?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge remote-tracking branch 'moduleh/module.h-split'
[linux-2.6/next.git] / fs / exofs / super.c
blob9049e27dcb07357fac3d5ddb97e32eafef45b452
1 /*
2 * Copyright (C) 2005, 2006
3 * Avishay Traeger (avishay@gmail.com)
4 * Copyright (C) 2008, 2009
5 * Boaz Harrosh <bharrosh@panasas.com>
6 *
7 * Copyrights for code taken from ext2:
8 * Copyright (C) 1992, 1993, 1994, 1995
9 * Remy Card (card@masi.ibp.fr)
10 * Laboratoire MASI - Institut Blaise Pascal
11 * Universite Pierre et Marie Curie (Paris VI)
12 * from
13 * linux/fs/minix/inode.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
15 *
16 * This file is part of exofs.
17 *
18 * exofs is free software; you can redistribute it and/or modify
19 * it under the terms of the GNU General Public License as published by
20 * the Free Software Foundation. Since it is based on ext2, and the only
21 * valid version of GPL for the Linux kernel is version 2, the only valid
22 * version of GPL for exofs is version 2.
23 *
24 * exofs is distributed in the hope that it will be useful,
25 * but WITHOUT ANY WARRANTY; without even the implied warranty of
26 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
27 * GNU General Public License for more details.
28 *
29 * You should have received a copy of the GNU General Public License
30 * along with exofs; if not, write to the Free Software
31 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
32 */
33
34 #include <linux/string.h>
35 #include <linux/parser.h>
36 #include <linux/vfs.h>
37 #include <linux/random.h>
38 #include <linux/module.h>
39 #include <linux/exportfs.h>
40 #include <linux/slab.h>
41
42 #include "exofs.h"
43
44 #define EXOFS_DBGMSG2(M...) do {} while (0)
45
46 /******************************************************************************
47 * MOUNT OPTIONS
48 *****************************************************************************/
49
50 /*
51 * struct to hold what we get from mount options
52 */
53 struct exofs_mountopt {
54 bool is_osdname;
55 const char *dev_name;
56 uint64_t pid;
57 int timeout;
58 };
59
60 /*
61 * exofs-specific mount-time options.
62 */
63 enum { Opt_name, Opt_pid, Opt_to, Opt_err };
64
65 /*
66 * Our mount-time options. These should ideally be 64-bit unsigned, but the
67 * kernel's parsing functions do not currently support that. 32-bit should be
68 * sufficient for most applications now.
69 */
70 static match_table_t tokens = {
71 {Opt_name, "osdname=%s"},
72 {Opt_pid, "pid=%u"},
73 {Opt_to, "to=%u"},
74 {Opt_err, NULL}
75 };
76
77 /*
78 * The main option parsing method. Also makes sure that all of the mandatory
79 * mount options were set.
80 */
81 static int parse_options(char *options, struct exofs_mountopt *opts)
82 {
83 char *p;
84 substring_t args[MAX_OPT_ARGS];
85 int option;
86 bool s_pid = false;
87
88 EXOFS_DBGMSG("parse_options %s\n", options);
89 /* defaults */
90 memset(opts, 0, sizeof(*opts));
91 opts->timeout = BLK_DEFAULT_SG_TIMEOUT;
92
93 while ((p = strsep(&options, ",")) != NULL) {
94 int token;
95 char str[32];
96
97 if (!*p)
98 continue;
99
100 token = match_token(p, tokens, args);
101 switch (token) {
102 case Opt_name:
103 opts->dev_name = match_strdup(&args[0]);
104 if (unlikely(!opts->dev_name)) {
105 EXOFS_ERR("Error allocating dev_name");
106 return -ENOMEM;
107 }
108 opts->is_osdname = true;
109 break;
110 case Opt_pid:
111 if (0 == match_strlcpy(str, &args[0], sizeof(str)))
112 return -EINVAL;
113 opts->pid = simple_strtoull(str, NULL, 0);
114 if (opts->pid < EXOFS_MIN_PID) {
115 EXOFS_ERR("Partition ID must be >= %u",
116 EXOFS_MIN_PID);
117 return -EINVAL;
118 }
119 s_pid = 1;
120 break;
121 case Opt_to:
122 if (match_int(&args[0], &option))
123 return -EINVAL;
124 if (option <= 0) {
125 EXOFS_ERR("Timout must be > 0");
126 return -EINVAL;
127 }
128 opts->timeout = option * HZ;
129 break;
130 }
131 }
132
133 if (!s_pid) {
134 EXOFS_ERR("Need to specify the following options:\n");
135 EXOFS_ERR(" -o pid=pid_no_to_use\n");
136 return -EINVAL;
137 }
138
139 return 0;
140 }
141
142 /******************************************************************************
143 * INODE CACHE
144 *****************************************************************************/
145
146 /*
147 * Our inode cache. Isn't it pretty?
