Back to home page

LXR

 
 

    


0001 /*
0002  *  gendisk handling
0003  */
0004 
0005 #include <linux/module.h>
0006 #include <linux/fs.h>
0007 #include <linux/genhd.h>
0008 #include <linux/kdev_t.h>
0009 #include <linux/kernel.h>
0010 #include <linux/blkdev.h>
0011 #include <linux/backing-dev.h>
0012 #include <linux/init.h>
0013 #include <linux/spinlock.h>
0014 #include <linux/proc_fs.h>
0015 #include <linux/seq_file.h>
0016 #include <linux/slab.h>
0017 #include <linux/kmod.h>
0018 #include <linux/kobj_map.h>
0019 #include <linux/mutex.h>
0020 #include <linux/idr.h>
0021 #include <linux/log2.h>
0022 #include <linux/pm_runtime.h>
0023 #include <linux/badblocks.h>
0024 
0025 #include "blk.h"
0026 
0027 static DEFINE_MUTEX(block_class_lock);
0028 struct kobject *block_depr;
0029 
0030 /* for extended dynamic devt allocation, currently only one major is used */
0031 #define NR_EXT_DEVT     (1 << MINORBITS)
0032 
0033 /* For extended devt allocation.  ext_devt_lock prevents look up
0034  * results from going away underneath its user.
0035  */
0036 static DEFINE_SPINLOCK(ext_devt_lock);
0037 static DEFINE_IDR(ext_devt_idr);
0038 
0039 static struct device_type disk_type;
0040 
0041 static void disk_check_events(struct disk_events *ev,
0042                   unsigned int *clearing_ptr);
0043 static void disk_alloc_events(struct gendisk *disk);
0044 static void disk_add_events(struct gendisk *disk);
0045 static void disk_del_events(struct gendisk *disk);
0046 static void disk_release_events(struct gendisk *disk);
0047 
0048 /**
0049  * disk_get_part - get partition
0050  * @disk: disk to look partition from
0051  * @partno: partition number
0052  *
0053  * Look for partition @partno from @disk.  If found, increment
0054  * reference count and return it.
0055  *
0056  * CONTEXT:
0057  * Don't care.
0058  *
0059  * RETURNS:
0060  * Pointer to the found partition on success, NULL if not found.
0061  */
0062 struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
0063 {
0064     struct hd_struct *part = NULL;
0065     struct disk_part_tbl *ptbl;
0066 
0067     if (unlikely(partno < 0))
0068         return NULL;
0069 
0070     rcu_read_lock();
0071 
0072     ptbl = rcu_dereference(disk->part_tbl);
0073     if (likely(partno < ptbl->len)) {
0074         part = rcu_dereference(ptbl->part[partno]);
0075         if (part)
0076             get_device(part_to_dev(part));
0077     }
0078 
0079     rcu_read_unlock();
0080 
0081     return part;
0082 }
0083 EXPORT_SYMBOL_GPL(disk_get_part);
0084 
0085 /**
0086  * disk_part_iter_init - initialize partition iterator
0087  * @piter: iterator to initialize
0088  * @disk: disk to iterate over
0089  * @flags: DISK_PITER_* flags
0090  *
0091  * Initialize @piter so that it iterates over partitions of @disk.
0092  *
0093  * CONTEXT:
0094  * Don't care.
0095  */
0096 void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
0097               unsigned int flags)
0098 {
0099     struct disk_part_tbl *ptbl;
0100 
0101     rcu_read_lock();
0102     ptbl = rcu_dereference(disk->part_tbl);
0103 
0104     piter->disk = disk;
0105     piter->part = NULL;
0106 
0107     if (flags & DISK_PITER_REVERSE)
0108         piter->idx = ptbl->len - 1;
0109     else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
0110         piter->idx = 0;
0111     else
0112         piter->idx = 1;
0113 
0114     piter->flags = flags;
0115 
0116     rcu_read_unlock();
0117 }
0118 EXPORT_SYMBOL_GPL(disk_part_iter_init);
0119 
0120 /**
0121  * disk_part_iter_next - proceed iterator to the next partition and return it
0122  * @piter: iterator of interest
0123  *
0124  * Proceed @piter to the next partition and return it.
0125  *
0126  * CONTEXT:
0127  * Don't care.
0128  */
0129 struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
0130 {
0131     struct disk_part_tbl *ptbl;
0132     int inc, end;
0133 
0134     /* put the last partition */
0135     disk_put_part(piter->part);
0136     piter->part = NULL;
0137 
0138     /* get part_tbl */
0139     rcu_read_lock();
0140     ptbl = rcu_dereference(piter->disk->part_tbl);
0141 
0142     /* determine iteration parameters */
0143     if (piter->flags & DISK_PITER_REVERSE) {
0144         inc = -1;
0145         if (piter->flags & (DISK_PITER_INCL_PART0 |
0146                     DISK_PITER_INCL_EMPTY_PART0))
0147             end = -1;
0148         else
0149             end = 0;
0150     } else {
0151         inc = 1;
0152         end = ptbl->len;
0153     }
0154 
0155     /* iterate to the next partition */
0156     for (; piter->idx != end; piter->idx += inc) {
0157         struct hd_struct *part;
0158 
0159         part = rcu_dereference(ptbl->part[piter->idx]);
0160         if (!part)
0161             continue;
0162         if (!part_nr_sects_read(part) &&
0163             !(piter->flags & DISK_PITER_INCL_EMPTY) &&
0164             !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
0165               piter->idx == 0))
0166             continue;
0167 
0168         get_device(part_to_dev(part));
0169         piter->part = part;
0170         piter->idx += inc;
0171         break;
0172     }
0173 
0174     rcu_read_unlock();
0175 
0176     return piter->part;
0177 }
0178 EXPORT_SYMBOL_GPL(disk_part_iter_next);
0179 
0180 /**
0181  * disk_part_iter_exit - finish up partition iteration
0182  * @piter: iter of interest
0183  *
0184  * Called when iteration is over.  Cleans up @piter.
0185  *
0186  * CONTEXT:
0187  * Don't care.
0188  */
0189 void disk_part_iter_exit(struct disk_part_iter *piter)
0190 {
0191     disk_put_part(piter->part);
0192     piter->part = NULL;
0193 }
0194 EXPORT_SYMBOL_GPL(disk_part_iter_exit);
0195 
0196 static inline int sector_in_part(struct hd_struct *part, sector_t sector)
0197 {
0198     return part->start_sect <= sector &&
0199         sector < part->start_sect + part_nr_sects_read(part);
0200 }
0201 
0202 /**
0203  * disk_map_sector_rcu - map sector to partition
0204  * @disk: gendisk of interest
0205  * @sector: sector to map
0206  *
0207  * Find out which partition @sector maps to on @disk.  This is
0208  * primarily used for stats accounting.
0209  *
0210  * CONTEXT:
0211  * RCU read locked.  The returned partition pointer is valid only
0212  * while preemption is disabled.
0213  *
0214  * RETURNS:
0215  * Found partition on success, part0 is returned if no partition matches
0216  */
0217 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
0218 {
0219     struct disk_part_tbl *ptbl;
0220     struct hd_struct *part;
0221     int i;
0222 
0223     ptbl = rcu_dereference(disk->part_tbl);
0224 
0225     part = rcu_dereference(ptbl->last_lookup);
0226     if (part && sector_in_part(part, sector))
0227         return part;
0228 
0229     for (i = 1; i < ptbl->len; i++) {
0230         part = rcu_dereference(ptbl->part[i]);
0231 
0232         if (part && sector_in_part(part, sector)) {
0233             rcu_assign_pointer(ptbl->last_lookup, part);
0234             return part;
0235         }
0236     }
0237     return &disk->part0;
0238 }
0239 EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
0240 
0241 /*
0242  * Can be deleted altogether. Later.
