }
554
555 /**
556 * create_ok - OK to create a new on-disk inode here?
557 * @dip: Directory in which dinode is to be created
558 * @name: Name of new dinode
559 * @mode:
560 *
561 * Returns: errno
562 */
563
564 static int create_ok(struct gfs2_inode *dip, const struct qstr *name,
565 unsigned int mode)
566 {
567 int error;
568
569 error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL);
570 if (error)
571 return error;
572
573 /* Don't create entries in an unlinked directory */
574 if (!dip->i_inode.i_nlink)
575 return -EPERM;
576
577 error = gfs2_dir_search(&dip->i_inode, name, NULL, NULL);
578 switch (error) {
579 case -ENOENT:
580 error = 0;
581 break;
582 case 0:
583 return -EEXIST;
584 default:
585 return error;
586 }
587
588 if (dip->i_di.di_entries == (u32)-1)
589 return -EFBIG;
590 if (S_ISDIR(mode) && dip->i_inode.i_nlink == (u32)-1)
591 return -EMLINK;
592
593 return 0;
594 }
595
596 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode,
597 unsigned int *uid, unsigned int *gid)
598 {
599 if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir &&
600 (dip->i_inode.i_mode & S_ISUID) && dip->i_inode.i_uid) {
601 if (S_ISDIR(*mode))
602 *mode |= S_ISUID;
603 else if (dip->i_inode.i_uid != current->fsuid)
604 *mode &= ~07111;
605 *uid = dip->i_inode.i_uid;
606 } else
607 *uid = current->fsuid;
608
609 if (dip->i_inode.i_mode & S_ISGID) {
610 if (S_ISDIR(*mode))
611 *mode |= S_ISGID;
612 *gid = dip->i_inode.i_gid;
613 } else
614 *gid = current->fsgid;
615 }
616
617 static int alloc_dinode(struct gfs2_inode *dip, struct gfs2_inum_host *inum,
618 u64 *generation)
619 {
620 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
621 int error;
622
623 gfs2_alloc_get(dip);
624
625 dip->i_alloc.al_requested = RES_DINODE;
626 error = gfs2_inplace_reserve(dip);
627 if (error)
628 goto out;
629
630 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS, 0);
631 if (error)
632 goto out_ipreserv;
633
634 inum->no_addr = gfs2_alloc_di(dip, generation);
635
636 gfs2_trans_end(sdp);
637
638 out_ipreserv:
639 gfs2_inplace_release(dip);
640 out:
641 gfs2_alloc_put(dip);
642 return error;
643 }
644
645 /**
646 * init_dinode - Fill in a new dinode structure
647 * @dip: the directory this inode is being created in
648 * @gl: The glock covering the new inode
649 * @inum: the inode number
650 * @mode: the file permissions
651 * @uid:
652 * @gid:
653 *
654 */
655
656 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
657 const struct gfs2_inum_host *inum, unsigned int mode,
658 unsigned int uid, unsigned int gid,
659 const u64 *generation, dev_t dev)
660 {
661 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
662 struct gfs2_dinode *di;
663 struct buffer_head *dibh;
664
665 dibh = gfs2_meta_new(gl, inum->no_addr);
666 gfs2_trans_add_bh(gl, dibh, 1);
667 gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI);
668 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
669 di = (struct gfs2_dinode *)dibh->b_data;
670
671 di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino);
672 di->di_num.no_addr = cpu_to_be64(inum->no_addr);
673 di->di_mode = cpu_to_be32(mode);
674 di->di_uid = cpu_to_be32(uid);
675 di->di_gid = cpu_to_be32(gid);
676 di->di_nlink = cpu_to_be32(0);
677 di->di_size = cpu_to_be64(0);
678 di->di_blocks = cpu_to_be64(1);
679 di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(get_seconds());
680 di->di_major = cpu_to_be32(MAJOR(dev));
681 di->di_minor = cpu_to_be32(MINOR(dev));
682 di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr);
683 di->di_generation = cpu_to_be64(*generation);
684 di->di_flags = cpu_to_be32(0);
685
686 if (S_ISREG(mode)) {
687 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_JDATA) ||
688 gfs2_tune_get(sdp, gt_new_files_jdata))
689 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
690 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_DIRECTIO) ||
691 gfs2_tune_get(sdp, gt_new_files_directio))
692 di->di_flags |= cpu_to_be32(GFS2_DIF_DIRECTIO);
693 } else if (S_ISDIR(mode)) {
694 di->di_flags |= cpu_to_be32(dip->i_di.di_flags &
695 GFS2_DIF_INHERIT_DIRECTIO);
696 di->di_flags |= cpu_to_be32(dip->i_di.di_flags &
