blk-throttle: Make total_nr_queued unsigned
[linux-2.6.git] / fs / partitions / check.c
1 /*
2  *  fs/partitions/check.c
3  *
4  *  Code extracted from drivers/block/genhd.c
5  *  Copyright (C) 1991-1998  Linus Torvalds
6  *  Re-organised Feb 1998 Russell King
7  *
8  *  We now have independent partition support from the
9  *  block drivers, which allows all the partition code to
10  *  be grouped in one location, and it to be mostly self
11  *  contained.
12  *
13  *  Added needed MAJORS for new pairs, {hdi,hdj}, {hdk,hdl}
14  */
15
16 #include <linux/init.h>
17 #include <linux/module.h>
18 #include <linux/fs.h>
19 #include <linux/slab.h>
20 #include <linux/kmod.h>
21 #include <linux/ctype.h>
22 #include <linux/genhd.h>
23 #include <linux/blktrace_api.h>
24
25 #include "check.h"
26
27 #include "acorn.h"
28 #include "amiga.h"
29 #include "atari.h"
30 #include "ldm.h"
31 #include "mac.h"
32 #include "msdos.h"
33 #include "osf.h"
34 #include "sgi.h"
35 #include "sun.h"
36 #include "ibm.h"
37 #include "ultrix.h"
38 #include "efi.h"
39 #include "karma.h"
40 #include "sysv68.h"
41
42 #ifdef CONFIG_BLK_DEV_MD
43 extern void md_autodetect_dev(dev_t dev);
44 #endif
45
46 int warn_no_part = 1; /*This is ugly: should make genhd removable media aware*/
47
48 static int (*check_part[])(struct parsed_partitions *) = {
49         /*
50          * Probe partition formats with tables at disk address 0
51          * that also have an ADFS boot block at 0xdc0.
52          */
53 #ifdef CONFIG_ACORN_PARTITION_ICS
54         adfspart_check_ICS,
55 #endif
56 #ifdef CONFIG_ACORN_PARTITION_POWERTEC
57         adfspart_check_POWERTEC,
58 #endif
59 #ifdef CONFIG_ACORN_PARTITION_EESOX
60         adfspart_check_EESOX,
61 #endif
62
63         /*
64          * Now move on to formats that only have partition info at
65          * disk address 0xdc0.  Since these may also have stale
66          * PC/BIOS partition tables, they need to come before
67          * the msdos entry.
68          */
69 #ifdef CONFIG_ACORN_PARTITION_CUMANA
70         adfspart_check_CUMANA,
71 #endif
72 #ifdef CONFIG_ACORN_PARTITION_ADFS
73         adfspart_check_ADFS,
74 #endif
75
76 #ifdef CONFIG_EFI_PARTITION
77         efi_partition,          /* this must come before msdos */
78 #endif
79 #ifdef CONFIG_SGI_PARTITION
80         sgi_partition,
81 #endif
82 #ifdef CONFIG_LDM_PARTITION
83         ldm_partition,          /* this must come before msdos */
84 #endif
85 #ifdef CONFIG_MSDOS_PARTITION
86         msdos_partition,
87 #endif
88 #ifdef CONFIG_OSF_PARTITION
89         osf_partition,
90 #endif
91 #ifdef CONFIG_SUN_PARTITION
92         sun_partition,
93 #endif
94 #ifdef CONFIG_AMIGA_PARTITION
95         amiga_partition,
96 #endif
97 #ifdef CONFIG_ATARI_PARTITION
98         atari_partition,
99 #endif
100 #ifdef CONFIG_MAC_PARTITION
101         mac_partition,
102 #endif
103 #ifdef CONFIG_ULTRIX_PARTITION
104         ultrix_partition,
105 #endif
106 #ifdef CONFIG_IBM_PARTITION
107         ibm_partition,
108 #endif
109 #ifdef CONFIG_KARMA_PARTITION
110         karma_partition,
111 #endif
112 #ifdef CONFIG_SYSV68_PARTITION
113         sysv68_partition,
114 #endif
115         NULL
116 };
117  
118 /*
119  * disk_name() is used by partition check code and the genhd driver.