0243  *
0244  */
0245 static struct blk_major_name {
0246     struct blk_major_name *next;
0247     int major;
0248     char name[16];
0249 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
0250 
0251 /* index in the above - for now: assume no multimajor ranges */
0252 static inline int major_to_index(unsigned major)
0253 {
0254     return major % BLKDEV_MAJOR_HASH_SIZE;
0255 }
0256 
0257 #ifdef CONFIG_PROC_FS
0258 void blkdev_show(struct seq_file *seqf, off_t offset)
0259 {
0260     struct blk_major_name *dp;
0261 
0262     if (offset < BLKDEV_MAJOR_HASH_SIZE) {
0263         mutex_lock(&block_class_lock);
0264         for (dp = major_names[offset]; dp; dp = dp->next)
0265             seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
0266         mutex_unlock(&block_class_lock);
0267     }
0268 }
0269 #endif /* CONFIG_PROC_FS */
0270 
0271 /**
0272  * register_blkdev - register a new block device
0273  *
0274  * @major: the requested major device number [1..255]. If @major=0, try to
0275  *         allocate any unused major number.
0276  * @name: the name of the new block device as a zero terminated string
0277  *
0278  * The @name must be unique within the system.
0279  *
0280  * The return value depends on the @major input parameter.
0281  *  - if a major device number was requested in range [1..255] then the
0282  *    function returns zero on success, or a negative error code
0283  *  - if any unused major number was requested with @major=0 parameter
0284  *    then the return value is the allocated major number in range
0285  *    [1..255] or a negative error code otherwise
0286  */
0287 int register_blkdev(unsigned int major, const char *name)
0288 {
0289     struct blk_major_name **n, *p;
0290     int index, ret = 0;
0291 
0292     mutex_lock(&block_class_lock);
0293 
0294     /* temporary */
0295     if (major == 0) {
0296         for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
0297             if (major_names[index] == NULL)
0298                 break;
0299         }
0300 
0301         if (index == 0) {
0302             printk("register_blkdev: failed to get major for %s\n",
0303                    name);
0304             ret = -EBUSY;
0305             goto out;
0306         }
0307         major = index;
0308         ret = major;
0309     }
0310 
0311     p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
0312     if (p == NULL) {
0313         ret = -ENOMEM;
0314         goto out;
0315     }
0316 
0317     p->major = major;
0318     strlcpy(p->name, name, sizeof(p->name));
0319     p->next = NULL;
0320     index = major_to_index(major);
0321 
0322     for (n = &major_names[index]; *n; n = &(*n)->next) {
0323         if ((*n)->major == major)
0324             break;
0325     }
0326     if (!*n)
0327         *n = p;
0328     else
0329         ret = -EBUSY;
0330 
0331     if (ret < 0) {
0332         printk("register_blkdev: cannot get major %d for %s\n",
0333                major, name);
0334         kfree(p);
0335     }
0336 out:
0337     mutex_unlock(&block_class_lock);
0338     return ret;
0339 }
0340 
0341 EXPORT_SYMBOL(register_blkdev);
0342 
0343 void unregister_blkdev(unsigned int major, const char *name)
0344 {
0345     struct blk_major_name **n;
0346     struct blk_major_name *p = NULL;
0347     int index = major_to_index(major);
0348 
0349     mutex_lock(&block_class_lock);
0350     for (n = &major_names[index]; *n; n = &(*n)->next)
0351         if ((*n)->major == major)
0352             break;
0353     if (!*n || strcmp((*n)->name, name)) {
0354         WARN_ON(1);
0355     } else {
0356         p = *n;
0357         *n = p->next;
0358     }
0359     mutex_unlock(&block_class_lock);
0360     kfree(p);
0361 }
0362 
0363 EXPORT_SYMBOL(unregister_blkdev);
0364 
0365 static struct kobj_map *bdev_map;
0366 
0367 /**
0368  * blk_mangle_minor - scatter minor numbers apart
0369  * @minor: minor number to mangle
0370  *
0371  * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
0372  * is enabled.  Mangling twice gives the original value.
0373  *
0374  * RETURNS:
0375  * Mangled value.
0376  *
0377  * CONTEXT:
0378  * Don't care.
0379  */
0380 static int blk_mangle_minor(int minor)
0381 {
0382 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
0383     int i;
0384 
0385     for (i = 0; i < MINORBITS / 2; i++) {
0386         int low = minor & (1 << i);
0387         int high = minor & (1 << (MINORBITS - 1 - i));
0388         int distance = MINORBITS - 1 - 2 * i;
0389 
0390         minor ^= low | high;    /* clear both bits */
0391         low <<= distance;   /* swap the positions */
0392         high >>= distance;
0393         minor |= low | high;    /* and set */
0394     }
0395 #endif
0396     return minor;
0397 }
0398 
0399 /**
0400  * blk_alloc_devt - allocate a dev_t for a partition
0401  * @part: partition to allocate dev_t for
0402  * @devt: out parameter for resulting dev_t
0403  *
0404  * Allocate a dev_t for block device.
0405  *
0406  * RETURNS:
0407  * 0 on success, allocated dev_t is returned in *@devt.  -errno on
0408  * failure.
0409  *
0410  * CONTEXT:
0411  * Might sleep.
0412  */
0413 int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
0414 {
0415     struct gendisk *disk = part_to_disk(part);
0416     int idx;
0417 
0418     /* in consecutive minor range? */
0419     if (part->partno < disk->minors) {
0420         *devt = MKDEV(disk->major, disk->first_minor + part->partno);
0421         return 0;
0422     }
0423 
0424     /* allocate ext devt */
0425     idr_preload(GFP_KERNEL);
0426 
0427     spin_lock_bh(&ext_devt_lock);
0428     idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
0429     spin_unlock_bh(&ext_devt_lock);
0430 
0431     idr_preload_end();
0432     if (idx < 0)
0433         return idx == -ENOSPC ? -EBUSY : idx;
0434 
0435     *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
0436     return 0;
0437 }
0438 
0439 /**
0440  * blk_free_devt - free a dev_t
0441  * @devt: dev_t to free
0442  *
0443  * Free @devt which was allocated using blk_alloc_devt().
0444  *
0445  * CONTEXT:
0446  * Might sleep.
0447  */
0448 void blk_free_devt(dev_t devt)
0449 {
0450     if (devt == MKDEV(0, 0))
0451         return;
0452 
0453     if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
0454         spin_lock_bh(&ext_devt_lock);
0455         idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
0456         spin_unlock_bh(&ext_devt_lock);
0457     }
0458 }
0459 
0460 static char *bdevt_str(dev_t devt, char *buf)
0461 {
0462     if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
0463         char tbuf[BDEVT_SIZE];
0464         snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
0465         snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
0466     } else
0467         snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
0468 
0469     return buf;
0470 }
0471 
0472 /*
0473  * Register device numbers dev..(dev+range-1)
0474  * range must be nonzero
0475  * The hash chain is sorted on range, so that subranges can override.