697 GFS2_DIF_INHERIT_JDATA);
698 }
699
700 di->__pad1 = 0;
701 di->di_payload_format = cpu_to_be32(S_ISDIR(mode) ? GFS2_FORMAT_DE : 0);
702 di->di_height = cpu_to_be32(0);
703 di->__pad2 = 0;
704 di->__pad3 = 0;
705 di->di_depth = cpu_to_be16(0);
706 di->di_entries = cpu_to_be32(0);
707 memset(&di->__pad4, 0, sizeof(di->__pad4));
708 di->di_eattr = cpu_to_be64(0);
709 memset(&di->di_reserved, 0, sizeof(di->di_reserved));
710
711 brelse(dibh);
712 }
713
714 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
715 unsigned int mode, const struct gfs2_inum_host *inum,
716 const u64 *generation, dev_t dev)
717 {
718 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
719 unsigned int uid, gid;
720 int error;
721
722 munge_mode_uid_gid(dip, &mode, &uid, &gid);
723 gfs2_alloc_get(dip);
724
725 error = gfs2_quota_lock(dip, uid, gid);
726 if (error)
727 goto out;
728
729 error = gfs2_quota_check(dip, uid, gid);
730 if (error)
731 goto out_quota;
732
733 error = gfs2_trans_begin(sdp, RES_DINODE + RES_QUOTA, 0);
734 if (error)
735 goto out_quota;
736
737 init_dinode(dip, gl, inum, mode, uid, gid, generation, dev);
738 gfs2_quota_change(dip, +1, uid, gid);
739 gfs2_trans_end(sdp);
740
741 out_quota:
742 gfs2_quota_unlock(dip);
743 out:
744 gfs2_alloc_put(dip);
745 return error;
746 }
747
748 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name,
749 struct gfs2_inode *ip)
750 {
751 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
752 struct gfs2_alloc *al;
753 int alloc_required;
754 struct buffer_head *dibh;
755 int error;
756
757 al = gfs2_alloc_get(dip);
758
759 error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
760 if (error)
761 goto fail;
762
763 error = alloc_required = gfs2_diradd_alloc_required(&dip->i_inode, name);
764 if (alloc_required < 0)
765 goto fail;
766 if (alloc_required) {
767 error = gfs2_quota_check(dip, dip->i_inode.i_uid, dip->i_inode.i_gid);
768 if (error)
769 goto fail_quota_locks;
770
771 al->al_requested = sdp->sd_max_dirres;
772
773 error = gfs2_inplace_reserve(dip);
774 if (error)
775 goto fail_quota_locks;
776
777 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres +
778 al->al_rgd->rd_ri.ri_length +
779 2 * RES_DINODE +
780 RES_STATFS + RES_QUOTA, 0);
781 if (error)
782 goto fail_ipreserv;
783 } else {
784 error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0);
785 if (error)
786 goto fail_quota_locks;
787 }
788
789 error = gfs2_dir_add(&dip->i_inode, name, &ip->i_num, IF2DT(ip->i_inode.i_mode));
790 if (error)
791 goto fail_end_trans;
792
793 error = gfs2_meta_inode_buffer(ip, &dibh);
794 if (error)
795 goto fail_end_trans;
796 ip->i_inode.i_nlink = 1;
797 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
798 gfs2_dinode_out(ip, dibh->b_data);
799 brelse(dibh);
800 return 0;
801
802 fail_end_trans:
803 gfs2_trans_end(sdp);
804
805 fail_ipreserv:
806 if (dip->i_alloc.al_rgd)
807 gfs2_inplace_release(dip);
808
809 fail_quota_locks:
810 gfs2_quota_unlock(dip);
811
812 fail:
813 gfs2_alloc_put(dip);
814 return error;
815 }
816
817 static int gfs2_security_init(struct gfs2_inode *dip, struct gfs2_inode *ip)
818 {
819 int err;
820 size_t len;
821 void *value;
822 char *name;
823 struct gfs2_ea_request er;
824
825 err = security_inode_init_security(&ip->i_inode, &dip->i_inode,
826 &name, &value, &len);
827
828 if (err) {
829 if (err == -EOPNOTSUPP)
830 return 0;
831 return err;
832 }
833
834 memset(&er, 0, sizeof(struct gfs2_ea_request));
835
836 er.er_type = GFS2_EATYPE_SECURITY;
837 er.er_name = name;
838 er.er_data = value;
839 er.er_name_len = strlen(name);
840 er.er_data_len = len;
841
842 err = gfs2_ea_set_i(ip, &er);
843
844 kfree(value);
845 kfree(name);
846
847 return err;
848 }
849
850 /**
851 * gfs2_createi - Create a new inode
852 * @ghs: An array of two holders
853 * @name: The name of the new file
854 * @mode: the permissions on the new inode
855 *
856 * @ghs[0] is an initialized holder for the directory
857 * @ghs[1] is the holder for the inode lock
858 *
859 * If the return value is not NULL, the glocks on both the directory and the new
860 * file are held. A transaction has been started and an inplace reservation
861 * is held, as well.
