lightnvm: introduce factory reset

Now that a device can be managed using the system blocks, a method to
reset the device is necessary as well. This patch introduces logic to
reset the device easily to factory state and exposes it through an
ioctl.

The ioctl takes the following flags:

  NVM_FACTORY_ERASE_ONLY_USER
      By default all blocks, except host-reserved blocks are erased upon
      factory reset. Instead of this, only erase host-reserved blocks.
  NVM_FACTORY_RESET_HOST_BLKS
      Mark host-reserved blocks to be erased and set their type to free.
  NVM_FACTORY_RESET_GRWN_BBLKS
      Mark "grown bad blocks" to be erased and set their type to free.

Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>

drivers/lightnvm/core.c

@@ -1088,6 +1088,38 @@ static long nvm_ioctl_dev_init(struct file *file, void __user *arg)
 	return __nvm_ioctl_dev_init(&init);
 }
 
+static long nvm_ioctl_dev_factory(struct file *file, void __user *arg)
+{
+	struct nvm_ioctl_dev_factory fact;
+	struct nvm_dev *dev;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	if (copy_from_user(&fact, arg, sizeof(struct nvm_ioctl_dev_factory)))
+		return -EFAULT;
+
+	fact.dev[DISK_NAME_LEN - 1] = '\0';
+
+	if (fact.flags & ~(NVM_FACTORY_NR_BITS - 1))
+		return -EINVAL;
+
+	down_write(&nvm_lock);
+	dev = nvm_find_nvm_dev(fact.dev);
+	up_write(&nvm_lock);
+	if (!dev) {
+		pr_err("nvm: device not found\n");
+		return -EINVAL;
+	}
+
+	if (dev->mt) {
+		dev->mt->unregister_mgr(dev);
+		dev->mt = NULL;
+	}
+
+	return nvm_dev_factory(dev, fact.flags);
+}
+
 static long nvm_ctl_ioctl(struct file *file, uint cmd, unsigned long arg)
 {
 	void __user *argp = (void __user *)arg;
@@ -1103,6 +1135,8 @@ static long nvm_ctl_ioctl(struct file *file, uint cmd, unsigned long arg)
 		return nvm_ioctl_dev_remove(file, argp);
 	case NVM_DEV_INIT:
 		return nvm_ioctl_dev_init(file, argp);
+	case NVM_DEV_FACTORY:
+		return nvm_ioctl_dev_factory(file, argp);
 	}
 	return 0;
 }