148 */
149 static struct kmem_cache *exofs_inode_cachep;
150
151 /*
152 * Allocate an inode in the cache
153 */
154 static struct inode *exofs_alloc_inode(struct super_block *sb)
155 {
156 struct exofs_i_info *oi;
157
158 oi = kmem_cache_alloc(exofs_inode_cachep, GFP_KERNEL);
159 if (!oi)
160 return NULL;
161
162 oi->vfs_inode.i_version = 1;
163 return &oi->vfs_inode;
164 }
165
166 static void exofs_i_callback(struct rcu_head *head)
167 {
168 struct inode *inode = container_of(head, struct inode, i_rcu);
169 INIT_LIST_HEAD(&inode->i_dentry);
170 kmem_cache_free(exofs_inode_cachep, exofs_i(inode));
171 }
172
173 /*
174 * Remove an inode from the cache
175 */
176 static void exofs_destroy_inode(struct inode *inode)
177 {
178 call_rcu(&inode->i_rcu, exofs_i_callback);
179 }
180
181 /*
182 * Initialize the inode
183 */
184 static void exofs_init_once(void *foo)
185 {
186 struct exofs_i_info *oi = foo;
187
188 inode_init_once(&oi->vfs_inode);
189 }
190
191 /*
192 * Create and initialize the inode cache
193 */
194 static int init_inodecache(void)
195 {
196 exofs_inode_cachep = kmem_cache_create("exofs_inode_cache",
197 sizeof(struct exofs_i_info), 0,
198 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
199 exofs_init_once);
200 if (exofs_inode_cachep == NULL)
201 return -ENOMEM;
202 return 0;
203 }
204
205 /*
206 * Destroy the inode cache
207 */
208 static void destroy_inodecache(void)
209 {
210 kmem_cache_destroy(exofs_inode_cachep);
211 }
212
213 /******************************************************************************
214 * Some osd helpers
215 *****************************************************************************/
216 void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], const struct osd_obj_id *obj)
217 {
218 osd_sec_init_nosec_doall_caps(cred_a, obj, false, true);
219 }
220
221 static int exofs_read_kern(struct osd_dev *od, u8 *cred, struct osd_obj_id *obj,
222 u64 offset, void *p, unsigned length)
223 {
224 struct osd_request *or = osd_start_request(od, GFP_KERNEL);
225 /* struct osd_sense_info osi = {.key = 0};*/
226 int ret;
227
228 if (unlikely(!or)) {
229 EXOFS_DBGMSG("%s: osd_start_request failed.\n", __func__);
230 return -ENOMEM;
231 }
232 ret = osd_req_read_kern(or, obj, offset, p, length);
233 if (unlikely(ret)) {
234 EXOFS_DBGMSG("%s: osd_req_read_kern failed.\n", __func__);
235 goto out;
236 }
237
238 ret = osd_finalize_request(or, 0, cred, NULL);
239 if (unlikely(ret)) {
240 EXOFS_DBGMSG("Failed to osd_finalize_request() => %d\n", ret);
241 goto out;
242 }
243
244 ret = osd_execute_request(or);
245 if (unlikely(ret))
246 EXOFS_DBGMSG("osd_execute_request() => %d\n", ret);
247 /* osd_req_decode_sense(or, ret); */
248
249 out:
250 osd_end_request(or);
251 EXOFS_DBGMSG2("read_kern(0x%llx) offset=0x%llx "
252 "length=0x%llx dev=%p ret=>%d\n",
253 _LLU(obj->id), _LLU(offset), _LLU(length), od, ret);
254 return ret;
255 }
256
257 static const struct osd_attr g_attr_sb_stats = ATTR_DEF(
258 EXOFS_APAGE_SB_DATA,
259 EXOFS_ATTR_SB_STATS,
260 sizeof(struct exofs_sb_stats));
261
262 static int __sbi_read_stats(struct exofs_sb_info *sbi)
263 {
264 struct osd_attr attrs[] = {
265 [0] = g_attr_sb_stats,
266 };
267 struct ore_io_state *ios;
268 int ret;
269
270 ret = ore_get_io_state(&sbi->layout, &sbi->comps, &ios);
271 if (unlikely(ret)) {
272 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
273 return ret;
274 }
275
276 ios->in_attr = attrs;
277 ios->in_attr_len = ARRAY_SIZE(attrs);
278
279 ret = ore_read(ios);
280 if (unlikely(ret)) {
281 EXOFS_ERR("Error reading super_block stats => %d\n", ret);
282 goto out;
283 }
284
285 ret = extract_attr_from_ios(ios, &attrs[0]);
286 if (ret) {
287 EXOFS_ERR("%s: extract_attr of sb_stats failed\n", __func__);
288 goto out;
289 }
290 if (attrs[0].len) {
291 struct exofs_sb_stats *ess;
292
293 if (unlikely(attrs[0].len != sizeof(*ess))) {
294 EXOFS_ERR("%s: Wrong version of exofs_sb_stats "
295 "size(%d) != expected(%zd)\n",
296 __func__, attrs[0].len, sizeof(*ess));
297 goto out;
298 }
299
300 ess = attrs[0].val_ptr;
301 sbi->s_nextid = le64_to_cpu(ess->s_nextid);