862 *
863 * Returns: An inode
864 */
865
866 struct inode *gfs2_createi(struct gfs2_holder *ghs, const struct qstr *name,
867 unsigned int mode, dev_t dev)
868 {
869 struct inode *inode;
870 struct gfs2_inode *dip = ghs->gh_gl->gl_object;
871 struct inode *dir = &dip->i_inode;
872 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
873 struct gfs2_inum_host inum;
874 int error;
875 u64 generation;
876
877 if (!name->len || name->len > GFS2_FNAMESIZE)
878 return ERR_PTR(-ENAMETOOLONG);
879
880 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs);
881 error = gfs2_glock_nq(ghs);
882 if (error)
883 goto fail;
884
885 error = create_ok(dip, name, mode);
886 if (error)
887 goto fail_gunlock;
888
889 error = pick_formal_ino(sdp, &inum.no_formal_ino);
890 if (error)
891 goto fail_gunlock;
892
893 error = alloc_dinode(dip, &inum, &generation);
894 if (error)
895 goto fail_gunlock;
896
897 if (inum.no_addr < dip->i_num.no_addr) {
898 gfs2_glock_dq(ghs);
899
900 error = gfs2_glock_nq_num(sdp, inum.no_addr,
901 &gfs2_inode_glops, LM_ST_EXCLUSIVE,
902 GL_SKIP, ghs + 1);
903 if (error) {
904 return ERR_PTR(error);
905 }
906
907 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs);
908 error = gfs2_glock_nq(ghs);
909 if (error) {
910 gfs2_glock_dq_uninit(ghs + 1);
911 return ERR_PTR(error);
912 }
913
914 error = create_ok(dip, name, mode);
915 if (error)
916 goto fail_gunlock2;
917 } else {
918 error = gfs2_glock_nq_num(sdp, inum.no_addr,
919 &gfs2_inode_glops, LM_ST_EXCLUSIVE,
920 GL_SKIP, ghs + 1);
921 if (error)
922 goto fail_gunlock;
923 }
924
925 error = make_dinode(dip, ghs[1].gh_gl, mode, &inum, &generation, dev);
926 if (error)
927 goto fail_gunlock2;
928
929 inode = gfs2_inode_lookup(dir->i_sb, &inum, IF2DT(mode));
930 if (IS_ERR(inode))
931 goto fail_gunlock2;
932
933 error = gfs2_inode_refresh(GFS2_I(inode));
934 if (error)
935 goto fail_iput;
936
937 error = gfs2_acl_create(dip, GFS2_I(inode));
938 if (error)
939 goto fail_iput;
940
941 error = gfs2_security_init(dip, GFS2_I(inode));
942 if (error)
943 goto fail_iput;
944
945 error = link_dinode(dip, name, GFS2_I(inode));
946 if (error)
947 goto fail_iput;
948
949 if (!inode)
950 return ERR_PTR(-ENOMEM);
951 return inode;
952
953 fail_iput:
954 iput(inode);
955 fail_gunlock2:
956 gfs2_glock_dq_uninit(ghs + 1);
957 fail_gunlock:
958 gfs2_glock_dq(ghs);
959 fail:
960 return ERR_PTR(error);
961 }
962
963 /**
964 * gfs2_rmdiri - Remove a directory
965 * @dip: The parent directory of the directory to be removed
966 * @name: The name of the directory to be removed
967 * @ip: The GFS2 inode of the directory to be removed
968 *
969 * Assumes Glocks on dip and ip are held
970 *
971 * Returns: errno
972 */
973
974 int gfs2_rmdiri(struct gfs2_inode *dip, const struct qstr *name,
975 struct gfs2_inode *ip)
976 {
977 struct qstr dotname;
978 int error;
979
980 if (ip->i_di.di_entries != 2) {
981 if (gfs2_consist_inode(ip))
982 gfs2_dinode_print(ip);
983 return -EIO;
984 }
985
986 error = gfs2_dir_del(dip, name);
987 if (error)
988 return error;
989
