Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / fs / jffs2 / erase.c
diff --git a/kernel/fs/jffs2/erase.c b/kernel/fs/jffs2/erase.c
new file mode 100644 (file)
index 0000000..4a6cf28
--- /dev/null
@@ -0,0 +1,515 @@
+/*
+ * JFFS2 -- Journalling Flash File System, Version 2.
+ *
+ * Copyright © 2001-2007 Red Hat, Inc.
+ * Copyright © 2004-2010 David Woodhouse <dwmw2@infradead.org>
+ *
+ * Created by David Woodhouse <dwmw2@infradead.org>
+ *
+ * For licensing information, see the file 'LICENCE' in this directory.
+ *
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/kernel.h>
+#include <linux/slab.h>
+#include <linux/mtd/mtd.h>
+#include <linux/compiler.h>
+#include <linux/crc32.h>
+#include <linux/sched.h>
+#include <linux/pagemap.h>
+#include "nodelist.h"
+
+struct erase_priv_struct {
+       struct jffs2_eraseblock *jeb;
+       struct jffs2_sb_info *c;
+};
+
+#ifndef __ECOS
+static void jffs2_erase_callback(struct erase_info *);
+#endif
+static void jffs2_erase_failed(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t bad_offset);
+static void jffs2_erase_succeeded(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb);
+static void jffs2_mark_erased_block(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb);
+
+static void jffs2_erase_block(struct jffs2_sb_info *c,
+                             struct jffs2_eraseblock *jeb)
+{
+       int ret;
+       uint32_t bad_offset;
+#ifdef __ECOS
+       ret = jffs2_flash_erase(c, jeb);
+       if (!ret) {
+              jffs2_erase_succeeded(c, jeb);
+              return;
+       }
+       bad_offset = jeb->offset;
+#else /* Linux */
+       struct erase_info *instr;
+
+       jffs2_dbg(1, "%s(): erase block %#08x (range %#08x-%#08x)\n",
+                 __func__,
+                 jeb->offset, jeb->offset, jeb->offset + c->sector_size);
+       instr = kmalloc(sizeof(struct erase_info) + sizeof(struct erase_priv_struct), GFP_KERNEL);
+       if (!instr) {
+               pr_warn("kmalloc for struct erase_info in jffs2_erase_block failed. Refiling block for later\n");
+               mutex_lock(&c->erase_free_sem);
+               spin_lock(&c->erase_completion_lock);
+               list_move(&jeb->list, &c->erase_pending_list);
+               c->erasing_size -= c->sector_size;
+               c->dirty_size += c->sector_size;
+               jeb->dirty_size = c->sector_size;
+               spin_unlock(&c->erase_completion_lock);
+               mutex_unlock(&c->erase_free_sem);
+               return;
+       }
+
+       memset(instr, 0, sizeof(*instr));
+
+       instr->mtd = c->mtd;
+       instr->addr = jeb->offset;
+       instr->len = c->sector_size;
+       instr->callback = jffs2_erase_callback;
+       instr->priv = (unsigned long)(&instr[1]);
+
+       ((struct erase_priv_struct *)instr->priv)->jeb = jeb;
+       ((struct erase_priv_struct *)instr->priv)->c = c;
+
+       ret = mtd_erase(c->mtd, instr);
+       if (!ret)
+               return;
+
+       bad_offset = instr->fail_addr;
+       kfree(instr);
+#endif /* __ECOS */
+
+       if (ret == -ENOMEM || ret == -EAGAIN) {
+               /* Erase failed immediately. Refile it on the list */
+               jffs2_dbg(1, "Erase at 0x%08x failed: %d. Refiling on erase_pending_list\n",
+                         jeb->offset, ret);
+               mutex_lock(&c->erase_free_sem);
+               spin_lock(&c->erase_completion_lock);
+               list_move(&jeb->list, &c->erase_pending_list);
+               c->erasing_size -= c->sector_size;
+               c->dirty_size += c->sector_size;
+               jeb->dirty_size = c->sector_size;
+               spin_unlock(&c->erase_completion_lock);
+               mutex_unlock(&c->erase_free_sem);
+               return;
+       }
+
+       if (ret == -EROFS)
+               pr_warn("Erase at 0x%08x failed immediately: -EROFS. Is the sector locked?\n",
+                       jeb->offset);
+       else
+               pr_warn("Erase at 0x%08x failed immediately: errno %d\n",