drivers/lightnvm/sysblk.c

@@ -560,3 +560,182 @@ err_mark:
 	mutex_unlock(&dev->mlock);
 	return ret;
 }
+
+struct factory_blks {
+	struct nvm_dev *dev;
+	int flags;
+	unsigned long *blks;
+};
+
+static int factory_nblks(int nblks)
+{
+	/* Round up to nearest BITS_PER_LONG */
+	return (nblks + (BITS_PER_LONG - 1)) & ~(BITS_PER_LONG - 1);
+}
+
+static unsigned int factory_blk_offset(struct nvm_dev *dev, int ch, int lun)
+{
+	int nblks = factory_nblks(dev->blks_per_lun);
+
+	return ((ch * dev->luns_per_chnl * nblks) + (lun * nblks)) /
+								BITS_PER_LONG;
+}
+
+static int nvm_factory_blks(struct ppa_addr ppa, int nr_blks, u8 *blks,
+								void *private)
+{
+	struct factory_blks *f = private;
+	struct nvm_dev *dev = f->dev;
+	int i, lunoff;
+
+	lunoff = factory_blk_offset(dev, ppa.g.ch, ppa.g.lun);
+
+	/* non-set bits correspond to the block must be erased */
+	for (i = 0; i < nr_blks; i++) {
+		switch (blks[i]) {
+		case NVM_BLK_T_FREE:
+			if (f->flags & NVM_FACTORY_ERASE_ONLY_USER)
+				set_bit(i, &f->blks[lunoff]);
+			break;
+		case NVM_BLK_T_HOST:
+			if (!(f->flags & NVM_FACTORY_RESET_HOST_BLKS))
+				set_bit(i, &f->blks[lunoff]);
+			break;
+		case NVM_BLK_T_GRWN_BAD:
+			if (!(f->flags & NVM_FACTORY_RESET_GRWN_BBLKS))
+				set_bit(i, &f->blks[lunoff]);
+			break;
+		default:
+			set_bit(i, &f->blks[lunoff]);
+			break;
+		}
+	}
+
+	return 0;
+}
+
+static int nvm_fact_get_blks(struct nvm_dev *dev, struct ppa_addr *erase_list,
+					int max_ppas, struct factory_blks *f)
+{
+	struct ppa_addr ppa;
+	int ch, lun, blkid, idx, done = 0, ppa_cnt = 0;
+	unsigned long *offset;
+
+	while (!done) {
+		done = 1;
+		for (ch = 0; ch < dev->nr_chnls; ch++) {
+			for (lun = 0; lun < dev->luns_per_chnl; lun++) {
+				idx = factory_blk_offset(dev, ch, lun);
+				offset = &f->blks[idx];
+
+				blkid = find_first_zero_bit(offset,
+							dev->blks_per_lun);
+				if (blkid >= dev->blks_per_lun)
+					continue;
+				set_bit(blkid, offset);
+
+				ppa.ppa = 0;
+				ppa.g.ch = ch;
+				ppa.g.lun = lun;
+				ppa.g.blk = blkid;
+				pr_debug("nvm: erase ppa (%u %u %u)\n",
+								ppa.g.ch,
+								ppa.g.lun,
+								ppa.g.blk);
+
+				erase_list[ppa_cnt] = ppa;
+				ppa_cnt++;
+				done = 0;
+
+				if (ppa_cnt == max_ppas)
+					return ppa_cnt;
+			}
+		}
+	}
+
+	return ppa_cnt;
+}
+
+static int nvm_fact_get_bb_tbl(struct nvm_dev *dev, struct ppa_addr ppa,
+					nvm_bb_update_fn *fn, void *priv)
+{
+	struct ppa_addr dev_ppa;
+	int ret;
+
+	dev_ppa = generic_to_dev_addr(dev, ppa);
+
+	ret = dev->ops->get_bb_tbl(dev, dev_ppa, dev->blks_per_lun, fn, priv);
+	if (ret)
+		pr_err("nvm: failed bb tbl for ch%u lun%u\n",
+							ppa.g.ch, ppa.g.blk);
+	return ret;
+}
+
+static int nvm_fact_select_blks(struct nvm_dev *dev, struct factory_blks *f)
+{
+	int ch, lun, ret;
+	struct ppa_addr ppa;
+
+	ppa.ppa = 0;
+	for (ch = 0; ch < dev->nr_chnls; ch++) {
+		for (lun = 0; lun < dev->luns_per_chnl; lun++) {
+			ppa.g.ch = ch;
+			ppa.g.lun = lun;
+
+			ret = nvm_fact_get_bb_tbl(dev, ppa, nvm_factory_blks,
+									f);
+			if (ret)
+				return ret;
+		}
+	}
+
+	return 0;
+}
+
+int nvm_dev_factory(struct nvm_dev *dev, int flags)
+{
+	struct factory_blks f;
+	struct ppa_addr *ppas;
+	int ppa_cnt, ret = -ENOMEM;
+	int max_ppas = dev->ops->max_phys_sect / dev->nr_planes;
+	struct ppa_addr sysblk_ppas[MAX_SYSBLKS];
+	struct sysblk_scan s;
+
+	f.blks = kzalloc(factory_nblks(dev->blks_per_lun) * dev->nr_luns,
+								GFP_KERNEL);
+	if (!f.blks)
+		return ret;
+
+	ppas = kcalloc(max_ppas, sizeof(struct ppa_addr), GFP_KERNEL);
+	if (!ppas)
+		goto err_blks;
+
+	f.dev = dev;
+	f.flags = flags;
+
+	/* create list of blks to be erased */
+	ret = nvm_fact_select_blks(dev, &f);
+	if (ret)
+		goto err_ppas;
+
+	/* continue to erase until list of blks until empty */
+	while ((ppa_cnt = nvm_fact_get_blks(dev, ppas, max_ppas, &f)) > 0)
+		nvm_erase_ppa(dev, ppas, ppa_cnt);
+
+	/* mark host reserved blocks free */
+	if (flags & NVM_FACTORY_RESET_HOST_BLKS) {
+		nvm_setup_sysblk_scan(dev, &s, sysblk_ppas);
+		mutex_lock(&dev->mlock);
+		ret = nvm_get_all_sysblks(dev, &s, sysblk_ppas,
+							sysblk_get_host_blks);
+		if (!ret)
+			ret = nvm_set_bb_tbl(dev, &s, NVM_BLK_T_FREE);
+		mutex_unlock(&dev->mlock);
+	}
+err_ppas:
+	kfree(ppas);
+err_blks:
+	kfree(f.blks);
+	return ret;
+}
+EXPORT_SYMBOL(nvm_dev_factory);

include/linux/lightnvm.h

@@ -527,6 +527,8 @@ struct nvm_system_block {
 extern int nvm_get_sysblock(struct nvm_dev *, struct nvm_sb_info *);
 extern int nvm_update_sysblock(struct nvm_dev *, struct nvm_sb_info *);
 extern int nvm_init_sysblock(struct nvm_dev *, struct nvm_sb_info *);
+
+extern int nvm_dev_factory(struct nvm_dev *, int flags);
 #else /* CONFIG_NVM */
 struct nvm_dev_ops;
 

include/uapi/linux/lightnvm.h

@@ -108,6 +108,20 @@ struct nvm_ioctl_dev_init {
 	__u32 flags;
 };
 
+enum {
+	NVM_FACTORY_ERASE_ONLY_USER	= 1 << 0, /* erase only blocks used as
+						   * host blks or grown blks */
+	NVM_FACTORY_RESET_HOST_BLKS	= 1 << 1, /* remove host blk marks */
+	NVM_FACTORY_RESET_GRWN_BBLKS	= 1 << 2, /* remove grown blk marks */
+	NVM_FACTORY_NR_BITS		= 1 << 3, /* stops here */
+};
+
+struct nvm_ioctl_dev_factory {
+	char dev[DISK_NAME_LEN];
+
+	__u32 flags;
+};
+
 /* The ioctl type, 'L', 0x20 - 0x2F documented in ioctl-number.txt */
 enum {
 	/* top level cmds */
@@ -120,6 +134,9 @@ enum {
 
 	/* Init a device to support LightNVM media managers */
 	NVM_DEV_INIT_CMD,
+
+	/* Factory reset device */
+	NVM_DEV_FACTORY_CMD,
 };
 
 #define NVM_IOCTL 'L' /* 0x4c */
@@ -134,6 +151,8 @@ enum {
 						struct nvm_ioctl_remove)
 #define NVM_DEV_INIT		_IOW(NVM_IOCTL, NVM_DEV_INIT_CMD, \
 						struct nvm_ioctl_dev_init)
+#define NVM_DEV_FACTORY		_IOW(NVM_IOCTL, NVM_DEV_FACTORY_CMD, \
+						struct nvm_ioctl_dev_factory)
 
 #define NVM_VERSION_MAJOR	1
 #define NVM_VERSION_MINOR	0