0476  */
0477 void blk_register_region(dev_t devt, unsigned long range, struct module *module,
0478              struct kobject *(*probe)(dev_t, int *, void *),
0479              int (*lock)(dev_t, void *), void *data)
0480 {
0481     kobj_map(bdev_map, devt, range, module, probe, lock, data);
0482 }
0483 
0484 EXPORT_SYMBOL(blk_register_region);
0485 
0486 void blk_unregister_region(dev_t devt, unsigned long range)
0487 {
0488     kobj_unmap(bdev_map, devt, range);
0489 }
0490 
0491 EXPORT_SYMBOL(blk_unregister_region);
0492 
0493 static struct kobject *exact_match(dev_t devt, int *partno, void *data)
0494 {
0495     struct gendisk *p = data;
0496 
0497     return &disk_to_dev(p)->kobj;
0498 }
0499 
0500 static int exact_lock(dev_t devt, void *data)
0501 {
0502     struct gendisk *p = data;
0503 
0504     if (!get_disk(p))
0505         return -1;
0506     return 0;
0507 }
0508 
0509 static void register_disk(struct device *parent, struct gendisk *disk)
0510 {
0511     struct device *ddev = disk_to_dev(disk);
0512     struct block_device *bdev;
0513     struct disk_part_iter piter;
0514     struct hd_struct *part;
0515     int err;
0516 
0517     ddev->parent = parent;
0518 
0519     dev_set_name(ddev, "%s", disk->disk_name);
0520 
0521     /* delay uevents, until we scanned partition table */
0522     dev_set_uevent_suppress(ddev, 1);
0523 
0524     if (device_add(ddev))
0525         return;
0526     if (!sysfs_deprecated) {
0527         err = sysfs_create_link(block_depr, &ddev->kobj,
0528                     kobject_name(&ddev->kobj));
0529         if (err) {
0530             device_del(ddev);
0531             return;
0532         }
0533     }
0534 
0535     /*
0536      * avoid probable deadlock caused by allocating memory with
0537      * GFP_KERNEL in runtime_resume callback of its all ancestor
0538      * devices
0539      */
0540     pm_runtime_set_memalloc_noio(ddev, true);
0541 
0542     disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
0543     disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
0544 
0545     /* No minors to use for partitions */
0546     if (!disk_part_scan_enabled(disk))
0547         goto exit;
0548 
0549     /* No such device (e.g., media were just removed) */
0550     if (!get_capacity(disk))
0551         goto exit;
0552 
0553     bdev = bdget_disk(disk, 0);
0554     if (!bdev)
0555         goto exit;
0556 
0557     bdev->bd_invalidated = 1;
0558     err = blkdev_get(bdev, FMODE_READ, NULL);
0559     if (err < 0)
0560         goto exit;
0561     blkdev_put(bdev, FMODE_READ);
0562 
0563 exit:
0564     /* announce disk after possible partitions are created */
0565     dev_set_uevent_suppress(ddev, 0);
0566     kobject_uevent(&ddev->kobj, KOBJ_ADD);
0567 
0568     /* announce possible partitions */
0569     disk_part_iter_init(&piter, disk, 0);
0570     while ((part = disk_part_iter_next(&piter)))
0571         kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
0572     disk_part_iter_exit(&piter);
0573 }
0574 
0575 /**
0576  * device_add_disk - add partitioning information to kernel list
0577  * @parent: parent device for the disk
0578  * @disk: per-device partitioning information
0579  *
0580  * This function registers the partitioning information in @disk
0581  * with the kernel.
0582  *
0583  * FIXME: error handling
0584  */
0585 void device_add_disk(struct device *parent, struct gendisk *disk)
0586 {
0587     struct backing_dev_info *bdi;
0588     dev_t devt;
0589     int retval;
0590 
0591     /* minors == 0 indicates to use ext devt from part0 and should
0592      * be accompanied with EXT_DEVT flag.  Make sure all
0593      * parameters make sense.
0594      */
0595     WARN_ON(disk->minors && !(disk->major || disk->first_minor));
0596     WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
0597 
0598     disk->flags |= GENHD_FL_UP;
0599 
0600     retval = blk_alloc_devt(&disk->part0, &devt);
0601     if (retval) {
0602         WARN_ON(1);
0603         return;
0604     }
0605     disk_to_dev(disk)->devt = devt;
0606 
0607     /* ->major and ->first_minor aren't supposed to be
0608      * dereferenced from here on, but set them just in case.
0609      */
0610     disk->major = MAJOR(devt);
0611     disk->first_minor = MINOR(devt);
0612 
0613     disk_alloc_events(disk);
0614 
0615     /* Register BDI before referencing it from bdev */
0616     bdi = &disk->queue->backing_dev_info;
0617     bdi_register_owner(bdi, disk_to_dev(disk));
0618 
0619     blk_register_region(disk_devt(disk), disk->minors, NULL,
0620                 exact_match, exact_lock, disk);
0621     register_disk(parent, disk);
0622     blk_register_queue(disk);
0623 
0624     /*
0625      * Take an extra ref on queue which will be put on disk_release()
0626      * so that it sticks around as long as @disk is there.
0627      */
0628     WARN_ON_ONCE(!blk_get_queue(disk->queue));
0629 
0630     retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
0631                    "bdi");
0632     WARN_ON(retval);
0633 
0634     disk_add_events(disk);
0635     blk_integrity_add(disk);
0636 }
0637 EXPORT_SYMBOL(device_add_disk);
0638 
0639 void del_gendisk(struct gendisk *disk)
0640 {
0641     struct disk_part_iter piter;
0642     struct hd_struct *part;
0643 
0644     blk_integrity_del(disk);
0645     disk_del_events(disk);
0646 
0647     /* invalidate stuff */
0648     disk_part_iter_init(&piter, disk,
0649                  DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
0650     while ((part = disk_part_iter_next(&piter))) {
0651         invalidate_partition(disk, part->partno);
0652         delete_partition(disk, part->partno);
0653     }
0654     disk_part_iter_exit(&piter);
0655 
0656     invalidate_partition(disk, 0);
0657     set_capacity(disk, 0);
0658     disk->flags &= ~GENHD_FL_UP;
0659 
0660     sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
0661     blk_unregister_queue(disk);
0662     blk_unregister_region(disk_devt(disk), disk->minors);
0663 
0664     part_stat_set_all(&disk->part0, 0);
0665     disk->part0.stamp = 0;
0666 
0667     kobject_put(disk->part0.holder_dir);
0668     kobject_put(disk->slave_dir);
0669     if (!sysfs_deprecated)
0670         sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
0671     pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
0672     device_del(disk_to_dev(disk));
0673 }
0674 EXPORT_SYMBOL(del_gendisk);
0675 
0676 /* sysfs access to bad-blocks list. */
0677 static ssize_t disk_badblocks_show(struct device *dev,
0678                     struct device_attribute *attr,
0679                     char *page)
0680 {
0681     struct gendisk *disk = dev_to_disk(dev);
0682 
0683     if (!disk->bb)
0684         return sprintf(page, "\n");
0685 
0686     return badblocks_show(disk->bb, page, 0);
0687 }
0688 
0689 static ssize_t disk_badblocks_store(struct device *dev,
0690                     struct device_attribute *attr,
0691                     const char *page, size_t len)
0692 {
0693     struct gendisk *disk = dev_to_disk(dev);
0694 
0695     if (!disk->bb)
0696         return -ENXIO;
0697 
0698     return badblocks_store(disk->bb, page, len, 0);
0699 }
0700 
0701 /**
0702  * get_gendisk - get partitioning information for a given device
0703  * @devt: device to get partitioning information for
0704  * @partno: returned partition index
0705  *
0706  * This function gets the structure containing partitioning
0707  * information for the given device @devt.