+                       jeb->offset, ret);
+
+       jffs2_erase_failed(c, jeb, bad_offset);
+}
+
+int jffs2_erase_pending_blocks(struct jffs2_sb_info *c, int count)
+{
+       struct jffs2_eraseblock *jeb;
+       int work_done = 0;
+
+       mutex_lock(&c->erase_free_sem);
+
+       spin_lock(&c->erase_completion_lock);
+
+       while (!list_empty(&c->erase_complete_list) ||
+              !list_empty(&c->erase_pending_list)) {
+
+               if (!list_empty(&c->erase_complete_list)) {
+                       jeb = list_entry(c->erase_complete_list.next, struct jffs2_eraseblock, list);
+                       list_move(&jeb->list, &c->erase_checking_list);
+                       spin_unlock(&c->erase_completion_lock);
+                       mutex_unlock(&c->erase_free_sem);
+                       jffs2_mark_erased_block(c, jeb);
+
+                       work_done++;
+                       if (!--count) {
+                               jffs2_dbg(1, "Count reached. jffs2_erase_pending_blocks leaving\n");
+                               goto done;
+                       }
+
+               } else if (!list_empty(&c->erase_pending_list)) {
+                       jeb = list_entry(c->erase_pending_list.next, struct jffs2_eraseblock, list);
+                       jffs2_dbg(1, "Starting erase of pending block 0x%08x\n",
+                                 jeb->offset);
+                       list_del(&jeb->list);
+                       c->erasing_size += c->sector_size;
+                       c->wasted_size -= jeb->wasted_size;
+                       c->free_size -= jeb->free_size;
+                       c->used_size -= jeb->used_size;
+                       c->dirty_size -= jeb->dirty_size;
+                       jeb->wasted_size = jeb->used_size = jeb->dirty_size = jeb->free_size = 0;
+                       jffs2_free_jeb_node_refs(c, jeb);
+                       list_add(&jeb->list, &c->erasing_list);
+                       spin_unlock(&c->erase_completion_lock);
+                       mutex_unlock(&c->erase_free_sem);
+
+                       jffs2_erase_block(c, jeb);
+
+               } else {
+                       BUG();
+               }
+
+               /* Be nice */
+               cond_resched();
+               mutex_lock(&c->erase_free_sem);
+               spin_lock(&c->erase_completion_lock);
+       }
+
+       spin_unlock(&c->erase_completion_lock);
+       mutex_unlock(&c->erase_free_sem);
+ done:
+       jffs2_dbg(1, "jffs2_erase_pending_blocks completed\n");
+       return work_done;
+}
+
+static void jffs2_erase_succeeded(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
+{
+       jffs2_dbg(1, "Erase completed successfully at 0x%08x\n", jeb->offset);
+       mutex_lock(&c->erase_free_sem);
+       spin_lock(&c->erase_completion_lock);
+       list_move_tail(&jeb->list, &c->erase_complete_list);
+       /* Wake the GC thread to mark them clean */
+       jffs2_garbage_collect_trigger(c);
+       spin_unlock(&c->erase_completion_lock);
+       mutex_unlock(&c->erase_free_sem);
+       wake_up(&c->erase_wait);
+}
+
+static void jffs2_erase_failed(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t bad_offset)
+{
+       /* For NAND, if the failure did not occur at the device level for a
+          specific physical page, don't bother updating the bad block table. */
+       if (jffs2_cleanmarker_oob(c) && (bad_offset != (uint32_t)MTD_FAIL_ADDR_UNKNOWN)) {
+               /* We had a device-level failure to erase.  Let's see if we've
+                  failed too many times. */
+               if (!jffs2_write_nand_badblock(c, jeb, bad_offset)) {
+                       /* We'd like to give this block another try. */
+                       mutex_lock(&c->erase_free_sem);
+                       spin_lock(&c->erase_completion_lock);
+                       list_move(&jeb->list, &c->erase_pending_list);
+                       c->erasing_size -= c->sector_size;
+                       c->dirty_size += c->sector_size;