990 error = gfs2_change_nlink(dip, -1);
991 if (error)
992 return error;
993
994 gfs2_str2qstr(&dotname, ".");
995 error = gfs2_dir_del(ip, &dotname);
996 if (error)
997 return error;
998
999 gfs2_str2qstr(&dotname, "..");
1000 error = gfs2_dir_del(ip, &dotname);
1001 if (error)
1002 return error;
1003
1004 /* It looks odd, but it really should be done twice */
1005 error = gfs2_change_nlink(ip, -1);
1006 if (error)
1007 return error;
1008
1009 error = gfs2_change_nlink(ip, -1);
1010 if (error)
1011 return error;
1012
1013 return error;
1014 }
1015
1016 /*
1017 * gfs2_unlink_ok - check to see that a inode is still in a directory
1018 * @dip: the directory
1019 * @name: the name of the file
1020 * @ip: the inode
1021 *
1022 * Assumes that the lock on (at least) @dip is held.
1023 *
1024 * Returns: 0 if the parent/child relationship is correct, errno if it isn't
1025 */
1026
1027 int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name,
1028 struct gfs2_inode *ip)
1029 {
1030 struct gfs2_inum_host inum;
1031 unsigned int type;
1032 int error;
1033
1034 if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode))
1035 return -EPERM;
1036
1037 if ((dip->i_inode.i_mode & S_ISVTX) &&
1038 dip->i_inode.i_uid != current->fsuid &&
1039 ip->i_inode.i_uid != current->fsuid && !capable(CAP_FOWNER))
1040 return -EPERM;
1041
1042 if (IS_APPEND(&dip->i_inode))
1043 return -EPERM;
1044
1045 error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL);
1046 if (error)
1047 return error;
1048
1049 error = gfs2_dir_search(&dip->i_inode, name, &inum, &type);
1050 if (error)
1051 return error;
1052
1053 if (!gfs2_inum_equal(&inum, &ip->i_num))
1054 return -ENOENT;
1055
1056 if (IF2DT(ip->i_inode.i_mode) != type) {
1057 gfs2_consist_inode(dip);
1058 return -EIO;
1059 }
1060
1061 return 0;
1062 }
1063
1064 /*
1065 * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1066 * @this: move this
1067 * @to: to here
1068 *
1069 * Follow @to back to the root and make sure we don't encounter @this
1070 * Assumes we already hold the rename lock.
1071 *
1072 * Returns: errno
1073 */
1074
1075 int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1076 {
1077 struct inode *dir = &to->i_inode;
1078 struct super_block *sb = dir->i_sb;
1079 struct inode *tmp;
1080 struct qstr dotdot;
1081 int error = 0;
1082
1083 gfs2_str2qstr(&dotdot, "..");
1084
1085 igrab(dir);
1086
1087 for (;;) {
1088 if (dir == &this->i_inode) {
1089 error = -EINVAL;
1090 break;
1091 }
1092 if (dir == sb->s_root->d_inode) {
1093 error = 0;
1094 break;
1095 }
1096
1097 tmp = gfs2_lookupi(dir, &dotdot, 1, NULL);
1098 if (IS_ERR(tmp)) {
1099 error = PTR_ERR(tmp);
1100 break;
1101 }
1102
1103 iput(dir);
1104 dir = tmp;
1105 }
1106
1107 iput(dir);
1108
1109 return error;
1110 }
1111
1112 /**
1113 * gfs2_readlinki - return the contents of a symlink
1114 * @ip: the symlink's inode
1115 * @buf: a pointer to the buffer to be filled
1116 * @len: a pointer to the length of @buf
1117 *
1118 * If @buf is too small, a piece of memory is kmalloc()ed and needs
1119 * to be freed by the caller.