0708  */
0709 struct gendisk *get_gendisk(dev_t devt, int *partno)
0710 {
0711     struct gendisk *disk = NULL;
0712 
0713     if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
0714         struct kobject *kobj;
0715 
0716         kobj = kobj_lookup(bdev_map, devt, partno);
0717         if (kobj)
0718             disk = dev_to_disk(kobj_to_dev(kobj));
0719     } else {
0720         struct hd_struct *part;
0721 
0722         spin_lock_bh(&ext_devt_lock);
0723         part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
0724         if (part && get_disk(part_to_disk(part))) {
0725             *partno = part->partno;
0726             disk = part_to_disk(part);
0727         }
0728         spin_unlock_bh(&ext_devt_lock);
0729     }
0730 
0731     return disk;
0732 }
0733 EXPORT_SYMBOL(get_gendisk);
0734 
0735 /**
0736  * bdget_disk - do bdget() by gendisk and partition number
0737  * @disk: gendisk of interest
0738  * @partno: partition number
0739  *
0740  * Find partition @partno from @disk, do bdget() on it.
0741  *
0742  * CONTEXT:
0743  * Don't care.
0744  *
0745  * RETURNS:
0746  * Resulting block_device on success, NULL on failure.
0747  */
0748 struct block_device *bdget_disk(struct gendisk *disk, int partno)
0749 {
0750     struct hd_struct *part;
0751     struct block_device *bdev = NULL;
0752 
0753     part = disk_get_part(disk, partno);
0754     if (part)
0755         bdev = bdget(part_devt(part));
0756     disk_put_part(part);
0757 
0758     return bdev;
0759 }
0760 EXPORT_SYMBOL(bdget_disk);
0761 
0762 /*
0763  * print a full list of all partitions - intended for places where the root
0764  * filesystem can't be mounted and thus to give the victim some idea of what
0765  * went wrong
0766  */
0767 void __init printk_all_partitions(void)
0768 {
0769     struct class_dev_iter iter;
0770     struct device *dev;
0771 
0772     class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
0773     while ((dev = class_dev_iter_next(&iter))) {
0774         struct gendisk *disk = dev_to_disk(dev);
0775         struct disk_part_iter piter;
0776         struct hd_struct *part;
0777         char name_buf[BDEVNAME_SIZE];
0778         char devt_buf[BDEVT_SIZE];
0779 
0780         /*
0781          * Don't show empty devices or things that have been
0782          * suppressed
0783          */
0784         if (get_capacity(disk) == 0 ||
0785             (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
0786             continue;
0787 
0788         /*
0789          * Note, unlike /proc/partitions, I am showing the
0790          * numbers in hex - the same format as the root=
0791          * option takes.
0792          */
0793         disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
0794         while ((part = disk_part_iter_next(&piter))) {
0795             bool is_part0 = part == &disk->part0;
0796 
0797             printk("%s%s %10llu %s %s", is_part0 ? "" : "  ",
0798                    bdevt_str(part_devt(part), devt_buf),
0799                    (unsigned long long)part_nr_sects_read(part) >> 1
0800                    , disk_name(disk, part->partno, name_buf),
0801                    part->info ? part->info->uuid : "");
0802             if (is_part0) {
0803                 if (dev->parent && dev->parent->driver)
0804                     printk(" driver: %s\n",
0805                           dev->parent->driver->name);
0806                 else
0807                     printk(" (driver?)\n");
0808             } else
0809                 printk("\n");
0810         }
0811         disk_part_iter_exit(&piter);
0812     }
0813     class_dev_iter_exit(&iter);
0814 }
0815 
0816 #ifdef CONFIG_PROC_FS
0817 /* iterator */
0818 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
0819 {
0820     loff_t skip = *pos;
0821     struct class_dev_iter *iter;
0822     struct device *dev;
0823 
0824     iter = kmalloc(sizeof(*iter), GFP_KERNEL);
0825     if (!iter)
0826         return ERR_PTR(-ENOMEM);
0827 
0828     seqf->private = iter;
0829     class_dev_iter_init(iter, &block_class, NULL, &disk_type);
0830     do {
0831         dev = class_dev_iter_next(iter);
0832         if (!dev)
0833             return NULL;
0834     } while (skip--);
0835 
0836     return dev_to_disk(dev);
0837 }
0838 
0839 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
0840 {
0841     struct device *dev;
0842 
0843     (*pos)++;
0844     dev = class_dev_iter_next(seqf->private);
0845     if (dev)
0846         return dev_to_disk(dev);
0847 
0848     return NULL;
0849 }
0850 
0851 static void disk_seqf_stop(struct seq_file *seqf, void *v)
0852 {
0853     struct class_dev_iter *iter = seqf->private;
0854 
0855     /* stop is called even after start failed :-( */
0856     if (iter) {
0857         class_dev_iter_exit(iter);
0858         kfree(iter);
0859         seqf->private = NULL;
0860     }
0861 }
0862 
0863 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
0864 {
0865     void *p;
0866 
0867     p = disk_seqf_start(seqf, pos);
0868     if (!IS_ERR_OR_NULL(p) && !*pos)
0869         seq_puts(seqf, "major minor  #blocks  name\n\n");
0870     return p;
0871 }
0872 
0873 static int show_partition(struct seq_file *seqf, void *v)
0874 {
0875     struct gendisk *sgp = v;
0876     struct disk_part_iter piter;
0877     struct hd_struct *part;
0878     char buf[BDEVNAME_SIZE];
0879 
0880     /* Don't show non-partitionable removeable devices or empty devices */
0881     if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
0882                    (sgp->flags & GENHD_FL_REMOVABLE)))
0883         return 0;
0884     if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
0885         return 0;
0886 
0887     /* show the full disk and all non-0 size partitions of it */
0888     disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
0889     while ((part = disk_part_iter_next(&piter)))
0890         seq_printf(seqf, "%4d  %7d %10llu %s\n",
0891                MAJOR(part_devt(part)), MINOR(part_devt(part)),
0892                (unsigned long long)part_nr_sects_read(part) >> 1,
0893                disk_name(sgp, part->partno, buf));
0894     disk_part_iter_exit(&piter);
0895 
0896     return 0;
0897 }
0898 
0899 static const struct seq_operations partitions_op = {
0900     .start  = show_partition_start,
0901     .next   = disk_seqf_next,
0902     .stop   = disk_seqf_stop,
0903     .show   = show_partition
0904 };
0905 
0906 static int partitions_open(struct inode *inode, struct file *file)
0907 {
0908     return seq_open(file, &partitions_op);
0909 }
0910 
0911 static const struct file_operations proc_partitions_operations = {
0912     .open       = partitions_open,
0913     .read       = seq_read,
0914     .llseek     = seq_lseek,
0915     .release    = seq_release,
0916 };
0917 #endif
0918 
0919 
0920 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
0921 {
0922     if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
0923         /* Make old-style 2.4 aliases work */
0924         request_module("block-major-%d", MAJOR(devt));
0925     return NULL;
0926 }
0927 
0928 static int __init genhd_device_init(void)
0929 {
0930     int error;
0931 
0932     block_class.dev_kobj = sysfs_dev_block_kobj;
0933     error = class_register(&block_class);
0934     if (unlikely(error))
0935         return error;
0936     bdev_map = kobj_map_init(base_probe, &block_class_lock);
0937     blk_dev_init();
0938 
0939     register_blkdev(BLOCK_EXT_MAJOR, "blkext");
0940 
0941     /* create top-level block dir */
0942     if (!sysfs_deprecated)
0943         block_depr = kobject_create_and_add("block", NULL);
0944     return 0;
0945 }