+                       jeb->dirty_size = c->sector_size;
+                       spin_unlock(&c->erase_completion_lock);
+                       mutex_unlock(&c->erase_free_sem);
+                       return;
+               }
+       }
+
+       mutex_lock(&c->erase_free_sem);
+       spin_lock(&c->erase_completion_lock);
+       c->erasing_size -= c->sector_size;
+       c->bad_size += c->sector_size;
+       list_move(&jeb->list, &c->bad_list);
+       c->nr_erasing_blocks--;
+       spin_unlock(&c->erase_completion_lock);
+       mutex_unlock(&c->erase_free_sem);
+       wake_up(&c->erase_wait);
+}
+
+#ifndef __ECOS
+static void jffs2_erase_callback(struct erase_info *instr)
+{
+       struct erase_priv_struct *priv = (void *)instr->priv;
+
+       if(instr->state != MTD_ERASE_DONE) {
+               pr_warn("Erase at 0x%08llx finished, but state != MTD_ERASE_DONE. State is 0x%x instead.\n",
+                       (unsigned long long)instr->addr, instr->state);
+               jffs2_erase_failed(priv->c, priv->jeb, instr->fail_addr);
+       } else {
+               jffs2_erase_succeeded(priv->c, priv->jeb);
+       }
+       kfree(instr);
+}
+#endif /* !__ECOS */
+
+/* Hmmm. Maybe we should accept the extra space it takes and make
+   this a standard doubly-linked list? */
+static inline void jffs2_remove_node_refs_from_ino_list(struct jffs2_sb_info *c,
+                       struct jffs2_raw_node_ref *ref, struct jffs2_eraseblock *jeb)
+{
+       struct jffs2_inode_cache *ic = NULL;
+       struct jffs2_raw_node_ref **prev;
+
+       prev = &ref->next_in_ino;
+
+       /* Walk the inode's list once, removing any nodes from this eraseblock */
+       while (1) {
+               if (!(*prev)->next_in_ino) {
+                       /* We're looking at the jffs2_inode_cache, which is
+                          at the end of the linked list. Stash it and continue
+                          from the beginning of the list */
+                       ic = (struct jffs2_inode_cache *)(*prev);
+                       prev = &ic->nodes;
+                       continue;
+               }
+
+               if (SECTOR_ADDR((*prev)->flash_offset) == jeb->offset) {
+                       /* It's in the block we're erasing */
+                       struct jffs2_raw_node_ref *this;
+
+                       this = *prev;
+                       *prev = this->next_in_ino;
+                       this->next_in_ino = NULL;
+
+                       if (this == ref)
+                               break;
+
+                       continue;
+               }
+               /* Not to be deleted. Skip */
+               prev = &((*prev)->next_in_ino);
+       }
+
+       /* PARANOIA */
+       if (!ic) {
+               JFFS2_WARNING("inode_cache/xattr_datum/xattr_ref"
+                             " not found in remove_node_refs()!!\n");
+               return;
+       }
+
+       jffs2_dbg(1, "Removed nodes in range 0x%08x-0x%08x from ino #%u\n",
+                 jeb->offset, jeb->offset + c->sector_size, ic->ino);
+
+       D2({
+               int i=0;
+               struct jffs2_raw_node_ref *this;
+               printk(KERN_DEBUG "After remove_node_refs_from_ino_list: \n");
+
+               this = ic->nodes;
+
+               printk(KERN_DEBUG);
+               while(this) {
+                       pr_cont("0x%08x(%d)->",
+                              ref_offset(this), ref_flags(this));
+                       if (++i == 5) {
+                               printk(KERN_DEBUG);
+                               i=0;
+                       }
+                       this = this->next_in_ino;
+               }
+               pr_cont("\n");
+       });
+
+       switch (ic->class) {
+#ifdef CONFIG_JFFS2_FS_XATTR
+               case RAWNODE_CLASS_XATTR_DATUM:
+                       jffs2_release_xattr_datum(c, (struct jffs2_xattr_datum *)ic);
+                       break;
+               case RAWNODE_CLASS_XATTR_REF:
