block: Expose discard granularity
[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/kmod.h>
20 #include <linux/ctype.h>
21 #include <linux/genhd.h>
22 #include <linux/blktrace_api.h>
23
24 #include "check.h"
25
26 #include "acorn.h"
27 #include "amiga.h"
28 #include "atari.h"
29 #include "ldm.h"
30 #include "mac.h"
31 #include "msdos.h"
32 #include "osf.h"
33 #include "sgi.h"
34 #include "sun.h"
35 #include "ibm.h"
36 #include "ultrix.h"
37 #include "efi.h"
38 #include "karma.h"
39 #include "sysv68.h"
40
41 #ifdef CONFIG_BLK_DEV_MD
42 extern void md_autodetect_dev(dev_t dev);
43 #endif
44
45 int warn_no_part = 1; /*This is ugly: should make genhd removable media aware*/
46
47 static int (*check_part[])(struct parsed_partitions *, struct block_device *) = {
48         /*
49          * Probe partition formats with tables at disk address 0
50          * that also have an ADFS boot block at 0xdc0.
51          */
52 #ifdef CONFIG_ACORN_PARTITION_ICS
53         adfspart_check_ICS,
54 #endif
55 #ifdef CONFIG_ACORN_PARTITION_POWERTEC
56         adfspart_check_POWERTEC,
57 #endif
58 #ifdef CONFIG_ACORN_PARTITION_EESOX
59         adfspart_check_EESOX,
60 #endif
61
62         /*
63          * Now move on to formats that only have partition info at
64          * disk address 0xdc0.  Since these may also have stale
65          * PC/BIOS partition tables, they need to come before
66          * the msdos entry.
67          */
68 #ifdef CONFIG_ACORN_PARTITION_CUMANA
69         adfspart_check_CUMANA,
70 #endif
71 #ifdef CONFIG_ACORN_PARTITION_ADFS
72         adfspart_check_ADFS,
73 #endif
74
75 #ifdef CONFIG_EFI_PARTITION
76         efi_partition,          /* this must come before msdos */
77 #endif
78 #ifdef CONFIG_SGI_PARTITION
79         sgi_partition,
80 #endif
81 #ifdef CONFIG_LDM_PARTITION
82         ldm_partition,          /* this must come before msdos */
83 #endif
84 #ifdef CONFIG_MSDOS_PARTITION
85         msdos_partition,
86 #endif
87 #ifdef CONFIG_OSF_PARTITION
88         osf_partition,
89 #endif
90 #ifdef CONFIG_SUN_PARTITION
91         sun_partition,
92 #endif
93 #ifdef CONFIG_AMIGA_PARTITION
94         amiga_partition,
95 #endif
96 #ifdef CONFIG_ATARI_PARTITION
97         atari_partition,
98 #endif
99 #ifdef CONFIG_MAC_PARTITION
100         mac_partition,
101 #endif
102 #ifdef CONFIG_ULTRIX_PARTITION
103         ultrix_partition,
104 #endif
105 #ifdef CONFIG_IBM_PARTITION
106         ibm_partition,
107 #endif
108 #ifdef CONFIG_KARMA_PARTITION
109         karma_partition,
110 #endif
111 #ifdef CONFIG_SYSV68_PARTITION
112         sysv68_partition,
113 #endif
114         NULL
115 };
116  
117 /*
118  * disk_name() is used by partition check code and the genhd driver.
119  * It formats the devicename of the indicated disk into
120  * the supplied buffer (of size at least 32), and returns
121  * a pointer to that same buffer (for convenience).