0946 
0947 subsys_initcall(genhd_device_init);
0948 
0949 static ssize_t disk_range_show(struct device *dev,
0950                    struct device_attribute *attr, char *buf)
0951 {
0952     struct gendisk *disk = dev_to_disk(dev);
0953 
0954     return sprintf(buf, "%d\n", disk->minors);
0955 }
0956 
0957 static ssize_t disk_ext_range_show(struct device *dev,
0958                    struct device_attribute *attr, char *buf)
0959 {
0960     struct gendisk *disk = dev_to_disk(dev);
0961 
0962     return sprintf(buf, "%d\n", disk_max_parts(disk));
0963 }
0964 
0965 static ssize_t disk_removable_show(struct device *dev,
0966                    struct device_attribute *attr, char *buf)
0967 {
0968     struct gendisk *disk = dev_to_disk(dev);
0969 
0970     return sprintf(buf, "%d\n",
0971                (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
0972 }
0973 
0974 static ssize_t disk_ro_show(struct device *dev,
0975                    struct device_attribute *attr, char *buf)
0976 {
0977     struct gendisk *disk = dev_to_disk(dev);
0978 
0979     return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
0980 }
0981 
0982 static ssize_t disk_capability_show(struct device *dev,
0983                     struct device_attribute *attr, char *buf)
0984 {
0985     struct gendisk *disk = dev_to_disk(dev);
0986 
0987     return sprintf(buf, "%x\n", disk->flags);
0988 }
0989 
0990 static ssize_t disk_alignment_offset_show(struct device *dev,
0991                       struct device_attribute *attr,
0992                       char *buf)
0993 {
0994     struct gendisk *disk = dev_to_disk(dev);
0995 
0996     return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
0997 }
0998 
0999 static ssize_t disk_discard_alignment_show(struct device *dev,
1000                        struct device_attribute *attr,
1001                        char *buf)
1002 {
1003     struct gendisk *disk = dev_to_disk(dev);
1004 
1005     return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1006 }
1007 
1008 static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
1009 static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
1010 static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
1011 static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
1012 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
1013 static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
1014 static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
1015            NULL);
1016 static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
1017 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
1018 static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
1019 static DEVICE_ATTR(badblocks, S_IRUGO | S_IWUSR, disk_badblocks_show,
1020         disk_badblocks_store);
1021 #ifdef CONFIG_FAIL_MAKE_REQUEST
1022 static struct device_attribute dev_attr_fail =
1023     __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
1024 #endif
1025 #ifdef CONFIG_FAIL_IO_TIMEOUT
1026 static struct device_attribute dev_attr_fail_timeout =
1027     __ATTR(io-timeout-fail,  S_IRUGO|S_IWUSR, part_timeout_show,
1028         part_timeout_store);
1029 #endif
1030 
1031 static struct attribute *disk_attrs[] = {
1032     &dev_attr_range.attr,
1033     &dev_attr_ext_range.attr,
1034     &dev_attr_removable.attr,
1035     &dev_attr_ro.attr,
1036     &dev_attr_size.attr,
1037     &dev_attr_alignment_offset.attr,
1038     &dev_attr_discard_alignment.attr,
1039     &dev_attr_capability.attr,
1040     &dev_attr_stat.attr,
1041     &dev_attr_inflight.attr,
1042     &dev_attr_badblocks.attr,
1043 #ifdef CONFIG_FAIL_MAKE_REQUEST
1044     &dev_attr_fail.attr,
1045 #endif
1046 #ifdef CONFIG_FAIL_IO_TIMEOUT
1047     &dev_attr_fail_timeout.attr,
1048 #endif
1049     NULL
1050 };
1051 
1052 static struct attribute_group disk_attr_group = {
1053     .attrs = disk_attrs,
1054 };
1055 
1056 static const struct attribute_group *disk_attr_groups[] = {
1057     &disk_attr_group,
1058     NULL
1059 };
1060 
1061 /**
1062  * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1063  * @disk: disk to replace part_tbl for
1064  * @new_ptbl: new part_tbl to install
1065  *
1066  * Replace disk->part_tbl with @new_ptbl in RCU-safe way.  The
1067  * original ptbl is freed using RCU callback.
1068  *
1069  * LOCKING:
1070  * Matching bd_mutx locked.
1071  */
1072 static void disk_replace_part_tbl(struct gendisk *disk,
1073                   struct disk_part_tbl *new_ptbl)
1074 {
1075     struct disk_part_tbl *old_ptbl = disk->part_tbl;
1076 
1077     rcu_assign_pointer(disk->part_tbl, new_ptbl);
1078 
1079     if (old_ptbl) {
1080         rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1081         kfree_rcu(old_ptbl, rcu_head);
1082     }
1083 }
1084 
1085 /**
1086  * disk_expand_part_tbl - expand disk->part_tbl
1087  * @disk: disk to expand part_tbl for
1088  * @partno: expand such that this partno can fit in
1089  *
1090  * Expand disk->part_tbl such that @partno can fit in.  disk->part_tbl
1091  * uses RCU to allow unlocked dereferencing for stats and other stuff.
1092  *
1093  * LOCKING:
1094  * Matching bd_mutex locked, might sleep.
1095  *
1096  * RETURNS:
1097  * 0 on success, -errno on failure.
1098  */
1099 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1100 {
1101     struct disk_part_tbl *old_ptbl = disk->part_tbl;
1102     struct disk_part_tbl *new_ptbl;
1103     int len = old_ptbl ? old_ptbl->len : 0;
1104     int i, target;
1105     size_t size;
1106 
1107     /*
1108      * check for int overflow, since we can get here from blkpg_ioctl()
1109      * with a user passed 'partno'.
1110      */
1111     target = partno + 1;
1112     if (target < 0)
1113         return -EINVAL;
1114 
1115     /* disk_max_parts() is zero during initialization, ignore if so */
1116     if (disk_max_parts(disk) && target > disk_max_parts(disk))
1117         return -EINVAL;
1118 
1119     if (target <= len)
1120         return 0;
1121 
1122     size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1123     new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1124     if (!new_ptbl)
1125         return -ENOMEM;
1126 
1127     new_ptbl->len = target;
1128 
1129     for (i = 0; i < len; i++)
1130         rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1131 
1132     disk_replace_part_tbl(disk, new_ptbl);
1133     return 0;
1134 }
1135 
1136 static void disk_release(struct device *dev)
1137 {
1138     struct gendisk *disk = dev_to_disk(dev);
1139 
1140     blk_free_devt(dev->devt);
1141     disk_release_events(disk);
1142     kfree(disk->random);
1143     disk_replace_part_tbl(disk, NULL);
1144     hd_free_part(&disk->part0);
1145     if (disk->queue)
1146         blk_put_queue(disk->queue);
1147     kfree(disk);
1148 }
1149 struct class block_class = {
1150     .name       = "block",
1151 };
1152 
1153 static char *block_devnode(struct device *dev, umode_t *mode,
1154                kuid_t *uid, kgid_t *gid)
1155 {
1156     struct gendisk *disk = dev_to_disk(dev);
1157 
1158     if (disk->devnode)
1159         return disk->devnode(disk, mode);
1160     return NULL;
1161 }
1162 
1163 static struct device_type disk_type = {
1164     .name       = "disk",
1165     .groups     = disk_attr_groups,
1166     .release    = disk_release,
1167     .devnode    = block_devnode,
1168 };
1169 
1170 #ifdef CONFIG_PROC_FS
1171 /*
1172  * aggregate disk stat collector.  Uses the same stats that the sysfs
1173  * entries do, above, but makes them available through one seq_file.
1174  *
1175  * The output looks suspiciously like /proc/partitions with a bunch of
1176  * extra fields.