120  * It formats the devicename of the indicated disk into
121  * the supplied buffer (of size at least 32), and returns
122  * a pointer to that same buffer (for convenience).
123  */
124
125 char *disk_name(struct gendisk *hd, int partno, char *buf)
126 {
127         if (!partno)
128                 snprintf(buf, BDEVNAME_SIZE, "%s", hd->disk_name);
129         else if (isdigit(hd->disk_name[strlen(hd->disk_name)-1]))
130                 snprintf(buf, BDEVNAME_SIZE, "%sp%d", hd->disk_name, partno);
131         else
132                 snprintf(buf, BDEVNAME_SIZE, "%s%d", hd->disk_name, partno);
133
134         return buf;
135 }
136
137 const char *bdevname(struct block_device *bdev, char *buf)
138 {
139         return disk_name(bdev->bd_disk, bdev->bd_part->partno, buf);
140 }
141
142 EXPORT_SYMBOL(bdevname);
143
144 /*
145  * There's very little reason to use this, you should really
146  * have a struct block_device just about everywhere and use
147  * bdevname() instead.
148  */
149 const char *__bdevname(dev_t dev, char *buffer)
150 {
151         scnprintf(buffer, BDEVNAME_SIZE, "unknown-block(%u,%u)",
152                                 MAJOR(dev), MINOR(dev));
153         return buffer;
154 }
155
156 EXPORT_SYMBOL(__bdevname);
157
158 static struct parsed_partitions *
159 check_partition(struct gendisk *hd, struct block_device *bdev)
160 {
161         struct parsed_partitions *state;
162         int i, res, err;
163
164         state = kzalloc(sizeof(struct parsed_partitions), GFP_KERNEL);
165         if (!state)
166                 return NULL;
167         state->pp_buf = (char *)__get_free_page(GFP_KERNEL);
168         if (!state->pp_buf) {
169                 kfree(state);
170                 return NULL;
171         }
172         state->pp_buf[0] = '\0';
173
174         state->bdev = bdev;
175         disk_name(hd, 0, state->name);
176         snprintf(state->pp_buf, PAGE_SIZE, " %s:", state->name);
177         if (isdigit(state->name[strlen(state->name)-1]))
178                 sprintf(state->name, "p");
179
180         state->limit = disk_max_parts(hd);
181         i = res = err = 0;
182         while (!res && check_part[i]) {
183                 memset(&state->parts, 0, sizeof(state->parts));
184                 res = check_part[i++](state);
185                 if (res < 0) {
186                         /* We have hit an I/O error which we don't report now.
187                         * But record it, and let the others do their job.
188                         */
189                         err = res;
190                         res = 0;
191                 }
192
193         }
194         if (res > 0) {
195                 printk(KERN_INFO "%s", state->pp_buf);
196
197                 free_page((unsigned long)state->pp_buf);
198                 return state;
199         }
200         if (state->access_beyond_eod)
201                 err = -ENOSPC;
202         if (err)
203         /* The partition is unrecognized. So report I/O errors if there were any */
204                 res = err;
205         if (!res)
206                 strlcat(state->pp_buf, " unknown partition table\n", PAGE_SIZE);
207         else if (warn_no_part)
208                 strlcat(state->pp_buf, " unable to read partition table\n", PAGE_SIZE);
209
210         printk(KERN_INFO "%s", state->pp_buf);
211
212         free_page((unsigned long)state->pp_buf);
213         kfree(state);
214         return ERR_PTR(res);
215 }
216
217 static ssize_t part_partition_show(struct device *dev,
218                                    struct device_attribute *attr, char *buf)
219 {
220         struct hd_struct *p = dev_to_part(dev);
221
222         return sprintf(buf, "%d\n", p->partno);
223 }
224
225 static ssize_t part_start_show(struct device *dev,
226                                struct device_attribute *attr, char *buf)
227 {
228         struct hd_struct *p = dev_to_part(dev);
229
230         return sprintf(buf, "%llu\n",(unsigned long long)p->start_sect);
231 }
232
233 ssize_t part_size_show(struct device *dev,