122  */
123
124 char *disk_name(struct gendisk *hd, int partno, char *buf)
125 {
126         if (!partno)
127                 snprintf(buf, BDEVNAME_SIZE, "%s", hd->disk_name);
128         else if (isdigit(hd->disk_name[strlen(hd->disk_name)-1]))
129                 snprintf(buf, BDEVNAME_SIZE, "%sp%d", hd->disk_name, partno);
130         else
131                 snprintf(buf, BDEVNAME_SIZE, "%s%d", hd->disk_name, partno);
132
133         return buf;
134 }
135
136 const char *bdevname(struct block_device *bdev, char *buf)
137 {
138         return disk_name(bdev->bd_disk, bdev->bd_part->partno, buf);
139 }
140
141 EXPORT_SYMBOL(bdevname);
142
143 /*
144  * There's very little reason to use this, you should really
145  * have a struct block_device just about everywhere and use
146  * bdevname() instead.
147  */
148 const char *__bdevname(dev_t dev, char *buffer)
149 {
150         scnprintf(buffer, BDEVNAME_SIZE, "unknown-block(%u,%u)",
151                                 MAJOR(dev), MINOR(dev));
152         return buffer;
153 }
154
155 EXPORT_SYMBOL(__bdevname);
156
157 static struct parsed_partitions *
158 check_partition(struct gendisk *hd, struct block_device *bdev)
159 {
160         struct parsed_partitions *state;
161         int i, res, err;
162
163         state = kmalloc(sizeof(struct parsed_partitions), GFP_KERNEL);
164         if (!state)
165                 return NULL;
166
167         disk_name(hd, 0, state->name);
168         printk(KERN_INFO " %s:", state->name);
169         if (isdigit(state->name[strlen(state->name)-1]))
170                 sprintf(state->name, "p");
171
172         state->limit = disk_max_parts(hd);
173         i = res = err = 0;
174         while (!res && check_part[i]) {
175                 memset(&state->parts, 0, sizeof(state->parts));
176                 res = check_part[i++](state, bdev);
177                 if (res < 0) {
178                         /* We have hit an I/O error which we don't report now.
179                         * But record it, and let the others do their job.
180                         */
181                         err = res;
182                         res = 0;
183                 }
184
185         }
186         if (res > 0)
187                 return state;
188         if (err)
189         /* The partition is unrecognized. So report I/O errors if there were any */
190                 res = err;
191         if (!res)
192                 printk(" unknown partition table\n");
193         else if (warn_no_part)
194                 printk(" unable to read partition table\n");
195         kfree(state);
196         return ERR_PTR(res);
197 }
198
199 static ssize_t part_partition_show(struct device *dev,
200                                    struct device_attribute *attr, char *buf)
201 {
202         struct hd_struct *p = dev_to_part(dev);
203
204         return sprintf(buf, "%d\n", p->partno);
205 }
206
207 static ssize_t part_start_show(struct device *dev,
208                                struct device_attribute *attr, char *buf)
209 {
210         struct hd_struct *p = dev_to_part(dev);
211
212         return sprintf(buf, "%llu\n",(unsigned long long)p->start_sect);
213 }
214
215 ssize_t part_size_show(struct device *dev,
216                        struct device_attribute *attr, char *buf)
217 {
218         struct hd_struct *p = dev_to_part(dev);
219         return sprintf(buf, "%llu\n",(unsigned long long)p->nr_sects);
220 }
221
222 ssize_t part_alignment_offset_show(struct device *dev,
223                                    struct device_attribute *attr, char *buf)
224 {
225         struct hd_struct *p = dev_to_part(dev);
226         return sprintf(buf, "%llu\n", (unsigned long long)p->alignment_offset);
227 }
228
229 ssize_t part_discard_alignment_show(struct device *dev,
230                                    struct device_attribute *attr, char *buf)
231 {
232         struct hd_struct *p = dev_to_part(dev);