302 sbi->s_numfiles = le32_to_cpu(ess->s_numfiles);
303 }
304
305 out:
306 ore_put_io_state(ios);
307 return ret;
308 }
309
310 static void stats_done(struct ore_io_state *ios, void *p)
311 {
312 ore_put_io_state(ios);
313 /* Good thanks nothing to do anymore */
314 }
315
316 /* Asynchronously write the stats attribute */
317 int exofs_sbi_write_stats(struct exofs_sb_info *sbi)
318 {
319 struct osd_attr attrs[] = {
320 [0] = g_attr_sb_stats,
321 };
322 struct ore_io_state *ios;
323 int ret;
324
325 ret = ore_get_io_state(&sbi->layout, &sbi->comps, &ios);
326 if (unlikely(ret)) {
327 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
328 return ret;
329 }
330
331 sbi->s_ess.s_nextid = cpu_to_le64(sbi->s_nextid);
332 sbi->s_ess.s_numfiles = cpu_to_le64(sbi->s_numfiles);
333 attrs[0].val_ptr = &sbi->s_ess;
334
335
336 ios->done = stats_done;
337 ios->private = sbi;
338 ios->out_attr = attrs;
339 ios->out_attr_len = ARRAY_SIZE(attrs);
340
341 ret = ore_write(ios);
342 if (unlikely(ret)) {
343 EXOFS_ERR("%s: ore_write failed.\n", __func__);
344 ore_put_io_state(ios);
345 }
346
347 return ret;
348 }
349
350 /******************************************************************************
351 * SUPERBLOCK FUNCTIONS
352 *****************************************************************************/
353 static const struct super_operations exofs_sops;
354 static const struct export_operations exofs_export_ops;
355
356 /*
357 * Write the superblock to the OSD
358 */
359 int exofs_sync_fs(struct super_block *sb, int wait)
360 {
361 struct exofs_sb_info *sbi;
362 struct exofs_fscb *fscb;
363 struct ore_comp one_comp;
364 struct ore_components comps;
365 struct ore_io_state *ios;
366 int ret = -ENOMEM;
367
368 fscb = kmalloc(sizeof(*fscb), GFP_KERNEL);
369 if (unlikely(!fscb))
370 return -ENOMEM;
371
372 sbi = sb->s_fs_info;
373
374 /* NOTE: We no longer dirty the super_block anywhere in exofs. The
375 * reason we write the fscb here on unmount is so we can stay backwards
376 * compatible with fscb->s_version == 1. (What we are not compatible
377 * with is if a new version FS crashed and then we try to mount an old
378 * version). Otherwise the exofs_fscb is read-only from mkfs time. All
379 * the writeable info is set in exofs_sbi_write_stats() above.
380 */
381
382 exofs_init_comps(&comps, &one_comp, sbi, EXOFS_SUPER_ID);
383
384 ret = ore_get_io_state(&sbi->layout, &comps, &ios);
385 if (unlikely(ret))
386 goto out;
387
388 lock_super(sb);
389
390 ios->length = offsetof(struct exofs_fscb, s_dev_table_oid);
391 memset(fscb, 0, ios->length);
392 fscb->s_nextid = cpu_to_le64(sbi->s_nextid);
393 fscb->s_numfiles = cpu_to_le32(sbi->s_numfiles);
394 fscb->s_magic = cpu_to_le16(sb->s_magic);
395 fscb->s_newfs = 0;
396 fscb->s_version = EXOFS_FSCB_VER;
397
398 ios->offset = 0;
399 ios->kern_buff = fscb;
400
401 ret = ore_write(ios);
402 if (unlikely(ret))
403 EXOFS_ERR("%s: ore_write failed.\n", __func__);
404 else
405 sb->s_dirt = 0;
406
407
408 unlock_super(sb);
409 out:
410 EXOFS_DBGMSG("s_nextid=0x%llx ret=%d\n", _LLU(sbi->s_nextid), ret);
411 ore_put_io_state(ios);
412 kfree(fscb);
413 return ret;
414 }
415
416 static void exofs_write_super(struct super_block *sb)
417 {
418 if (!(sb->s_flags & MS_RDONLY))
419 exofs_sync_fs(sb, 1);
420 else
421 sb->s_dirt = 0;
422 }
423
424 static void _exofs_print_device(const char *msg, const char *dev_path,
425 struct osd_dev *od, u64 pid)
426 {
427 const struct osd_dev_info *odi = osduld_device_info(od);
428
429 printk(KERN_NOTICE "exofs: %s %s osd_name-%s pid-0x%llx\n",
430 msg, dev_path ?: "", odi->osdname, _LLU(pid));
431 }
432
433 void exofs_free_sbi(struct exofs_sb_info *sbi)
434 {
435 while (sbi->comps.numdevs) {
436 int i = --sbi->comps.numdevs;
437 struct osd_dev *od = sbi->comps.ods[i];
438
439 if (od) {
440 sbi->comps.ods[i] = NULL;
441 osduld_put_device(od);
442 }
443 }
444 if (sbi->comps.ods != sbi->_min_one_dev)
445 kfree(sbi->comps.ods);
446 kfree(sbi);
447 }
448
449 /*
450 * This function is called when the vfs is freeing the superblock. We just
451 * need to free our own part.