234                        struct device_attribute *attr, char *buf)
235 {
236         struct hd_struct *p = dev_to_part(dev);
237         return sprintf(buf, "%llu\n",(unsigned long long)p->nr_sects);
238 }
239
240 static ssize_t part_ro_show(struct device *dev,
241                             struct device_attribute *attr, char *buf)
242 {
243         struct hd_struct *p = dev_to_part(dev);
244         return sprintf(buf, "%d\n", p->policy ? 1 : 0);
245 }
246
247 static ssize_t part_alignment_offset_show(struct device *dev,
248                                           struct device_attribute *attr, char *buf)
249 {
250         struct hd_struct *p = dev_to_part(dev);
251         return sprintf(buf, "%llu\n", (unsigned long long)p->alignment_offset);
252 }
253
254 static ssize_t part_discard_alignment_show(struct device *dev,
255                                            struct device_attribute *attr, char *buf)
256 {
257         struct hd_struct *p = dev_to_part(dev);
258         struct gendisk *disk = dev_to_disk(dev);
259         unsigned int alignment = 0;
260
261         if (disk->queue)
262                 alignment = queue_limit_discard_alignment(&disk->queue->limits,
263                                                                 p->start_sect);
264         return sprintf(buf, "%u\n", alignment);
265 }
266
267 ssize_t part_stat_show(struct device *dev,
268                        struct device_attribute *attr, char *buf)
269 {
270         struct hd_struct *p = dev_to_part(dev);
271         int cpu;
272
273         cpu = part_stat_lock();
274         part_round_stats(cpu, p);
275         part_stat_unlock();
276         return sprintf(buf,
277                 "%8lu %8lu %8llu %8u "
278                 "%8lu %8lu %8llu %8u "
279                 "%8u %8u %8u"
280                 "\n",
281                 part_stat_read(p, ios[READ]),
282                 part_stat_read(p, merges[READ]),
283                 (unsigned long long)part_stat_read(p, sectors[READ]),
284                 jiffies_to_msecs(part_stat_read(p, ticks[READ])),
285                 part_stat_read(p, ios[WRITE]),
286                 part_stat_read(p, merges[WRITE]),
287                 (unsigned long long)part_stat_read(p, sectors[WRITE]),
288                 jiffies_to_msecs(part_stat_read(p, ticks[WRITE])),
289                 part_in_flight(p),
290                 jiffies_to_msecs(part_stat_read(p, io_ticks)),
291                 jiffies_to_msecs(part_stat_read(p, time_in_queue)));
292 }
293
294 ssize_t part_inflight_show(struct device *dev,
295                         struct device_attribute *attr, char *buf)
296 {
297         struct hd_struct *p = dev_to_part(dev);
298
299         return sprintf(buf, "%8u %8u\n", atomic_read(&p->in_flight[0]),
300                 atomic_read(&p->in_flight[1]));
301 }
302
303 #ifdef CONFIG_FAIL_MAKE_REQUEST
304 ssize_t part_fail_show(struct device *dev,
305                        struct device_attribute *attr, char *buf)
306 {
307         struct hd_struct *p = dev_to_part(dev);
308
309         return sprintf(buf, "%d\n", p->make_it_fail);
310 }
311
312 ssize_t part_fail_store(struct device *dev,
313                         struct device_attribute *attr,
314                         const char *buf, size_t count)
315 {
316         struct hd_struct *p = dev_to_part(dev);
317         int i;
318
319         if (count > 0 && sscanf(buf, "%d", &i) > 0)
320                 p->make_it_fail = (i == 0) ? 0 : 1;
321
322         return count;
323 }
324 #endif
325
326 static DEVICE_ATTR(partition, S_IRUGO, part_partition_show, NULL);
327 static DEVICE_ATTR(start, S_IRUGO, part_start_show, NULL);
328 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
329 static DEVICE_ATTR(ro, S_IRUGO, part_ro_show, NULL);
330 static DEVICE_ATTR(alignment_offset, S_IRUGO, part_alignment_offset_show, NULL);
331 static DEVICE_ATTR(discard_alignment, S_IRUGO, part_discard_alignment_show,
332                    NULL);
333 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