233         return sprintf(buf, "%u\n", p->discard_alignment);
234 }
235
236 ssize_t part_stat_show(struct device *dev,
237                        struct device_attribute *attr, char *buf)
238 {
239         struct hd_struct *p = dev_to_part(dev);
240         int cpu;
241
242         cpu = part_stat_lock();
243         part_round_stats(cpu, p);
244         part_stat_unlock();
245         return sprintf(buf,
246                 "%8lu %8lu %8llu %8u "
247                 "%8lu %8lu %8llu %8u "
248                 "%8u %8u %8u"
249                 "\n",
250                 part_stat_read(p, ios[READ]),
251                 part_stat_read(p, merges[READ]),
252                 (unsigned long long)part_stat_read(p, sectors[READ]),
253                 jiffies_to_msecs(part_stat_read(p, ticks[READ])),
254                 part_stat_read(p, ios[WRITE]),
255                 part_stat_read(p, merges[WRITE]),
256                 (unsigned long long)part_stat_read(p, sectors[WRITE]),
257                 jiffies_to_msecs(part_stat_read(p, ticks[WRITE])),
258                 part_in_flight(p),
259                 jiffies_to_msecs(part_stat_read(p, io_ticks)),
260                 jiffies_to_msecs(part_stat_read(p, time_in_queue)));
261 }
262
263 ssize_t part_inflight_show(struct device *dev,
264                         struct device_attribute *attr, char *buf)
265 {
266         struct hd_struct *p = dev_to_part(dev);
267
268         return sprintf(buf, "%8u %8u\n", p->in_flight[0], p->in_flight[1]);
269 }
270
271 #ifdef CONFIG_FAIL_MAKE_REQUEST
272 ssize_t part_fail_show(struct device *dev,
273                        struct device_attribute *attr, char *buf)
274 {
275         struct hd_struct *p = dev_to_part(dev);
276
277         return sprintf(buf, "%d\n", p->make_it_fail);
278 }
279
280 ssize_t part_fail_store(struct device *dev,
281                         struct device_attribute *attr,
282                         const char *buf, size_t count)
283 {
284         struct hd_struct *p = dev_to_part(dev);
285         int i;
286
287         if (count > 0 && sscanf(buf, "%d", &i) > 0)
288                 p->make_it_fail = (i == 0) ? 0 : 1;
289
290         return count;
291 }
292 #endif
293
294 static DEVICE_ATTR(partition, S_IRUGO, part_partition_show, NULL);
295 static DEVICE_ATTR(start, S_IRUGO, part_start_show, NULL);
296 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
297 static DEVICE_ATTR(alignment_offset, S_IRUGO, part_alignment_offset_show, NULL);
298 static DEVICE_ATTR(discard_alignment, S_IRUGO, part_discard_alignment_show,
299                    NULL);
300 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
301 static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
302 #ifdef CONFIG_FAIL_MAKE_REQUEST
303 static struct device_attribute dev_attr_fail =
304         __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
305 #endif
306
307 static struct attribute *part_attrs[] = {
308         &dev_attr_partition.attr,
309         &dev_attr_start.attr,
310         &dev_attr_size.attr,
311         &dev_attr_alignment_offset.attr,
312         &dev_attr_discard_alignment.attr,
313         &dev_attr_stat.attr,
314         &dev_attr_inflight.attr,
315 #ifdef CONFIG_FAIL_MAKE_REQUEST
316         &dev_attr_fail.attr,
317 #endif
318         NULL
319 };
320
321 static struct attribute_group part_attr_group = {
322         .attrs = part_attrs,
323 };
324
325 static const struct attribute_group *part_attr_groups[] = {
326         &part_attr_group,
327 #ifdef CONFIG_BLK_DEV_IO_TRACE
328         &blk_trace_attr_group,
329 #endif
330         NULL
331 };
332
333 static void part_release(struct device *dev)
334 {
335         struct hd_struct *p = dev_to_part(dev);
336         free_part_stats(p);
337         kfree(p);
338 }
339
340 struct device_type part_type = {
341         .name           = "partition",
342         .groups         = part_attr_groups,