452 */
453 static void exofs_put_super(struct super_block *sb)
454 {
455 int num_pend;
456 struct exofs_sb_info *sbi = sb->s_fs_info;
457
458 /* make sure there are no pending commands */
459 for (num_pend = atomic_read(&sbi->s_curr_pending); num_pend > 0;
460 num_pend = atomic_read(&sbi->s_curr_pending)) {
461 wait_queue_head_t wq;
462
463 printk(KERN_NOTICE "%s: !!Pending operations in flight. "
464 "This is a BUG. please report to osd-dev@open-osd.org\n",
465 __func__);
466 init_waitqueue_head(&wq);
467 wait_event_timeout(wq,
468 (atomic_read(&sbi->s_curr_pending) == 0),
469 msecs_to_jiffies(100));
470 }
471
472 _exofs_print_device("Unmounting", NULL, sbi->comps.ods[0],
473 sbi->one_comp.obj.partition);
474
475 bdi_destroy(&sbi->bdi);
476 exofs_free_sbi(sbi);
477 sb->s_fs_info = NULL;
478 }
479
480 static int _read_and_match_data_map(struct exofs_sb_info *sbi, unsigned numdevs,
481 struct exofs_device_table *dt)
482 {
483 u64 stripe_length;
484
485 sbi->layout.stripe_unit =
486 le64_to_cpu(dt->dt_data_map.cb_stripe_unit);
487 sbi->layout.group_width =
488 le32_to_cpu(dt->dt_data_map.cb_group_width);
489 sbi->layout.group_depth =
490 le32_to_cpu(dt->dt_data_map.cb_group_depth);
491 sbi->layout.mirrors_p1 =
492 le32_to_cpu(dt->dt_data_map.cb_mirror_cnt) + 1;
493 sbi->layout.raid_algorithm =
494 le32_to_cpu(dt->dt_data_map.cb_raid_algorithm);
495
496 /* FIXME: Only raid0 for now. if not so, do not mount */
497 if (sbi->layout.raid_algorithm != PNFS_OSD_RAID_0) {
498 EXOFS_ERR("Only RAID_0 for now\n");
499 return -EINVAL;
500 }
501 if (numdevs < (sbi->layout.group_width * sbi->layout.mirrors_p1)) {
502 EXOFS_ERR("Data Map wrong, "
503 "numdevs=%d < group_width=%d * mirrors=%d\n",
504 numdevs, sbi->layout.group_width,
505 sbi->layout.mirrors_p1);
506 return -EINVAL;
507 }
508
509 if (0 != (sbi->layout.stripe_unit & ~PAGE_MASK)) {
510 EXOFS_ERR("Stripe Unit(0x%llx)"
511 " must be Multples of PAGE_SIZE(0x%lx)\n",
512 _LLU(sbi->layout.stripe_unit), PAGE_SIZE);
513 return -EINVAL;
514 }
515
516 if (sbi->layout.group_width) {
517 if (!sbi->layout.group_depth) {
518 EXOFS_ERR("group_depth == 0 && group_width != 0\n");
519 return -EINVAL;
520 }
521 sbi->layout.group_count = numdevs / sbi->layout.mirrors_p1 /
522 sbi->layout.group_width;
523 } else {
524 if (sbi->layout.group_depth) {
525 printk(KERN_NOTICE "Warning: group_depth ignored "
526 "group_width == 0 && group_depth == %lld\n",
527 _LLU(sbi->layout.group_depth));
528 }
529 sbi->layout.group_width = numdevs / sbi->layout.mirrors_p1;
530 sbi->layout.group_depth = -1;
531 sbi->layout.group_count = 1;
532 }
533
534 stripe_length = (u64)sbi->layout.group_width * sbi->layout.stripe_unit;
535 if (stripe_length >= (1ULL << 32)) {
536 EXOFS_ERR("Total Stripe length(0x%llx)"
537 " >= 32bit is not supported\n", _LLU(stripe_length));
538 return -EINVAL;
539 }
540
541 EXOFS_DBGMSG("exofs: layout: "
542 "num_comps=%u stripe_unit=0x%x group_width=%u "
543 "group_depth=0x%llx mirrors_p1=%u raid_algorithm=%u\n",
544 numdevs,
545 sbi->layout.stripe_unit,
546 sbi->layout.group_width,
547 _LLU(sbi->layout.group_depth),
548 sbi->layout.mirrors_p1,
549 sbi->layout.raid_algorithm);
550 return 0;
551 }
552
553 static unsigned __ra_pages(struct ore_layout *layout)
554 {
555 const unsigned _MIN_RA = 32; /* min 128K read-ahead */