1120 *
1121 * Returns: errno
1122 */
1123
1124 int gfs2_readlinki(struct gfs2_inode *ip, char **buf, unsigned int *len)
1125 {
1126 struct gfs2_holder i_gh;
1127 struct buffer_head *dibh;
1128 unsigned int x;
1129 int error;
1130
1131 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh);
1132 error = gfs2_glock_nq_atime(&i_gh);
1133 if (error) {
1134 gfs2_holder_uninit(&i_gh);
1135 return error;
1136 }
1137
1138 if (!ip->i_di.di_size) {
1139 gfs2_consist_inode(ip);
1140 error = -EIO;
1141 goto out;
1142 }
1143
1144 error = gfs2_meta_inode_buffer(ip, &dibh);
1145 if (error)
1146 goto out;
1147
1148 x = ip->i_di.di_size + 1;
1149 if (x > *len) {
1150 *buf = kmalloc(x, GFP_KERNEL);
1151 if (!*buf) {
1152 error = -ENOMEM;
1153 goto out_brelse;
1154 }
1155 }
1156
1157 memcpy(*buf, dibh->b_data + sizeof(struct gfs2_dinode), x);
1158 *len = x;
1159
1160 out_brelse:
1161 brelse(dibh);
1162 out:
1163 gfs2_glock_dq_uninit(&i_gh);
1164 return error;
1165 }
1166
1167 /**
1168 * gfs2_glock_nq_atime - Acquire a hold on an inode's glock, and
1169 * conditionally update the inode's atime
1170 * @gh: the holder to acquire
1171 *
1172 * Tests atime (access time) for gfs2_read, gfs2_readdir and gfs2_mmap
1173 * Update if the difference between the current time and the inode's current
1174 * atime is greater than an interval specified at mount.
1175 *
1176 * Returns: errno
1177 */
1178
1179 int gfs2_glock_nq_atime(struct gfs2_holder *gh)
1180 {
1181 struct gfs2_glock *gl = gh->gh_gl;
1182 struct gfs2_sbd *sdp = gl->gl_sbd;
1183 struct gfs2_inode *ip = gl->gl_object;
1184 s64 curtime, quantum = gfs2_tune_get(sdp, gt_atime_quantum);
1185 unsigned int state;
1186 int flags;
1187 int error;
1188
1189 if (gfs2_assert_warn(sdp, gh->gh_flags & GL_ATIME) ||
1190 gfs2_assert_warn(sdp, !(gh->gh_flags & GL_ASYNC)) ||
1191 gfs2_assert_warn(sdp, gl->gl_ops == &gfs2_inode_glops))
1192 return -EINVAL;
1193
1194 state = gh->gh_state;
1195 flags = gh->gh_flags;
1196
1197 error = gfs2_glock_nq(gh);
1198 if (error)
1199 return error;
1200
1201 if (test_bit(SDF_NOATIME, &sdp->sd_flags) ||
1202 (sdp->sd_vfs->s_flags & MS_RDONLY))
1203 return 0;
1204
1205 curtime = get_seconds();
1206 if (curtime - ip->i_inode.i_atime.tv_sec >= quantum) {
1207 gfs2_glock_dq(gh);
1208 gfs2_holder_reinit(LM_ST_EXCLUSIVE, gh->gh_flags & ~LM_FLAG_ANY,
1209 gh);
1210 error = gfs2_glock_nq(gh);
1211 if (error)
1212 return error;
1213
1214 /* Verify that atime hasn't been updated while we were
1215 trying to get exclusive lock. */
1216
1217 curtime = get_seconds();
1218 if (curtime - ip->i_inode.i_atime.tv_sec >= quantum) {
1219 struct buffer_head *dibh;
1220 struct gfs2_dinode *di;
1221
1222 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1223 if (error == -EROFS)
1224 return 0;
1225 if (error)
1226 goto fail;
1227
1228 error = gfs2_meta_inode_buffer(ip, &dibh);
1229 if (error)
1230 goto fail_end_trans;
1231
1232 ip->i_inode.i_atime.tv_sec = curtime;
1233
1234 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1235 di = (struct gfs2_dinode *)dibh->b_data;
1236 di->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
1237 brelse(dibh);
1238
1239 gfs2_trans_end(sdp);
1240 }
1241
1242 /* If someone else has asked for the glock,
1243 unlock and let them have it. Then reacquire
1244 in the original state. */
1245 if (gfs2_glock_is_blocking(gl)) {
1246 gfs2_glock_dq(gh);
1247 gfs2_holder_reinit(state, flags, gh);
1248 return gfs2_glock_nq(gh);