343         .release        = part_release,
344 };
345
346 static void delete_partition_rcu_cb(struct rcu_head *head)
347 {
348         struct hd_struct *part = container_of(head, struct hd_struct, rcu_head);
349
350         part->start_sect = 0;
351         part->nr_sects = 0;
352         part_stat_set_all(part, 0);
353         put_device(part_to_dev(part));
354 }
355
356 void delete_partition(struct gendisk *disk, int partno)
357 {
358         struct disk_part_tbl *ptbl = disk->part_tbl;
359         struct hd_struct *part;
360
361         if (partno >= ptbl->len)
362                 return;
363
364         part = ptbl->part[partno];
365         if (!part)
366                 return;
367
368         blk_free_devt(part_devt(part));
369         rcu_assign_pointer(ptbl->part[partno], NULL);
370         rcu_assign_pointer(ptbl->last_lookup, NULL);
371         kobject_put(part->holder_dir);
372         device_del(part_to_dev(part));
373
374         call_rcu(&part->rcu_head, delete_partition_rcu_cb);
375 }
376
377 static ssize_t whole_disk_show(struct device *dev,
378                                struct device_attribute *attr, char *buf)
379 {
380         return 0;
381 }
382 static DEVICE_ATTR(whole_disk, S_IRUSR | S_IRGRP | S_IROTH,
383                    whole_disk_show, NULL);
384
385 struct hd_struct *add_partition(struct gendisk *disk, int partno,
386                                 sector_t start, sector_t len, int flags)
387 {
388         struct hd_struct *p;
389         dev_t devt = MKDEV(0, 0);
390         struct device *ddev = disk_to_dev(disk);
391         struct device *pdev;
392         struct disk_part_tbl *ptbl;
393         const char *dname;
394         int err;
395
396         err = disk_expand_part_tbl(disk, partno);
397         if (err)
398                 return ERR_PTR(err);
399         ptbl = disk->part_tbl;
400
401         if (ptbl->part[partno])
402                 return ERR_PTR(-EBUSY);
403
404         p = kzalloc(sizeof(*p), GFP_KERNEL);
405         if (!p)
406                 return ERR_PTR(-EBUSY);
407
408         if (!init_part_stats(p)) {
409                 err = -ENOMEM;
410                 goto out_free;
411         }
412         pdev = part_to_dev(p);
413
414         p->start_sect = start;
415         p->alignment_offset = queue_sector_alignment_offset(disk->queue, start);
416         p->discard_alignment = queue_sector_discard_alignment(disk->queue,
417                                                               start);
418         p->nr_sects = len;
419         p->partno = partno;
420         p->policy = get_disk_ro(disk);
421
422         dname = dev_name(ddev);
423         if (isdigit(dname[strlen(dname) - 1]))
424                 dev_set_name(pdev, "%sp%d", dname, partno);
425         else
426                 dev_set_name(pdev, "%s%d", dname, partno);
427
428         device_initialize(pdev);
429         pdev->class = &block_class;
430         pdev->type = &part_type;
431         pdev->parent = ddev;
432
433         err = blk_alloc_devt(p, &devt);
434         if (err)
435                 goto out_free_stats;
436         pdev->devt = devt;
437
438         /* delay uevent until 'holders' subdir is created */
439         dev_set_uevent_suppress(pdev, 1);
440         err = device_add(pdev);
441         if (err)
442                 goto out_put;
443
444         err = -ENOMEM;
445         p->holder_dir = kobject_create_and_add("holders", &pdev->kobj);
446         if (!p->holder_dir)
447                 goto out_del;
448
449         dev_set_uevent_suppress(pdev, 0);
450         if (flags & ADDPART_FLAG_WHOLEDISK) {
451                 err = device_create_file(pdev, &dev_attr_whole_disk);
452                 if (err)
453                         goto out_del;
454         }
455
456         /* everything is up and running, commence */
457         INIT_RCU_HEAD(&p->rcu_head);
458         rcu_assign_pointer(ptbl->part[partno], p);