556 unsigned ra_pages = layout->group_width * layout->stripe_unit /
557 PAGE_SIZE;
558 unsigned max_io_pages = exofs_max_io_pages(layout, ~0);
559
560 ra_pages *= 2; /* two stripes */
561 if (ra_pages < _MIN_RA)
562 ra_pages = roundup(_MIN_RA, ra_pages / 2);
563
564 if (ra_pages > max_io_pages)
565 ra_pages = max_io_pages;
566
567 return ra_pages;
568 }
569
570 /* @odi is valid only as long as @fscb_dev is valid */
571 static int exofs_devs_2_odi(struct exofs_dt_device_info *dt_dev,
572 struct osd_dev_info *odi)
573 {
574 odi->systemid_len = le32_to_cpu(dt_dev->systemid_len);
575 memcpy(odi->systemid, dt_dev->systemid, odi->systemid_len);
576
577 odi->osdname_len = le32_to_cpu(dt_dev->osdname_len);
578 odi->osdname = dt_dev->osdname;
579
580 /* FIXME support long names. Will need a _put function */
581 if (dt_dev->long_name_offset)
582 return -EINVAL;
583
584 /* Make sure osdname is printable!
585 * mkexofs should give us space for a null-terminator else the
586 * device-table is invalid.
587 */
588 if (unlikely(odi->osdname_len >= sizeof(dt_dev->osdname)))
589 odi->osdname_len = sizeof(dt_dev->osdname) - 1;
590 dt_dev->osdname[odi->osdname_len] = 0;
591
592 /* If it's all zeros something is bad we read past end-of-obj */
593 return !(odi->systemid_len || odi->osdname_len);
594 }
595
596 static int exofs_read_lookup_dev_table(struct exofs_sb_info *sbi,
597 struct osd_dev *fscb_od,
598 unsigned table_count)
599 {
600 struct ore_comp comp;
601 struct exofs_device_table *dt;
602 unsigned table_bytes = table_count * sizeof(dt->dt_dev_table[0]) +
603 sizeof(*dt);
604 unsigned numdevs, i;
605 int ret;
606
607 dt = kmalloc(table_bytes, GFP_KERNEL);
608 if (unlikely(!dt)) {
609 EXOFS_ERR("ERROR: allocating %x bytes for device table\n",
610 table_bytes);
611 return -ENOMEM;
612 }
613
614 sbi->comps.numdevs = 0;
615
616 comp.obj.partition = sbi->one_comp.obj.partition;
617 comp.obj.id = EXOFS_DEVTABLE_ID;
618 exofs_make_credential(comp.cred, &comp.obj);
619
620 ret = exofs_read_kern(fscb_od, comp.cred, &comp.obj, 0, dt,
621 table_bytes);
622 if (unlikely(ret)) {
623 EXOFS_ERR("ERROR: reading device table\n");
624 goto out;
625 }
626
627 numdevs = le64_to_cpu(dt->dt_num_devices);
628 if (unlikely(!numdevs)) {
629 ret = -EINVAL;
630 goto out;
631 }
632 WARN_ON(table_count != numdevs);
633
634 ret = _read_and_match_data_map(sbi, numdevs, dt);
635 if (unlikely(ret))
636 goto out;
637
638 if (likely(numdevs > 1)) {
639 unsigned size = numdevs * sizeof(sbi->comps.ods[0]);
640
641 /* Twice bigger table: See exofs_init_comps() and below
642 * comment
643 */
644 sbi->comps.ods = kzalloc(size + size - 1, GFP_KERNEL);
645 if (unlikely(!sbi->comps.ods)) {
646 EXOFS_ERR("ERROR: faild allocating Device array[%d]\n",
647 numdevs);
648 ret = -ENOMEM;
649 goto out;
650 }
651 }
652
653 for (i = 0; i < numdevs; i++) {
654 struct exofs_fscb fscb;
655 struct osd_dev_info odi;
656 struct osd_dev *od;
657
658 if (exofs_devs_2_odi(&dt->dt_dev_table[i], &odi)) {
659 EXOFS_ERR("ERROR: Read all-zeros device entry\n");
660 ret = -EINVAL;
661 goto out;
662 }
663
664 printk(KERN_NOTICE "Add device[%d]: osd_name-%s\n",
665 i, odi.osdname);
666
667 /* On all devices the device table is identical. The user can
668 * specify any one of the participating devices on the command
669 * line. We always keep them in device-table order.