1249 }
1250 }
1251
1252 return 0;
1253
1254 fail_end_trans:
1255 gfs2_trans_end(sdp);
1256 fail:
1257 gfs2_glock_dq(gh);
1258 return error;
1259 }
1260
1261 /**
1262 * glock_compare_atime - Compare two struct gfs2_glock structures for sort
1263 * @arg_a: the first structure
1264 * @arg_b: the second structure
1265 *
1266 * Returns: 1 if A > B
1267 * -1 if A < B
1268 * 0 if A == B
1269 */
1270
1271 static int glock_compare_atime(const void *arg_a, const void *arg_b)
1272 {
1273 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a;
1274 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b;
1275 const struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1276 const struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1277
1278 if (a->ln_number > b->ln_number)
1279 return 1;
1280 if (a->ln_number < b->ln_number)
1281 return -1;
1282 if (gh_a->gh_state == LM_ST_SHARED && gh_b->gh_state == LM_ST_EXCLUSIVE)
1283 return 1;
1284 if (gh_a->gh_state == LM_ST_SHARED && (gh_b->gh_flags & GL_ATIME))
1285 return 1;
1286
1287 return 0;
1288 }
1289
1290 /**
1291 * gfs2_glock_nq_m_atime - acquire multiple glocks where one may need an
1292 * atime update
1293 * @num_gh: the number of structures
1294 * @ghs: an array of struct gfs2_holder structures
1295 *
1296 * Returns: 0 on success (all glocks acquired),
1297 * errno on failure (no glocks acquired)
1298 */
1299
1300 int gfs2_glock_nq_m_atime(unsigned int num_gh, struct gfs2_holder *ghs)
1301 {
1302 struct gfs2_holder **p;
1303 unsigned int x;
1304 int error = 0;
1305
1306 if (!num_gh)
1307 return 0;
1308
1309 if (num_gh == 1) {
1310 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1311 if (ghs->gh_flags & GL_ATIME)
1312 error = gfs2_glock_nq_atime(ghs);
1313 else
1314 error = gfs2_glock_nq(ghs);
1315 return error;
1316 }
1317
1318 p = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL);
1319 if (!p)
1320 return -ENOMEM;
1321
1322 for (x = 0; x < num_gh; x++)
1323 p[x] = &ghs[x];
1324
1325 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare_atime,NULL);
1326
1327 for (x = 0; x < num_gh; x++) {
1328 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1329
1330 if (p[x]->gh_flags & GL_ATIME)
1331 error = gfs2_glock_nq_atime(p[x]);
1332 else
1333 error = gfs2_glock_nq(p[x]);
1334
1335 if (error) {
1336 while (x--)
1337 gfs2_glock_dq(p[x]);
1338 break;
1339 }
1340 }
1341
1342 kfree(p);
1343 return error;
1344 }
1345
1346
1347 static int
1348 __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1349 {
1350 struct buffer_head *dibh;
1351 int error;
1352
1353 error = gfs2_meta_inode_buffer(ip, &dibh);
1354 if (!error) {
1355 error = inode_setattr(&ip->i_inode, attr);
1356 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1357 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
1358 gfs2_dinode_out(ip, dibh->b_data);
1359 brelse(dibh);
1360 }
1361 return error;
1362 }
1363
1364 /**
1365 * gfs2_setattr_simple -
1366 * @ip:
1367 * @attr:
1368 *
1369 * Called with a reference on the vnode.
1370 *
1371 * Returns: errno
1372 */
1373
1374 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1375 {
1376 int error;
1377
1378 if (current->journal_info)
1379 return __gfs2_setattr_simple(ip, attr);
1380
1381 error = gfs2_trans_begin(GFS2_SB(&ip->i_inode), RES_DINODE, 0);
1382 if (error)
1383 return error;
1384
1385 error = __gfs2_setattr_simple(ip, attr);
1386 gfs2_trans_end(GFS2_SB(&ip->i_inode));
1387 return error;
1388 }
1389
Linus' kernel tree