459
460         /* suppress uevent if the disk supresses it */
461         if (!dev_get_uevent_suppress(ddev))
462                 kobject_uevent(&pdev->kobj, KOBJ_ADD);
463
464         return p;
465
466 out_free_stats:
467         free_part_stats(p);
468 out_free:
469         kfree(p);
470         return ERR_PTR(err);
471 out_del:
472         kobject_put(p->holder_dir);
473         device_del(pdev);
474 out_put:
475         put_device(pdev);
476         blk_free_devt(devt);
477         return ERR_PTR(err);
478 }
479
480 /* Not exported, helper to add_disk(). */
481 void register_disk(struct gendisk *disk)
482 {
483         struct device *ddev = disk_to_dev(disk);
484         struct block_device *bdev;
485         struct disk_part_iter piter;
486         struct hd_struct *part;
487         int err;
488
489         ddev->parent = disk->driverfs_dev;
490
491         dev_set_name(ddev, disk->disk_name);
492
493         /* delay uevents, until we scanned partition table */
494         dev_set_uevent_suppress(ddev, 1);
495
496         if (device_add(ddev))
497                 return;
498 #ifndef CONFIG_SYSFS_DEPRECATED
499         err = sysfs_create_link(block_depr, &ddev->kobj,
500                                 kobject_name(&ddev->kobj));
501         if (err) {
502                 device_del(ddev);
503                 return;
504         }
505 #endif
506         disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
507         disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
508
509         /* No minors to use for partitions */
510         if (!disk_partitionable(disk))
511                 goto exit;
512
513         /* No such device (e.g., media were just removed) */
514         if (!get_capacity(disk))
515                 goto exit;
516
517         bdev = bdget_disk(disk, 0);
518         if (!bdev)
519                 goto exit;
520
521         bdev->bd_invalidated = 1;
522         err = blkdev_get(bdev, FMODE_READ);
523         if (err < 0)
524                 goto exit;
525         blkdev_put(bdev, FMODE_READ);
526
527 exit:
528         /* announce disk after possible partitions are created */
529         dev_set_uevent_suppress(ddev, 0);
530         kobject_uevent(&ddev->kobj, KOBJ_ADD);
531
532         /* announce possible partitions */
533         disk_part_iter_init(&piter, disk, 0);
534         while ((part = disk_part_iter_next(&piter)))
535                 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
536         disk_part_iter_exit(&piter);
537 }
538
539 int rescan_partitions(struct gendisk *disk, struct block_device *bdev)
540 {
541         struct disk_part_iter piter;
542         struct hd_struct *part;
543         struct parsed_partitions *state;
544         int p, highest, res;
545
546         if (bdev->bd_part_count)
547                 return -EBUSY;
548         res = invalidate_partition(disk, 0);
549         if (res)
550                 return res;
551
552         disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
553         while ((part = disk_part_iter_next(&piter)))
554                 delete_partition(disk, part->partno);
555         disk_part_iter_exit(&piter);
556
557         if (disk->fops->revalidate_disk)
558                 disk->fops->revalidate_disk(disk);
559         check_disk_size_change(disk, bdev);
560         bdev->bd_invalidated = 0;
561         if (!get_capacity(disk) || !(state = check_partition(disk, bdev)))
562                 return 0;
563         if (IS_ERR(state))      /* I/O error reading the partition table */
564                 return -EIO;
565
566         /* tell userspace that the media / partition table may have changed */
567         kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
568
569         /* Detect the highest partition number and preallocate
570          * disk->part_tbl.  This is an optimization and not strictly
571          * necessary.