1177  */
1178 static int diskstats_show(struct seq_file *seqf, void *v)
1179 {
1180     struct gendisk *gp = v;
1181     struct disk_part_iter piter;
1182     struct hd_struct *hd;
1183     char buf[BDEVNAME_SIZE];
1184     int cpu;
1185 
1186     /*
1187     if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1188         seq_puts(seqf,  "major minor name"
1189                 "     rio rmerge rsect ruse wio wmerge "
1190                 "wsect wuse running use aveq"
1191                 "\n\n");
1192     */
1193 
1194     disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1195     while ((hd = disk_part_iter_next(&piter))) {
1196         cpu = part_stat_lock();
1197         part_round_stats(cpu, hd);
1198         part_stat_unlock();
1199         seq_printf(seqf, "%4d %7d %s %lu %lu %lu "
1200                "%u %lu %lu %lu %u %u %u %u\n",
1201                MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1202                disk_name(gp, hd->partno, buf),
1203                part_stat_read(hd, ios[READ]),
1204                part_stat_read(hd, merges[READ]),
1205                part_stat_read(hd, sectors[READ]),
1206                jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1207                part_stat_read(hd, ios[WRITE]),
1208                part_stat_read(hd, merges[WRITE]),
1209                part_stat_read(hd, sectors[WRITE]),
1210                jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1211                part_in_flight(hd),
1212                jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1213                jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1214             );
1215     }
1216     disk_part_iter_exit(&piter);
1217 
1218     return 0;
1219 }
1220 
1221 static const struct seq_operations diskstats_op = {
1222     .start  = disk_seqf_start,
1223     .next   = disk_seqf_next,
1224     .stop   = disk_seqf_stop,
1225     .show   = diskstats_show
1226 };
1227 
1228 static int diskstats_open(struct inode *inode, struct file *file)
1229 {
1230     return seq_open(file, &diskstats_op);
1231 }
1232 
1233 static const struct file_operations proc_diskstats_operations = {
1234     .open       = diskstats_open,
1235     .read       = seq_read,
1236     .llseek     = seq_lseek,
1237     .release    = seq_release,
1238 };
1239 
1240 static int __init proc_genhd_init(void)
1241 {
1242     proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1243     proc_create("partitions", 0, NULL, &proc_partitions_operations);
1244     return 0;
1245 }
1246 module_init(proc_genhd_init);
1247 #endif /* CONFIG_PROC_FS */
1248 
1249 dev_t blk_lookup_devt(const char *name, int partno)
1250 {
1251     dev_t devt = MKDEV(0, 0);
1252     struct class_dev_iter iter;
1253     struct device *dev;
1254 
1255     class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1256     while ((dev = class_dev_iter_next(&iter))) {
1257         struct gendisk *disk = dev_to_disk(dev);
1258         struct hd_struct *part;
1259 
1260         if (strcmp(dev_name(dev), name))
1261             continue;
1262 
1263         if (partno < disk->minors) {
1264             /* We need to return the right devno, even
1265              * if the partition doesn't exist yet.
1266              */
1267             devt = MKDEV(MAJOR(dev->devt),
1268                      MINOR(dev->devt) + partno);
1269             break;
1270         }
1271         part = disk_get_part(disk, partno);
1272         if (part) {
1273             devt = part_devt(part);
1274             disk_put_part(part);
1275             break;
1276         }
1277         disk_put_part(part);
1278     }
1279     class_dev_iter_exit(&iter);
1280     return devt;
1281 }
1282 EXPORT_SYMBOL(blk_lookup_devt);
1283 
1284 struct gendisk *alloc_disk(int minors)
1285 {
1286     return alloc_disk_node(minors, NUMA_NO_NODE);
1287 }
1288 EXPORT_SYMBOL(alloc_disk);
1289 
1290 struct gendisk *alloc_disk_node(int minors, int node_id)
1291 {
1292     struct gendisk *disk;
1293 
1294     disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1295     if (disk) {
1296         if (!init_part_stats(&disk->part0)) {
1297             kfree(disk);
1298             return NULL;
1299         }
1300         disk->node_id = node_id;
1301         if (disk_expand_part_tbl(disk, 0)) {
1302             free_part_stats(&disk->part0);
1303             kfree(disk);
1304             return NULL;
1305         }
1306         disk->part_tbl->part[0] = &disk->part0;
1307 
1308         /*
1309          * set_capacity() and get_capacity() currently don't use
1310          * seqcounter to read/update the part0->nr_sects. Still init
1311          * the counter as we can read the sectors in IO submission
1312          * patch using seqence counters.
1313          *
1314          * TODO: Ideally set_capacity() and get_capacity() should be
1315          * converted to make use of bd_mutex and sequence counters.
1316          */
1317         seqcount_init(&disk->part0.nr_sects_seq);
1318         if (hd_ref_init(&disk->part0)) {
1319             hd_free_part(&disk->part0);
1320             kfree(disk);
1321             return NULL;
1322         }
1323 
1324         disk->minors = minors;
1325         rand_initialize_disk(disk);
1326         disk_to_dev(disk)->class = &block_class;
1327         disk_to_dev(disk)->type = &disk_type;
1328         device_initialize(disk_to_dev(disk));
1329     }
1330     return disk;
1331 }
1332 EXPORT_SYMBOL(alloc_disk_node);
1333 
1334 struct kobject *get_disk(struct gendisk *disk)
1335 {
1336     struct module *owner;
1337     struct kobject *kobj;
1338 
1339     if (!disk->fops)
1340         return NULL;
1341     owner = disk->fops->owner;
1342     if (owner && !try_module_get(owner))
1343         return NULL;
1344     kobj = kobject_get(&disk_to_dev(disk)->kobj);
1345     if (kobj == NULL) {
1346         module_put(owner);
1347         return NULL;
1348     }
1349     return kobj;
1350 
1351 }
1352 
1353 EXPORT_SYMBOL(get_disk);
1354 
1355 void put_disk(struct gendisk *disk)
1356 {
1357     if (disk)
1358         kobject_put(&disk_to_dev(disk)->kobj);
1359 }
1360 
1361 EXPORT_SYMBOL(put_disk);
1362 
1363 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1364 {
1365     char event[] = "DISK_RO=1";
1366     char *envp[] = { event, NULL };
1367 
1368     if (!ro)
1369         event[8] = '0';
1370     kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1371 }
1372 
1373 void set_device_ro(struct block_device *bdev, int flag)
1374 {
1375     bdev->bd_part->policy = flag;
1376 }
1377 
1378 EXPORT_SYMBOL(set_device_ro);
1379 
1380 void set_disk_ro(struct gendisk *disk, int flag)
1381 {
1382     struct disk_part_iter piter;
1383     struct hd_struct *part;
1384 
1385     if (disk->part0.policy != flag) {
1386         set_disk_ro_uevent(disk, flag);
1387         disk->part0.policy = flag;
1388     }
1389 
1390     disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1391     while ((part = disk_part_iter_next(&piter)))
1392         part->policy = flag;
1393     disk_part_iter_exit(&piter);
1394 }
1395 
1396 EXPORT_SYMBOL(set_disk_ro);
1397 
1398 int bdev_read_only(struct block_device *bdev)
1399 {
1400     if (!bdev)
1401         return 0;
1402     return bdev->bd_part->policy;
1403 }
1404 
1405 EXPORT_SYMBOL(bdev_read_only);
1406 
1407 int invalidate_partition(struct gendisk *disk, int partno)
1408 {
1409     int res = 0;
1410     struct block_device *bdev = bdget_disk(disk, partno);
1411     if (bdev) {
1412         fsync_bdev(bdev);
1413         res = __invalidate_device(bdev, true);
1414         bdput(bdev);
1415     }
1416     return res;
1417 }
1418 
1419 EXPORT_SYMBOL(invalidate_partition);
1420 
1421 /*
1422  * Disk events - monitor disk events like media change and eject request.
