}
/**
- * ubifs_calc_min_idx_lebs - calculate amount of eraseblocks for the index.
+ * ubifs_calc_min_idx_lebs - calculate amount of LEBs for the index.
* @c: UBIFS file-system description object
*
- * This function calculates and returns the number of eraseblocks which should
- * be kept for index usage.
+ * This function calculates and returns the number of LEBs which should be kept
+ * for index usage.
*/
int ubifs_calc_min_idx_lebs(struct ubifs_info *c)
{
- int idx_lebs, eff_leb_size = c->leb_size - c->max_idx_node_sz;
+ int idx_lebs;
long long idx_size;
idx_size = c->old_idx_sz + c->budg_idx_growth + c->budg_uncommitted_idx;
-
/* And make sure we have thrice the index size of space reserved */
- idx_size = idx_size + (idx_size << 1);
-
+ idx_size += idx_size << 1;
/*
* We do not maintain 'old_idx_size' as 'old_idx_lebs'/'old_idx_bytes'
* pair, nor similarly the two variables for the new index size, so we
* have to do this costly 64-bit division on fast-path.
*/
- idx_size += eff_leb_size - 1;
- idx_lebs = div_u64(idx_size, eff_leb_size);
+ idx_lebs = div_u64(idx_size + c->idx_leb_size - 1, c->idx_leb_size);
/*
* The index head is not available for the in-the-gaps method, so add an
* extra LEB to compensate.
* do_budget_space - reserve flash space for index and data growth.
* @c: UBIFS file-system description object
*
- * This function makes sure UBIFS has enough free eraseblocks for index growth
- * and data.
+ * This function makes sure UBIFS has enough free LEBs for index growth and
+ * data.
*
* When budgeting index space, UBIFS reserves thrice as many LEBs as the index
* would take if it was consolidated and written to the flash. This guarantees
* that the "in-the-gaps" commit method always succeeds and UBIFS will always
* be able to commit dirty index. So this function basically adds amount of
* budgeted index space to the size of the current index, multiplies this by 3,
- * and makes sure this does not exceed the amount of free eraseblocks.
+ * and makes sure this does not exceed the amount of free LEBs.
*
* Notes about @c->min_idx_lebs and @c->lst.idx_lebs variables:
* o @c->lst.idx_lebs is the number of LEBs the index currently uses. It might
* be large, because UBIFS does not do any index consolidation as long as
* there is free space. IOW, the index may take a lot of LEBs, but the LEBs
* will contain a lot of dirt.
- * o @c->min_idx_lebs is the the index presumably takes. IOW, the index may be
- * consolidated to take up to @c->min_idx_lebs LEBs.
+ * o @c->min_idx_lebs is the number of LEBS the index presumably takes. IOW,
+ * the index may be consolidated to take up to @c->min_idx_lebs LEBs.
*
* This function returns zero in case of success, and %-ENOSPC in case of
* failure.
* This function calculates amount of free space to report to user-space.
*
* Because UBIFS may introduce substantial overhead (the index, node headers,
- * alignment, wastage at the end of eraseblocks, etc), it cannot report real
- * amount of free flash space it has (well, because not all dirty space is
- * reclaimable, UBIFS does not actually know the real amount). If UBIFS did so,
- * it would bread user expectations about what free space is. Users seem to
- * accustomed to assume that if the file-system reports N bytes of free space,
- * they would be able to fit a file of N bytes to the FS. This almost works for
+ * alignment, wastage at the end of LEBs, etc), it cannot report real amount of
+ * free flash space it has (well, because not all dirty space is reclaimable,
+ * UBIFS does not actually know the real amount). If UBIFS did so, it would
+ * bread user expectations about what free space is. Users seem to accustomed
+ * to assume that if the file-system reports N bytes of free space, they would
+ * be able to fit a file of N bytes to the FS. This almost works for
* traditional file-systems, because they have way less overhead than UBIFS.
* So, to keep users happy, UBIFS tries to take the overhead into account.
*/
"bad or corrupted node)");
else {
for (i = 0; i < nlen && dent->name[i]; i++)
- printk("%c", dent->name[i]);
+ printk(KERN_CONT "%c", dent->name[i]);
}
- printk("\n");
+ printk(KERN_CONT "\n");
break;
}
/*
* Make sure the last key in our znode is less or
- * equivalent than the the key in zbranch which goes
+ * equivalent than the key in the zbranch which goes
* after our pointing zbranch.