334 static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
335 #ifdef CONFIG_FAIL_MAKE_REQUEST
336 static struct device_attribute dev_attr_fail =
337         __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
338 #endif
339
340 static struct attribute *part_attrs[] = {
341         &dev_attr_partition.attr,
342         &dev_attr_start.attr,
343         &dev_attr_size.attr,
344         &dev_attr_ro.attr,
345         &dev_attr_alignment_offset.attr,
346         &dev_attr_discard_alignment.attr,
347         &dev_attr_stat.attr,
348         &dev_attr_inflight.attr,
349 #ifdef CONFIG_FAIL_MAKE_REQUEST
350         &dev_attr_fail.attr,
351 #endif
352         NULL
353 };
354
355 static struct attribute_group part_attr_group = {
356         .attrs = part_attrs,
357 };
358
359 static const struct attribute_group *part_attr_groups[] = {
360         &part_attr_group,
361 #ifdef CONFIG_BLK_DEV_IO_TRACE
362         &blk_trace_attr_group,
363 #endif
364         NULL
365 };
366
367 static void part_release(struct device *dev)
368 {
369         struct hd_struct *p = dev_to_part(dev);
370         free_part_stats(p);
371         free_part_info(p);
372         kfree(p);
373 }
374
375 struct device_type part_type = {
376         .name           = "partition",
377         .groups         = part_attr_groups,
378         .release        = part_release,
379 };
380
381 static void delete_partition_rcu_cb(struct rcu_head *head)
382 {
383         struct hd_struct *part = container_of(head, struct hd_struct, rcu_head);
384
385         part->start_sect = 0;
386         part->nr_sects = 0;
387         part_stat_set_all(part, 0);
388         put_device(part_to_dev(part));
389 }
390
391 void __delete_partition(struct hd_struct *part)
392 {
393         call_rcu(&part->rcu_head, delete_partition_rcu_cb);
394 }
395
396 void delete_partition(struct gendisk *disk, int partno)
397 {
398         struct disk_part_tbl *ptbl = disk->part_tbl;
399         struct hd_struct *part;
400
401         if (partno >= ptbl->len)
402                 return;
403
404         part = ptbl->part[partno];
405         if (!part)
406                 return;
407
408         blk_free_devt(part_devt(part));
409         rcu_assign_pointer(ptbl->part[partno], NULL);
410         rcu_assign_pointer(ptbl->last_lookup, NULL);
411         kobject_put(part->holder_dir);
412         device_del(part_to_dev(part));
413
414         hd_struct_put(part);
415 }
416
417 static ssize_t whole_disk_show(struct device *dev,
418                                struct device_attribute *attr, char *buf)
419 {
420         return 0;
421 }
422 static DEVICE_ATTR(whole_disk, S_IRUSR | S_IRGRP | S_IROTH,
423                    whole_disk_show, NULL);
424
425 struct hd_struct *add_partition(struct gendisk *disk, int partno,
426                                 sector_t start, sector_t len, int flags,
427                                 struct partition_meta_info *info)
428 {
429         struct hd_struct *p;
430         dev_t devt = MKDEV(0, 0);
431         struct device *ddev = disk_to_dev(disk);
432         struct device *pdev;
433         struct disk_part_tbl *ptbl;
434         const char *dname;
435         int err;
436
437         err = disk_expand_part_tbl(disk, partno);
438         if (err)
439                 return ERR_PTR(err);
440         ptbl = disk->part_tbl;
441
442         if (ptbl->part[partno])
443                 return ERR_PTR(-EBUSY);
444
445         p = kzalloc(sizeof(*p), GFP_KERNEL);
446         if (!p)
447                 return ERR_PTR(-EBUSY);
448
449         if (!init_part_stats(p)) {
450                 err = -ENOMEM;
451                 goto out_free;
452         }
453         pdev = part_to_dev(p);
454
455         p->start_sect = start;
456         p->alignment_offset =
457                 queue_limit_alignment_offset(&disk->queue->limits, start);
458         p->nr_sects = len;
459         p->partno = partno;
460         p->policy = get_disk_ro(disk);
461
462         if (info) {
463                 struct partition_meta_info *pinfo = alloc_part_info(disk);