572          */
573         for (p = 1, highest = 0; p < state->limit; p++)
574                 if (state->parts[p].size)
575                         highest = p;
576
577         disk_expand_part_tbl(disk, highest);
578
579         /* add partitions */
580         for (p = 1; p < state->limit; p++) {
581                 sector_t size, from;
582 try_scan:
583                 size = state->parts[p].size;
584                 if (!size)
585                         continue;
586
587                 from = state->parts[p].from;
588                 if (from >= get_capacity(disk)) {
589                         printk(KERN_WARNING
590                                "%s: p%d ignored, start %llu is behind the end of the disk\n",
591                                disk->disk_name, p, (unsigned long long) from);
592                         continue;
593                 }
594
595                 if (from + size > get_capacity(disk)) {
596                         const struct block_device_operations *bdops = disk->fops;
597                         unsigned long long capacity;
598
599                         printk(KERN_WARNING
600                                "%s: p%d size %llu exceeds device capacity, ",
601                                disk->disk_name, p, (unsigned long long) size);
602
603                         if (bdops->set_capacity &&
604                             (disk->flags & GENHD_FL_NATIVE_CAPACITY) == 0) {
605                                 printk(KERN_CONT "enabling native capacity\n");
606                                 capacity = bdops->set_capacity(disk, ~0ULL);
607                                 disk->flags |= GENHD_FL_NATIVE_CAPACITY;
608                                 if (capacity > get_capacity(disk)) {
609                                         set_capacity(disk, capacity);
610                                         check_disk_size_change(disk, bdev);
611                                         bdev->bd_invalidated = 0;
612                                 }
613                                 goto try_scan;
614                         } else {
615                                 /*
616                                  * we can not ignore partitions of broken tables
617                                  * created by for example camera firmware, but
618                                  * we limit them to the end of the disk to avoid
619                                  * creating invalid block devices
620                                  */
621                                 printk(KERN_CONT "limited to end of disk\n");
622                                 size = get_capacity(disk) - from;
623                         }
624                 }
625                 part = add_partition(disk, p, from, size,
626                                      state->parts[p].flags);
627                 if (IS_ERR(part)) {
628                         printk(KERN_ERR " %s: p%d could not be added: %ld\n",
629                                disk->disk_name, p, -PTR_ERR(part));
630                         continue;
631                 }
632 #ifdef CONFIG_BLK_DEV_MD
633                 if (state->parts[p].flags & ADDPART_FLAG_RAID)
634                         md_autodetect_dev(part_to_dev(part)->devt);
635 #endif
636         }
637         kfree(state);
638         return 0;
639 }
640
641 unsigned char *read_dev_sector(struct block_device *bdev, sector_t n, Sector *p)
642 {
643         struct address_space *mapping = bdev->bd_inode->i_mapping;
644         struct page *page;
645
646         page = read_mapping_page(mapping, (pgoff_t)(n >> (PAGE_CACHE_SHIFT-9)),
647                                  NULL);
648         if (!IS_ERR(page)) {
649                 if (PageError(page))
650                         goto fail;
651                 p->v = page;
652                 return (unsigned char *)page_address(page) +  ((n & ((1 << (PAGE_CACHE_SHIFT - 9)) - 1)) << 9);
653 fail:
654                 page_cache_release(page);
655         }
656         p->v = NULL;
657         return NULL;
658 }
659
660 EXPORT_SYMBOL(read_dev_sector);
661
662 void del_gendisk(struct gendisk *disk)
663 {
664         struct disk_part_iter piter;
665         struct hd_struct *part;
666
667         /* invalidate stuff */
668         disk_part_iter_init(&piter, disk,
669                              DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
670         while ((part = disk_part_iter_next(&piter))) {
671                 invalidate_partition(disk, part->partno);
672                 delete_partition(disk, part->partno);
673         }
674         disk_part_iter_exit(&piter);
675
676         invalidate_partition(disk, 0);
677         blk_free_devt(disk_to_dev(disk)->devt);
678         set_capacity(disk, 0);
679         disk->flags &= ~GENHD_FL_UP;
680         unlink_gendisk(disk);
681         part_stat_set_all(&disk->part0, 0);
682         disk->part0.stamp = 0;
683
684         kobject_put(disk->part0.holder_dir);
685         kobject_put(disk->slave_dir);
686         disk->driverfs_dev = NULL;
687 #ifndef CONFIG_SYSFS_DEPRECATED
688         sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
689 #endif
690         device_del(disk_to_dev(disk));
691 }