670 */
671 if (fscb_od && osduld_device_same(fscb_od, &odi)) {
672 sbi->comps.ods[i] = fscb_od;
673 ++sbi->comps.numdevs;
674 fscb_od = NULL;
675 continue;
676 }
677
678 od = osduld_info_lookup(&odi);
679 if (IS_ERR(od)) {
680 ret = PTR_ERR(od);
681 EXOFS_ERR("ERROR: device requested is not found "
682 "osd_name-%s =>%d\n", odi.osdname, ret);
683 goto out;
684 }
685
686 sbi->comps.ods[i] = od;
687 ++sbi->comps.numdevs;
688
689 /* Read the fscb of the other devices to make sure the FS
690 * partition is there.
691 */
692 ret = exofs_read_kern(od, comp.cred, &comp.obj, 0, &fscb,
693 sizeof(fscb));
694 if (unlikely(ret)) {
695 EXOFS_ERR("ERROR: Malformed participating device "
696 "error reading fscb osd_name-%s\n",
697 odi.osdname);
698 goto out;
699 }
700
701 /* TODO: verify other information is correct and FS-uuid
702 * matches. Benny what did you say about device table
703 * generation and old devices?
704 */
705 }
706
707 out:
708 kfree(dt);
709 if (likely(!ret)) {
710 unsigned numdevs = sbi->comps.numdevs;
711
712 if (unlikely(fscb_od)) {
713 EXOFS_ERR("ERROR: Bad device-table container device not present\n");
714 osduld_put_device(fscb_od);
715 return -EINVAL;
716 }
717 /* exofs round-robins the device table view according to inode
718 * number. We hold a: twice bigger table hence inodes can point
719 * to any device and have a sequential view of the table
720 * starting at this device. See exofs_init_comps()
721 */
722 for (i = 0; i < numdevs - 1; ++i)
723 sbi->comps.ods[i + numdevs] = sbi->comps.ods[i];
724 }
725 return ret;
726 }
727
728 /*
729 * Read the superblock from the OSD and fill in the fields
730 */
731 static int exofs_fill_super(struct super_block *sb, void *data, int silent)
732 {
733 struct inode *root;
734 struct exofs_mountopt *opts = data;
735 struct exofs_sb_info *sbi; /*extended info */
736 struct osd_dev *od; /* Master device */
737 struct exofs_fscb fscb; /*on-disk superblock info */
738 struct ore_comp comp;
739 unsigned table_count;
740 int ret;
741
742 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
743 if (!sbi)
744 return -ENOMEM;
745
746 /* use mount options to fill superblock */
747 if (opts->is_osdname) {
748 struct osd_dev_info odi = {.systemid_len = 0};
749
750 odi.osdname_len = strlen(opts->dev_name);
751 odi.osdname = (u8 *)opts->dev_name;
752 od = osduld_info_lookup(&odi);
753 kfree(opts->dev_name);
754 opts->dev_name = NULL;
755 } else {
756 od = osduld_path_lookup(opts->dev_name);
757 }
758 if (IS_ERR(od)) {
759 ret = -EINVAL;
760 goto free_sbi;
761 }
762
763 /* Default layout in case we do not have a device-table */
764 sbi->layout.stripe_unit = PAGE_SIZE;
765 sbi->layout.mirrors_p1 = 1;
766 sbi->layout.group_width = 1;
767 sbi->layout.group_depth = -1;
768 sbi->layout.group_count = 1;
769 sbi->s_timeout = opts->timeout;
770
771 sbi->one_comp.obj.partition = opts->pid;
772 sbi->one_comp.obj.id = 0;
773 exofs_make_credential(sbi->one_comp.cred, &sbi->one_comp.obj);
774 sbi->comps.numdevs = 1;
775 sbi->comps.single_comp = EC_SINGLE_COMP;
776 sbi->comps.comps = &sbi->one_comp;
777 sbi->comps.ods = sbi->_min_one_dev;
778
779 /* fill in some other data by hand */
780 memset(sb->s_id, 0, sizeof(sb->s_id));
781 strcpy(sb->s_id, "exofs");
782 sb->s_blocksize = EXOFS_BLKSIZE;
783 sb->s_blocksize_bits = EXOFS_BLKSHIFT;