464                 if (!pinfo)
465                         goto out_free_stats;
466                 memcpy(pinfo, info, sizeof(*info));
467                 p->info = pinfo;
468         }
469
470         dname = dev_name(ddev);
471         if (isdigit(dname[strlen(dname) - 1]))
472                 dev_set_name(pdev, "%sp%d", dname, partno);
473         else
474                 dev_set_name(pdev, "%s%d", dname, partno);
475
476         device_initialize(pdev);
477         pdev->class = &block_class;
478         pdev->type = &part_type;
479         pdev->parent = ddev;
480
481         err = blk_alloc_devt(p, &devt);
482         if (err)
483                 goto out_free_info;
484         pdev->devt = devt;
485
486         /* delay uevent until 'holders' subdir is created */
487         dev_set_uevent_suppress(pdev, 1);
488         err = device_add(pdev);
489         if (err)
490                 goto out_put;
491
492         err = -ENOMEM;
493         p->holder_dir = kobject_create_and_add("holders", &pdev->kobj);
494         if (!p->holder_dir)
495                 goto out_del;
496
497         dev_set_uevent_suppress(pdev, 0);
498         if (flags & ADDPART_FLAG_WHOLEDISK) {
499                 err = device_create_file(pdev, &dev_attr_whole_disk);
500                 if (err)
501                         goto out_del;
502         }
503
504         /* everything is up and running, commence */
505         rcu_assign_pointer(ptbl->part[partno], p);
506
507         /* suppress uevent if the disk suppresses it */
508         if (!dev_get_uevent_suppress(ddev))
509                 kobject_uevent(&pdev->kobj, KOBJ_ADD);
510
511         hd_ref_init(p);
512         return p;
513
514 out_free_info:
515         free_part_info(p);
516 out_free_stats:
517         free_part_stats(p);
518 out_free:
519         kfree(p);
520         return ERR_PTR(err);
521 out_del:
522         kobject_put(p->holder_dir);
523         device_del(pdev);
524 out_put:
525         put_device(pdev);
526         blk_free_devt(devt);
527         return ERR_PTR(err);
528 }
529
530 static bool disk_unlock_native_capacity(struct gendisk *disk)
531 {
532         const struct block_device_operations *bdops = disk->fops;
533
534         if (bdops->unlock_native_capacity &&
535             !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) {
536                 printk(KERN_CONT "enabling native capacity\n");
537                 bdops->unlock_native_capacity(disk);
538                 disk->flags |= GENHD_FL_NATIVE_CAPACITY;
539                 return true;
540         } else {
541                 printk(KERN_CONT "truncated\n");
542                 return false;
543         }
544 }
545
546 int rescan_partitions(struct gendisk *disk, struct block_device *bdev)
547 {
548         struct parsed_partitions *state = NULL;
549         struct disk_part_iter piter;
550         struct hd_struct *part;
551         int p, highest, res;
552 rescan:
553         if (state && !IS_ERR(state)) {
554                 kfree(state);
555                 state = NULL;
556         }
557
558         if (bdev->bd_part_count)
559                 return -EBUSY;
560         res = invalidate_partition(disk, 0);
561         if (res)
562                 return res;
563
564         disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
565         while ((part = disk_part_iter_next(&piter)))
566                 delete_partition(disk, part->partno);
567         disk_part_iter_exit(&piter);
568
569         if (disk->fops->revalidate_disk)
570                 disk->fops->revalidate_disk(disk);
571         check_disk_size_change(disk, bdev);
572         bdev->bd_invalidated = 0;
573         if (!get_capacity(disk) || !(state = check_partition(disk, bdev)))
574                 return 0;
575         if (IS_ERR(state)) {
576                 /*
577                  * I/O error reading the partition table.  If any
578                  * partition code tried to read beyond EOD, retry
579                  * after unlocking native capacity.