*/
cmp = keys_cmp(c, max,
struct ubifs_inode *ui = ubifs_inode(inode);
pgoff_t index = pos >> PAGE_CACHE_SHIFT;
int uninitialized_var(err), appending = !!(pos + len > inode->i_size);
+ int skipped_read = 0;
struct page *page;
ubifs_assert(ubifs_inode(inode)->ui_size == inode->i_size);
if (!PageUptodate(page)) {
/* The page is not loaded from the flash */
- if (!(pos & ~PAGE_CACHE_MASK) && len == PAGE_CACHE_SIZE)
+ if (!(pos & ~PAGE_CACHE_MASK) && len == PAGE_CACHE_SIZE) {
/*
* We change whole page so no need to load it. But we
* have to set the @PG_checked flag to make the further
* the media.
*/
SetPageChecked(page);
- else {
+ skipped_read = 1;
+ } else {
err = do_readpage(page);
if (err) {
unlock_page(page);
err = allocate_budget(c, page, ui, appending);
if (unlikely(err)) {
ubifs_assert(err == -ENOSPC);
+ /*
+ * If we skipped reading the page because we were going to
+ * write all of it, then it is not up to date.
+ */
+ if (skipped_read) {
+ ClearPageChecked(page);
+ ClearPageUptodate(page);
+ }
/*
* Budgeting failed which means it would have to force
* write-back but didn't, because we set the @fast flag in the
* whole index and correct all inode sizes, which is long an unacceptable.
*
* To prevent situations like this, UBIFS writes pages back only if they are
- * within last synchronized inode size, i.e. the the size which has been
+ * within the last synchronized inode size, i.e. the size which has been
* written to the flash media last time. Otherwise, UBIFS forces inode
* write-back, thus making sure the on-flash inode contains current inode size,
* and then keeps writing pages back.
* ubifs_find_free_space - find a data LEB with free space.
* @c: the UBIFS file-system description object
* @min_space: minimum amount of required free space
- * @free: contains amount of free space in the LEB on exit
+ * @offs: contains offset of where free space starts on exit
* @squeeze: whether to try to find space in a non-empty LEB first
*
* This function looks for an LEB with at least @min_space bytes of free space.
* failed to find a LEB with @min_space bytes of free space and other a negative
* error codes in case of failure.
*/
-int ubifs_find_free_space(struct ubifs_info *c, int min_space, int *free,
+int ubifs_find_free_space(struct ubifs_info *c, int min_space, int *offs,
int squeeze)
{
const struct ubifs_lprops *lprops;
spin_unlock(&c->space_lock);
}
- *free = lprops->free;
+ *offs = c->leb_size - lprops->free;
ubifs_release_lprops(c);
- if (*free == c->leb_size) {
+ if (*offs == 0) {
/*
* Ensure that empty LEBs have been unmapped. They may not have
* been, for example, because of an unclean unmount. Also
return err;
}
- dbg_find("found LEB %d, free %d", lnum, *free);
- ubifs_assert(*free >= min_space);
+ dbg_find("found LEB %d, free %d", lnum, c->leb_size - *offs);
+ ubifs_assert(*offs <= c->leb_size - min_space);
return lnum;
out:
* have to waste large pieces of free space at the end of LEB B, because nodes
* from LEB A would not fit. And the worst situation is when all nodes are of
* maximum size. So dark watermark is the amount of free + dirty space in LEB
- * which are guaranteed to be reclaimable. If LEB has less space, the GC migh
+ * which are guaranteed to be reclaimable. If LEB has less space, the GC might
* be unable to reclaim it. So, LEBs with free + dirty greater than dark
* watermark are "good" LEBs from GC's point of few. The other LEBs are not so
* good, and GC takes extra care when moving them.
#include <linux/pagemap.h>
#include "ubifs.h"
-/*
- * GC tries to optimize the way it fit nodes to available space, and it sorts
- * nodes a little. The below constants are watermarks which define "large",
- * "medium", and "small" nodes.
- */
-#define MEDIUM_NODE_WM (UBIFS_BLOCK_SIZE / 4)
-#define SMALL_NODE_WM UBIFS_MAX_DENT_NODE_SZ
-
/*
* GC may need to move more than one LEB to make progress. The below constants
* define "soft" and "hard" limits on the number of LEBs the garbage collector
}
/**
- * joinup - bring data nodes for an inode together.