784 sb->s_maxbytes = MAX_LFS_FILESIZE;
785 atomic_set(&sbi->s_curr_pending, 0);
786 sb->s_bdev = NULL;
787 sb->s_dev = 0;
788
789 comp.obj.partition = sbi->one_comp.obj.partition;
790 comp.obj.id = EXOFS_SUPER_ID;
791 exofs_make_credential(comp.cred, &comp.obj);
792
793 ret = exofs_read_kern(od, comp.cred, &comp.obj, 0, &fscb, sizeof(fscb));
794 if (unlikely(ret))
795 goto free_sbi;
796
797 sb->s_magic = le16_to_cpu(fscb.s_magic);
798 /* NOTE: we read below to be backward compatible with old versions */
799 sbi->s_nextid = le64_to_cpu(fscb.s_nextid);
800 sbi->s_numfiles = le32_to_cpu(fscb.s_numfiles);
801
802 /* make sure what we read from the object store is correct */
803 if (sb->s_magic != EXOFS_SUPER_MAGIC) {
804 if (!silent)
805 EXOFS_ERR("ERROR: Bad magic value\n");
806 ret = -EINVAL;
807 goto free_sbi;
808 }
809 if (le32_to_cpu(fscb.s_version) > EXOFS_FSCB_VER) {
810 EXOFS_ERR("ERROR: Bad FSCB version expected-%d got-%d\n",
811 EXOFS_FSCB_VER, le32_to_cpu(fscb.s_version));
812 ret = -EINVAL;
813 goto free_sbi;
814 }
815
816 /* start generation numbers from a random point */
817 get_random_bytes(&sbi->s_next_generation, sizeof(u32));
818 spin_lock_init(&sbi->s_next_gen_lock);
819
820 table_count = le64_to_cpu(fscb.s_dev_table_count);
821 if (table_count) {
822 ret = exofs_read_lookup_dev_table(sbi, od, table_count);
823 if (unlikely(ret))
824 goto free_sbi;
825 } else {
826 sbi->comps.ods[0] = od;
827 }
828
829 __sbi_read_stats(sbi);
830
831 /* set up operation vectors */
832 sbi->bdi.ra_pages = __ra_pages(&sbi->layout);
833 sb->s_bdi = &sbi->bdi;
834 sb->s_fs_info = sbi;
835 sb->s_op = &exofs_sops;
836 sb->s_export_op = &exofs_export_ops;
837 root = exofs_iget(sb, EXOFS_ROOT_ID - EXOFS_OBJ_OFF);
838 if (IS_ERR(root)) {
839 EXOFS_ERR("ERROR: exofs_iget failed\n");
840 ret = PTR_ERR(root);
841 goto free_sbi;
842 }
843 sb->s_root = d_alloc_root(root);
844 if (!sb->s_root) {
845 iput(root);
846 EXOFS_ERR("ERROR: get root inode failed\n");
847 ret = -ENOMEM;
848 goto free_sbi;
849 }
850
851 if (!S_ISDIR(root->i_mode)) {
852 dput(sb->s_root);
853 sb->s_root = NULL;
854 EXOFS_ERR("ERROR: corrupt root inode (mode = %hd)\n",
855 root->i_mode);
856 ret = -EINVAL;
857 goto free_sbi;
858 }
859
860 ret = bdi_setup_and_register(&sbi->bdi, "exofs", BDI_CAP_MAP_COPY);
861 if (ret) {
862 EXOFS_DBGMSG("Failed to bdi_setup_and_register\n");
863 goto free_sbi;
864 }
865
866 _exofs_print_device("Mounting", opts->dev_name, sbi->comps.ods[0],
867 sbi->one_comp.obj.partition);
868 return 0;
869
870 free_sbi:
871 EXOFS_ERR("Unable to mount exofs on %s pid=0x%llx err=%d\n",
872 opts->dev_name, sbi->one_comp.obj.partition, ret);
873 exofs_free_sbi(sbi);
874 return ret;
875 }
876
877 /*
878 * Set up the superblock (calls exofs_fill_super eventually)
879 */
880 static struct dentry *exofs_mount(struct file_system_type *type,
881 int flags, const char *dev_name,
882 void *data)
883 {
884 struct exofs_mountopt opts;
885 int ret;
886
887 ret = parse_options(data, &opts);
888 if (ret)
889 return ERR_PTR(ret);
890
891 if (!opts.dev_name)
892 opts.dev_name = dev_name;
893 return mount_nodev(type, flags, &opts, exofs_fill_super);
894 }
895
896 /*
897 * Return information about the file system state in the buffer. This is used
898 * by the 'df' command, for example.