580                  */
581                 if (PTR_ERR(state) == -ENOSPC) {
582                         printk(KERN_WARNING "%s: partition table beyond EOD, ",
583                                disk->disk_name);
584                         if (disk_unlock_native_capacity(disk))
585                                 goto rescan;
586                 }
587                 return -EIO;
588         }
589         /*
590          * If any partition code tried to read beyond EOD, try
591          * unlocking native capacity even if partition table is
592          * successfully read as we could be missing some partitions.
593          */
594         if (state->access_beyond_eod) {
595                 printk(KERN_WARNING
596                        "%s: partition table partially beyond EOD, ",
597                        disk->disk_name);
598                 if (disk_unlock_native_capacity(disk))
599                         goto rescan;
600         }
601
602         /* tell userspace that the media / partition table may have changed */
603         kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
604
605         /* Detect the highest partition number and preallocate
606          * disk->part_tbl.  This is an optimization and not strictly
607          * necessary.
608          */
609         for (p = 1, highest = 0; p < state->limit; p++)
610                 if (state->parts[p].size)
611                         highest = p;
612
613         disk_expand_part_tbl(disk, highest);
614
615         /* add partitions */
616         for (p = 1; p < state->limit; p++) {
617                 sector_t size, from;
618                 struct partition_meta_info *info = NULL;
619
620                 size = state->parts[p].size;
621                 if (!size)
622                         continue;
623
624                 from = state->parts[p].from;
625                 if (from >= get_capacity(disk)) {
626                         printk(KERN_WARNING
627                                "%s: p%d start %llu is beyond EOD, ",
628                                disk->disk_name, p, (unsigned long long) from);
629                         if (disk_unlock_native_capacity(disk))
630                                 goto rescan;
631                         continue;
632                 }
633
634                 if (from + size > get_capacity(disk)) {
635                         printk(KERN_WARNING
636                                "%s: p%d size %llu extends beyond EOD, ",
637                                disk->disk_name, p, (unsigned long long) size);
638
639                         if (disk_unlock_native_capacity(disk)) {
640                                 /* free state and restart */
641                                 goto rescan;
642                         } else {
643                                 /*
644                                  * we can not ignore partitions of broken tables
645                                  * created by for example camera firmware, but
646                                  * we limit them to the end of the disk to avoid
647                                  * creating invalid block devices
648                                  */
649                                 size = get_capacity(disk) - from;
650                         }
651                 }
652
653                 if (state->parts[p].has_info)
654                         info = &state->parts[p].info;
655                 part = add_partition(disk, p, from, size,
656                                      state->parts[p].flags,
657                                      &state->parts[p].info);
658                 if (IS_ERR(part)) {
659                         printk(KERN_ERR " %s: p%d could not be added: %ld\n",
660                                disk->disk_name, p, -PTR_ERR(part));
661                         continue;
662                 }
663 #ifdef CONFIG_BLK_DEV_MD
664                 if (state->parts[p].flags & ADDPART_FLAG_RAID)
665                         md_autodetect_dev(part_to_dev(part)->devt);
666 #endif
667         }
668         kfree(state);
669         return 0;
670 }
671
672 unsigned char *read_dev_sector(struct block_device *bdev, sector_t n, Sector *p)
673 {
674         struct address_space *mapping = bdev->bd_inode->i_mapping;
675         struct page *page;
676
677         page = read_mapping_page(mapping, (pgoff_t)(n >> (PAGE_CACHE_SHIFT-9)),
678                                  NULL);
679         if (!IS_ERR(page)) {
680                 if (PageError(page))
681                         goto fail;
682                 p->v = page;
683                 return (unsigned char *)page_address(page) +  ((n & ((1 << (PAGE_CACHE_SHIFT - 9)) - 1)) << 9);
684 fail:
685                 page_cache_release(page);
686         }
687         p->v = NULL;
688         return NULL;
689 }
690
691 EXPORT_SYMBOL(read_dev_sector);