1423  */
1424 struct disk_events {
1425     struct list_head    node;       /* all disk_event's */
1426     struct gendisk      *disk;      /* the associated disk */
1427     spinlock_t      lock;
1428 
1429     struct mutex        block_mutex;    /* protects blocking */
1430     int         block;      /* event blocking depth */
1431     unsigned int        pending;    /* events already sent out */
1432     unsigned int        clearing;   /* events being cleared */
1433 
1434     long            poll_msecs; /* interval, -1 for default */
1435     struct delayed_work dwork;
1436 };
1437 
1438 static const char *disk_events_strs[] = {
1439     [ilog2(DISK_EVENT_MEDIA_CHANGE)]    = "media_change",
1440     [ilog2(DISK_EVENT_EJECT_REQUEST)]   = "eject_request",
1441 };
1442 
1443 static char *disk_uevents[] = {
1444     [ilog2(DISK_EVENT_MEDIA_CHANGE)]    = "DISK_MEDIA_CHANGE=1",
1445     [ilog2(DISK_EVENT_EJECT_REQUEST)]   = "DISK_EJECT_REQUEST=1",
1446 };
1447 
1448 /* list of all disk_events */
1449 static DEFINE_MUTEX(disk_events_mutex);
1450 static LIST_HEAD(disk_events);
1451 
1452 /* disable in-kernel polling by default */
1453 static unsigned long disk_events_dfl_poll_msecs;
1454 
1455 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1456 {
1457     struct disk_events *ev = disk->ev;
1458     long intv_msecs = 0;
1459 
1460     /*
1461      * If device-specific poll interval is set, always use it.  If
1462      * the default is being used, poll iff there are events which
1463      * can't be monitored asynchronously.
1464      */
1465     if (ev->poll_msecs >= 0)
1466         intv_msecs = ev->poll_msecs;
1467     else if (disk->events & ~disk->async_events)
1468         intv_msecs = disk_events_dfl_poll_msecs;
1469 
1470     return msecs_to_jiffies(intv_msecs);
1471 }
1472 
1473 /**
1474  * disk_block_events - block and flush disk event checking
1475  * @disk: disk to block events for
1476  *
1477  * On return from this function, it is guaranteed that event checking
1478  * isn't in progress and won't happen until unblocked by
1479  * disk_unblock_events().  Events blocking is counted and the actual
1480  * unblocking happens after the matching number of unblocks are done.
1481  *
1482  * Note that this intentionally does not block event checking from
1483  * disk_clear_events().
1484  *
1485  * CONTEXT:
1486  * Might sleep.
1487  */
1488 void disk_block_events(struct gendisk *disk)
1489 {
1490     struct disk_events *ev = disk->ev;
1491     unsigned long flags;
1492     bool cancel;
1493 
1494     if (!ev)
1495         return;
1496 
1497     /*
1498      * Outer mutex ensures that the first blocker completes canceling
1499      * the event work before further blockers are allowed to finish.
1500      */
1501     mutex_lock(&ev->block_mutex);
1502 
1503     spin_lock_irqsave(&ev->lock, flags);
1504     cancel = !ev->block++;
1505     spin_unlock_irqrestore(&ev->lock, flags);
1506 
1507     if (cancel)
1508         cancel_delayed_work_sync(&disk->ev->dwork);
1509 
1510     mutex_unlock(&ev->block_mutex);
1511 }
1512 
1513 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1514 {
1515     struct disk_events *ev = disk->ev;
1516     unsigned long intv;
1517     unsigned long flags;
1518 
1519     spin_lock_irqsave(&ev->lock, flags);
1520 
1521     if (WARN_ON_ONCE(ev->block <= 0))
1522         goto out_unlock;
1523 
1524     if (--ev->block)
1525         goto out_unlock;
1526 
1527     intv = disk_events_poll_jiffies(disk);
1528     if (check_now)
1529         queue_delayed_work(system_freezable_power_efficient_wq,
1530                 &ev->dwork, 0);
1531     else if (intv)
1532         queue_delayed_work(system_freezable_power_efficient_wq,
1533                 &ev->dwork, intv);
1534 out_unlock:
1535     spin_unlock_irqrestore(&ev->lock, flags);
1536 }
1537 
1538 /**
1539  * disk_unblock_events - unblock disk event checking
1540  * @disk: disk to unblock events for
1541  *
1542  * Undo disk_block_events().  When the block count reaches zero, it
1543  * starts events polling if configured.
1544  *
1545  * CONTEXT:
1546  * Don't care.  Safe to call from irq context.
1547  */
1548 void disk_unblock_events(struct gendisk *disk)
1549 {
1550     if (disk->ev)
1551         __disk_unblock_events(disk, false);
1552 }
1553 
1554 /**
1555  * disk_flush_events - schedule immediate event checking and flushing
1556  * @disk: disk to check and flush events for
1557  * @mask: events to flush
1558  *
1559  * Schedule immediate event checking on @disk if not blocked.  Events in
1560  * @mask are scheduled to be cleared from the driver.  Note that this
1561  * doesn't clear the events from @disk->ev.
1562  *
1563  * CONTEXT:
1564  * If @mask is non-zero must be called with bdev->bd_mutex held.
1565  */
1566 void disk_flush_events(struct gendisk *disk, unsigned int mask)
1567 {
1568     struct disk_events *ev = disk->ev;
1569 
1570     if (!ev)
1571         return;
1572 
1573     spin_lock_irq(&ev->lock);
1574     ev->clearing |= mask;
1575     if (!ev->block)
1576         mod_delayed_work(system_freezable_power_efficient_wq,
1577                 &ev->dwork, 0);
1578     spin_unlock_irq(&ev->lock);
1579 }
1580 
1581 /**
1582  * disk_clear_events - synchronously check, clear and return pending events
1583  * @disk: disk to fetch and clear events from
1584  * @mask: mask of events to be fetched and cleared
1585  *
1586  * Disk events are synchronously checked and pending events in @mask
1587  * are cleared and returned.  This ignores the block count.
1588  *
1589  * CONTEXT:
1590  * Might sleep.
1591  */
1592 unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1593 {
1594     const struct block_device_operations *bdops = disk->fops;
1595     struct disk_events *ev = disk->ev;
1596     unsigned int pending;
1597     unsigned int clearing = mask;
1598 
1599     if (!ev) {
1600         /* for drivers still using the old ->media_changed method */
1601         if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1602             bdops->media_changed && bdops->media_changed(disk))
1603             return DISK_EVENT_MEDIA_CHANGE;
1604         return 0;
1605     }
1606 
1607     disk_block_events(disk);
1608 
1609     /*
1610      * store the union of mask and ev->clearing on the stack so that the
1611      * race with disk_flush_events does not cause ambiguity (ev->clearing
1612      * can still be modified even if events are blocked).
1613      */
1614     spin_lock_irq(&ev->lock);
1615     clearing |= ev->clearing;
1616     ev->clearing = 0;
1617     spin_unlock_irq(&ev->lock);
1618 
1619     disk_check_events(ev, &clearing);
1620     /*
1621      * if ev->clearing is not 0, the disk_flush_events got called in the
1622      * middle of this function, so we want to run the workfn without delay.
1623      */
1624     __disk_unblock_events(disk, ev->clearing ? true : false);
1625 
1626     /* then, fetch and clear pending events */
1627     spin_lock_irq(&ev->lock);
1628     pending = ev->pending & mask;
1629     ev->pending &= ~mask;
1630     spin_unlock_irq(&ev->lock);
1631     WARN_ON_ONCE(clearing & mask);
1632 
1633     return pending;
1634 }
1635 
1636 /*
1637  * Separate this part out so that a different pointer for clearing_ptr can be
1638  * passed in for disk_clear_events.