+                       jffs2_release_xattr_ref(c, (struct jffs2_xattr_ref *)ic);
+                       break;
+#endif
+               default:
+                       if (ic->nodes == (void *)ic && ic->pino_nlink == 0)
+                               jffs2_del_ino_cache(c, ic);
+       }
+}
+
+void jffs2_free_jeb_node_refs(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
+{
+       struct jffs2_raw_node_ref *block, *ref;
+       jffs2_dbg(1, "Freeing all node refs for eraseblock offset 0x%08x\n",
+                 jeb->offset);
+
+       block = ref = jeb->first_node;
+
+       while (ref) {
+               if (ref->flash_offset == REF_LINK_NODE) {
+                       ref = ref->next_in_ino;
+                       jffs2_free_refblock(block);
+                       block = ref;
+                       continue;
+               }
+               if (ref->flash_offset != REF_EMPTY_NODE && ref->next_in_ino)
+                       jffs2_remove_node_refs_from_ino_list(c, ref, jeb);
+               /* else it was a non-inode node or already removed, so don't bother */
+
+               ref++;
+       }
+       jeb->first_node = jeb->last_node = NULL;
+}
+
+static int jffs2_block_check_erase(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb, uint32_t *bad_offset)
+{
+       void *ebuf;
+       uint32_t ofs;
+       size_t retlen;
+       int ret;
+       unsigned long *wordebuf;
+
+       ret = mtd_point(c->mtd, jeb->offset, c->sector_size, &retlen,
+                       &ebuf, NULL);
+       if (ret != -EOPNOTSUPP) {
+               if (ret) {
+                       jffs2_dbg(1, "MTD point failed %d\n", ret);
+                       goto do_flash_read;
+               }
+               if (retlen < c->sector_size) {
+                       /* Don't muck about if it won't let us point to the whole erase sector */
+                       jffs2_dbg(1, "MTD point returned len too short: 0x%zx\n",
+                                 retlen);
+                       mtd_unpoint(c->mtd, jeb->offset, retlen);
+                       goto do_flash_read;
+               }
+               wordebuf = ebuf-sizeof(*wordebuf);
+               retlen /= sizeof(*wordebuf);
+               do {
+                  if (*++wordebuf != ~0)
+                          break;
+               } while(--retlen);
+               mtd_unpoint(c->mtd, jeb->offset, c->sector_size);
+               if (retlen) {
+                       pr_warn("Newly-erased block contained word 0x%lx at offset 0x%08tx\n",
+                               *wordebuf,
+                               jeb->offset +
+                               c->sector_size-retlen * sizeof(*wordebuf));
+                       return -EIO;
+               }
+               return 0;
+       }
+ do_flash_read:
+       ebuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
+       if (!ebuf) {
+               pr_warn("Failed to allocate page buffer for verifying erase at 0x%08x. Refiling\n",
+                       jeb->offset);
+               return -EAGAIN;
+       }
+
+       jffs2_dbg(1, "Verifying erase at 0x%08x\n", jeb->offset);
+
+       for (ofs = jeb->offset; ofs < jeb->offset + c->sector_size; ) {
+               uint32_t readlen = min((uint32_t)PAGE_SIZE, jeb->offset + c->sector_size - ofs);
+               int i;
+
+               *bad_offset = ofs;
+
+               ret = mtd_read(c->mtd, ofs, readlen, &retlen, ebuf);
+               if (ret) {
+                       pr_warn("Read of newly-erased block at 0x%08x failed: %d. Putting on bad_list\n",
+                               ofs, ret);
+                       ret = -EIO;
+                       goto fail;
+               }
+               if (retlen != readlen) {
+                       pr_warn("Short read from newly-erased block at 0x%08x. Wanted %d, got %zd\n",
+                               ofs, readlen, retlen);
+                       ret = -EIO;
+                       goto fail;
+               }
+               for (i=0; i<readlen; i += sizeof(unsigned long)) {