899 */
900 static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
901 {
902 struct super_block *sb = dentry->d_sb;
903 struct exofs_sb_info *sbi = sb->s_fs_info;
904 struct ore_io_state *ios;
905 struct osd_attr attrs[] = {
906 ATTR_DEF(OSD_APAGE_PARTITION_QUOTAS,
907 OSD_ATTR_PQ_CAPACITY_QUOTA, sizeof(__be64)),
908 ATTR_DEF(OSD_APAGE_PARTITION_INFORMATION,
909 OSD_ATTR_PI_USED_CAPACITY, sizeof(__be64)),
910 };
911 uint64_t capacity = ULLONG_MAX;
912 uint64_t used = ULLONG_MAX;
913 int ret;
914
915 ret = ore_get_io_state(&sbi->layout, &sbi->comps, &ios);
916 if (ret) {
917 EXOFS_DBGMSG("ore_get_io_state failed.\n");
918 return ret;
919 }
920
921 ios->in_attr = attrs;
922 ios->in_attr_len = ARRAY_SIZE(attrs);
923
924 ret = ore_read(ios);
925 if (unlikely(ret))
926 goto out;
927
928 ret = extract_attr_from_ios(ios, &attrs[0]);
929 if (likely(!ret)) {
930 capacity = get_unaligned_be64(attrs[0].val_ptr);
931 if (unlikely(!capacity))
932 capacity = ULLONG_MAX;
933 } else
934 EXOFS_DBGMSG("exofs_statfs: get capacity failed.\n");
935
936 ret = extract_attr_from_ios(ios, &attrs[1]);
937 if (likely(!ret))
938 used = get_unaligned_be64(attrs[1].val_ptr);
939 else
940 EXOFS_DBGMSG("exofs_statfs: get used-space failed.\n");
941
942 /* fill in the stats buffer */
943 buf->f_type = EXOFS_SUPER_MAGIC;
944 buf->f_bsize = EXOFS_BLKSIZE;
945 buf->f_blocks = capacity >> 9;
946 buf->f_bfree = (capacity - used) >> 9;
947 buf->f_bavail = buf->f_bfree;
948 buf->f_files = sbi->s_numfiles;
949 buf->f_ffree = EXOFS_MAX_ID - sbi->s_numfiles;
950 buf->f_namelen = EXOFS_NAME_LEN;
951
952 out:
953 ore_put_io_state(ios);
954 return ret;
955 }
956
957 static const struct super_operations exofs_sops = {
958 .alloc_inode = exofs_alloc_inode,
959 .destroy_inode = exofs_destroy_inode,
960 .write_inode = exofs_write_inode,
961 .evict_inode = exofs_evict_inode,
962 .put_super = exofs_put_super,
963 .write_super = exofs_write_super,
964 .sync_fs = exofs_sync_fs,
965 .statfs = exofs_statfs,
966 };
967
968 /******************************************************************************
969 * EXPORT OPERATIONS
970 *****************************************************************************/
971
972 struct dentry *exofs_get_parent(struct dentry *child)
973 {
974 unsigned long ino = exofs_parent_ino(child);
975
976 if (!ino)
977 return ERR_PTR(-ESTALE);
978
979 return d_obtain_alias(exofs_iget(child->d_inode->i_sb, ino));
980 }
981
982 static struct inode *exofs_nfs_get_inode(struct super_block *sb,
983 u64 ino, u32 generation)
984 {
985 struct inode *inode;
986
987 inode = exofs_iget(sb, ino);
988 if (IS_ERR(inode))
989 return ERR_CAST(inode);
990 if (generation && inode->i_generation != generation) {
991 /* we didn't find the right inode.. */
992 iput(inode);
993 return ERR_PTR(-ESTALE);
994 }
995 return inode;
996 }
997
998 static struct dentry *exofs_fh_to_dentry(struct super_block *sb,
999 struct fid *fid, int fh_len, int fh_type)
1000 {
1001 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1002 exofs_nfs_get_inode);
1003 }
1004
1005 static struct dentry *exofs_fh_to_parent(struct super_block *sb,
1006 struct fid *fid, int fh_len, int fh_type)
1007 {
1008 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1009 exofs_nfs_get_inode);
1010 }
1011
1012 static const struct export_operations exofs_export_ops = {
1013 .fh_to_dentry = exofs_fh_to_dentry,
1014 .fh_to_parent = exofs_fh_to_parent,
1015 .get_parent = exofs_get_parent,
1016 };
1017
1018 /******************************************************************************
1019 * INSMOD/RMMOD
1020 *****************************************************************************/
1021
1022 /*
1023 * struct that describes this file system
1024 */
1025 static struct file_system_type exofs_type = {
1026 .owner = THIS_MODULE,
1027 .name = "exofs",
1028 .mount = exofs_mount,
1029 .kill_sb = generic_shutdown_super,
1030 };
1031
1032 static int __init init_exofs(void)
1033 {
1034 int err;
1035
1036 err = init_inodecache();
1037 if (err)
1038 goto out;
1039
1040 err = register_filesystem(&exofs_type);
1041 if (err)
1042 goto out_d;
1043
1044 return 0;
1045 out_d:
1046 destroy_inodecache();
1047 out:
1048 return err;
1049 }
1050
1051 static void __exit exit_exofs(void)
1052 {
1053 unregister_filesystem(&exofs_type);
1054 destroy_inodecache();
1055 }
1056
1057 MODULE_AUTHOR("Avishay Traeger <avishay@gmail.com>");
1058 MODULE_DESCRIPTION("exofs");
1059 MODULE_LICENSE("GPL");
1060
1061 module_init(init_exofs)
1062 module_exit(exit_exofs)
linux-next