1639  */
1640 static void disk_events_workfn(struct work_struct *work)
1641 {
1642     struct delayed_work *dwork = to_delayed_work(work);
1643     struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1644 
1645     disk_check_events(ev, &ev->clearing);
1646 }
1647 
1648 static void disk_check_events(struct disk_events *ev,
1649                   unsigned int *clearing_ptr)
1650 {
1651     struct gendisk *disk = ev->disk;
1652     char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1653     unsigned int clearing = *clearing_ptr;
1654     unsigned int events;
1655     unsigned long intv;
1656     int nr_events = 0, i;
1657 
1658     /* check events */
1659     events = disk->fops->check_events(disk, clearing);
1660 
1661     /* accumulate pending events and schedule next poll if necessary */
1662     spin_lock_irq(&ev->lock);
1663 
1664     events &= ~ev->pending;
1665     ev->pending |= events;
1666     *clearing_ptr &= ~clearing;
1667 
1668     intv = disk_events_poll_jiffies(disk);
1669     if (!ev->block && intv)
1670         queue_delayed_work(system_freezable_power_efficient_wq,
1671                 &ev->dwork, intv);
1672 
1673     spin_unlock_irq(&ev->lock);
1674 
1675     /*
1676      * Tell userland about new events.  Only the events listed in
1677      * @disk->events are reported.  Unlisted events are processed the
1678      * same internally but never get reported to userland.
1679      */
1680     for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1681         if (events & disk->events & (1 << i))
1682             envp[nr_events++] = disk_uevents[i];
1683 
1684     if (nr_events)
1685         kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1686 }
1687 
1688 /*
1689  * A disk events enabled device has the following sysfs nodes under
1690  * its /sys/block/X/ directory.
1691  *
1692  * events       : list of all supported events
1693  * events_async     : list of events which can be detected w/o polling
1694  * events_poll_msecs    : polling interval, 0: disable, -1: system default
1695  */
1696 static ssize_t __disk_events_show(unsigned int events, char *buf)
1697 {
1698     const char *delim = "";
1699     ssize_t pos = 0;
1700     int i;
1701 
1702     for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1703         if (events & (1 << i)) {
1704             pos += sprintf(buf + pos, "%s%s",
1705                        delim, disk_events_strs[i]);
1706             delim = " ";
1707         }
1708     if (pos)
1709         pos += sprintf(buf + pos, "\n");
1710     return pos;
1711 }
1712 
1713 static ssize_t disk_events_show(struct device *dev,
1714                 struct device_attribute *attr, char *buf)
1715 {
1716     struct gendisk *disk = dev_to_disk(dev);
1717 
1718     return __disk_events_show(disk->events, buf);
1719 }
1720 
1721 static ssize_t disk_events_async_show(struct device *dev,
1722                       struct device_attribute *attr, char *buf)
1723 {
1724     struct gendisk *disk = dev_to_disk(dev);
1725 
1726     return __disk_events_show(disk->async_events, buf);
1727 }
1728 
1729 static ssize_t disk_events_poll_msecs_show(struct device *dev,
1730                        struct device_attribute *attr,
1731                        char *buf)
1732 {
1733     struct gendisk *disk = dev_to_disk(dev);
1734 
1735     return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1736 }
1737 
1738 static ssize_t disk_events_poll_msecs_store(struct device *dev,
1739                         struct device_attribute *attr,
1740                         const char *buf, size_t count)
1741 {
1742     struct gendisk *disk = dev_to_disk(dev);
1743     long intv;
1744 
1745     if (!count || !sscanf(buf, "%ld", &intv))
1746         return -EINVAL;
1747 
1748     if (intv < 0 && intv != -1)
1749         return -EINVAL;
1750 
1751     disk_block_events(disk);
1752     disk->ev->poll_msecs = intv;
1753     __disk_unblock_events(disk, true);
1754 
1755     return count;
1756 }
1757 
1758 static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1759 static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1760 static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1761              disk_events_poll_msecs_show,
1762              disk_events_poll_msecs_store);
1763 
1764 static const struct attribute *disk_events_attrs[] = {
1765     &dev_attr_events.attr,
1766     &dev_attr_events_async.attr,
1767     &dev_attr_events_poll_msecs.attr,
1768     NULL,
1769 };
1770 
1771 /*
1772  * The default polling interval can be specified by the kernel
1773  * parameter block.events_dfl_poll_msecs which defaults to 0
1774  * (disable).  This can also be modified runtime by writing to
1775  * /sys/module/block/events_dfl_poll_msecs.
1776  */
1777 static int disk_events_set_dfl_poll_msecs(const char *val,
1778                       const struct kernel_param *kp)
1779 {
1780     struct disk_events *ev;
1781     int ret;
1782 
1783     ret = param_set_ulong(val, kp);
1784     if (ret < 0)
1785         return ret;
1786 
1787     mutex_lock(&disk_events_mutex);
1788 
1789     list_for_each_entry(ev, &disk_events, node)
1790         disk_flush_events(ev->disk, 0);
1791 
1792     mutex_unlock(&disk_events_mutex);
1793 
1794     return 0;
1795 }
1796 
1797 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1798     .set    = disk_events_set_dfl_poll_msecs,
1799     .get    = param_get_ulong,
1800 };
1801 
1802 #undef MODULE_PARAM_PREFIX
1803 #define MODULE_PARAM_PREFIX "block."
1804 
1805 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1806         &disk_events_dfl_poll_msecs, 0644);
1807 
1808 /*
1809  * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
1810  */
1811 static void disk_alloc_events(struct gendisk *disk)
1812 {
1813     struct disk_events *ev;
1814 
1815     if (!disk->fops->check_events)
1816         return;
1817 
1818     ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1819     if (!ev) {
1820         pr_warn("%s: failed to initialize events\n", disk->disk_name);
1821         return;
1822     }
1823 
1824     INIT_LIST_HEAD(&ev->node);
1825     ev->disk = disk;
1826     spin_lock_init(&ev->lock);
1827     mutex_init(&ev->block_mutex);
1828     ev->block = 1;
1829     ev->poll_msecs = -1;
1830     INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1831 
1832     disk->ev = ev;
1833 }
1834 
1835 static void disk_add_events(struct gendisk *disk)
1836 {
1837     if (!disk->ev)
1838         return;
1839 
1840     /* FIXME: error handling */
1841     if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
1842         pr_warn("%s: failed to create sysfs files for events\n",
1843             disk->disk_name);
1844 
1845     mutex_lock(&disk_events_mutex);
1846     list_add_tail(&disk->ev->node, &disk_events);
1847     mutex_unlock(&disk_events_mutex);
1848 
1849     /*
1850      * Block count is initialized to 1 and the following initial
1851      * unblock kicks it into action.
1852      */
1853     __disk_unblock_events(disk, true);
1854 }
1855 
1856 static void disk_del_events(struct gendisk *disk)
1857 {
1858     if (!disk->ev)
1859         return;
1860 
1861     disk_block_events(disk);
1862 
1863     mutex_lock(&disk_events_mutex);
1864     list_del_init(&disk->ev->node);
1865     mutex_unlock(&disk_events_mutex);
1866 
1867     sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1868 }
1869 
1870 static void disk_release_events(struct gendisk *disk)
1871 {
1872     /* the block count should be 1 from disk_del_events() */
1873     WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
1874     kfree(disk->ev);
1875 }