+                       /* It's OK. We know it's properly aligned */
+                       unsigned long *datum = ebuf + i;
+                       if (*datum + 1) {
+                               *bad_offset += i;
+                               pr_warn("Newly-erased block contained word 0x%lx at offset 0x%08x\n",
+                                       *datum, *bad_offset);
+                               ret = -EIO;
+                               goto fail;
+                       }
+               }
+               ofs += readlen;
+               cond_resched();
+       }
+       ret = 0;
+fail:
+       kfree(ebuf);
+       return ret;
+}
+
+static void jffs2_mark_erased_block(struct jffs2_sb_info *c, struct jffs2_eraseblock *jeb)
+{
+       size_t retlen;
+       int ret;
+       uint32_t uninitialized_var(bad_offset);
+
+       switch (jffs2_block_check_erase(c, jeb, &bad_offset)) {
+       case -EAGAIN:   goto refile;
+       case -EIO:      goto filebad;
+       }
+
+       /* Write the erase complete marker */
+       jffs2_dbg(1, "Writing erased marker to block at 0x%08x\n", jeb->offset);
+       bad_offset = jeb->offset;
+
+       /* Cleanmarker in oob area or no cleanmarker at all ? */
+       if (jffs2_cleanmarker_oob(c) || c->cleanmarker_size == 0) {
+
+               if (jffs2_cleanmarker_oob(c)) {
+                       if (jffs2_write_nand_cleanmarker(c, jeb))
+                               goto filebad;
+               }
+       } else {
+
+               struct kvec vecs[1];
+               struct jffs2_unknown_node marker = {
+                       .magic =        cpu_to_je16(JFFS2_MAGIC_BITMASK),
+                       .nodetype =     cpu_to_je16(JFFS2_NODETYPE_CLEANMARKER),
+                       .totlen =       cpu_to_je32(c->cleanmarker_size)
+               };
+
+               jffs2_prealloc_raw_node_refs(c, jeb, 1);
+
+               marker.hdr_crc = cpu_to_je32(crc32(0, &marker, sizeof(struct jffs2_unknown_node)-4));
+
+               vecs[0].iov_base = (unsigned char *) &marker;
+               vecs[0].iov_len = sizeof(marker);
+               ret = jffs2_flash_direct_writev(c, vecs, 1, jeb->offset, &retlen);
+
+               if (ret || retlen != sizeof(marker)) {
+                       if (ret)
+                               pr_warn("Write clean marker to block at 0x%08x failed: %d\n",
+                                      jeb->offset, ret);
+                       else
+                               pr_warn("Short write to newly-erased block at 0x%08x: Wanted %zd, got %zd\n",
+                                      jeb->offset, sizeof(marker), retlen);
+
+                       goto filebad;
+               }
+       }
+       /* Everything else got zeroed before the erase */
+       jeb->free_size = c->sector_size;
+
+       mutex_lock(&c->erase_free_sem);
+       spin_lock(&c->erase_completion_lock);
+
+       c->erasing_size -= c->sector_size;
+       c->free_size += c->sector_size;
+
+       /* Account for cleanmarker now, if it's in-band */
+       if (c->cleanmarker_size && !jffs2_cleanmarker_oob(c))
+               jffs2_link_node_ref(c, jeb, jeb->offset | REF_NORMAL, c->cleanmarker_size, NULL);
+
+       list_move_tail(&jeb->list, &c->free_list);
+       c->nr_erasing_blocks--;
+       c->nr_free_blocks++;
+
+       jffs2_dbg_acct_sanity_check_nolock(c, jeb);
+       jffs2_dbg_acct_paranoia_check_nolock(c, jeb);
+
+       spin_unlock(&c->erase_completion_lock);
+       mutex_unlock(&c->erase_free_sem);
+       wake_up(&c->erase_wait);
+       return;
+
+filebad:
+       jffs2_erase_failed(c, jeb, bad_offset);
+       return;
+
+refile:
+       /* Stick it back on the list from whence it came and come back later */
+       mutex_lock(&c->erase_free_sem);
+       spin_lock(&c->erase_completion_lock);
+       jffs2_garbage_collect_trigger(c);
+       list_move(&jeb->list, &c->erase_complete_list);
+       spin_unlock(&c->erase_completion_lock);
+       mutex_unlock(&c->erase_free_sem);
+       return;
+}