- * @c: UBIFS file-system description object
- * @sleb: describes scanned LEB
- * @inum: inode number
- * @blk: block number
- * @data: list to which to add data nodes
+ * list_sort - sort a list.
+ * @priv: private data, passed to @cmp
+ * @head: the list to sort
+ * @cmp: the elements comparison function
*
- * This function looks at the first few nodes in the scanned LEB @sleb and adds
- * them to @data if they are data nodes from @inum and have a larger block
- * number than @blk. This function returns %0 on success and a negative error
- * code on failure.
+ * This function has been implemented by Mark J Roberts <mjr@znex.org>. It
+ * implements "merge sort" which has O(nlog(n)) complexity. The list is sorted
+ * in ascending order.
+ *
+ * The comparison function @cmp is supposed to return a negative value if @a is
+ * than @b, and a positive value if @a is greater than @b. If @a and @b are
+ * equivalent, then it does not matter what this function returns.
*/
-static int joinup(struct ubifs_info *c, struct ubifs_scan_leb *sleb, ino_t inum,
- unsigned int blk, struct list_head *data)
+static void list_sort(void *priv, struct list_head *head,
+ int (*cmp)(void *priv, struct list_head *a,
+ struct list_head *b))
{
- int err, cnt = 6, lnum = sleb->lnum, offs;
- struct ubifs_scan_node *snod, *tmp;
- union ubifs_key *key;
+ struct list_head *p, *q, *e, *list, *tail, *oldhead;
+ int insize, nmerges, psize, qsize, i;
+
+ if (list_empty(head))
+ return;
+
+ list = head->next;
+ list_del(head);
+ insize = 1;
+ for (;;) {
+ p = oldhead = list;
+ list = tail = NULL;
+ nmerges = 0;
+
+ while (p) {
+ nmerges++;
+ q = p;
+ psize = 0;
+ for (i = 0; i < insize; i++) {
+ psize++;
+ q = q->next == oldhead ? NULL : q->next;
+ if (!q)
+ break;
+ }
- list_for_each_entry_safe(snod, tmp, &sleb->nodes, list) {
- key = &snod->key;
- if (key_inum(c, key) == inum &&
- key_type(c, key) == UBIFS_DATA_KEY &&
- key_block(c, key) > blk) {
- offs = snod->offs;
- err = ubifs_tnc_has_node(c, key, 0, lnum, offs, 0);
- if (err < 0)
- return err;
- list_del(&snod->list);
- if (err) {
- list_add_tail(&snod->list, data);
- blk = key_block(c, key);
- } else
- kfree(snod);
- cnt = 6;
- } else if (--cnt == 0)
+ qsize = insize;
+ while (psize > 0 || (qsize > 0 && q)) {
+ if (!psize) {
+ e = q;
+ q = q->next;
+ qsize--;
+ if (q == oldhead)
+ q = NULL;
+ } else if (!qsize || !q) {
+ e = p;
+ p = p->next;
+ psize--;
+ if (p == oldhead)
+ p = NULL;
+ } else if (cmp(priv, p, q) <= 0) {
+ e = p;
+ p = p->next;
+ psize--;
+ if (p == oldhead)
+ p = NULL;
+ } else {
+ e = q;
+ q = q->next;
+ qsize--;
+ if (q == oldhead)
+ q = NULL;
+ }
+ if (tail)
+ tail->next = e;
+ else
+ list = e;
+ e->prev = tail;
+ tail = e;
+ }
+ p = q;
+ }
+
+ tail->next = list;
+ list->prev = tail;
+
+ if (nmerges <= 1)
break;
+
+ insize *= 2;
}
- return 0;
+
+ head->next = list;
+ head->prev = list->prev;
+ list->prev->next = head;
+ list->prev = head;
}
/**
- * move_nodes - move nodes.
+ * data_nodes_cmp - compare 2 data nodes.
+ * @priv: UBIFS file-system description object
+ * @a: first data node
+ * @a: second data node
+ *
+ * This function compares data nodes @a and @b. Returns %1 if @a has greater
+ * inode or block number, and %-1 otherwise.
+ */
+int data_nodes_cmp(void *priv, struct list_head *a, struct list_head *b)
+{
+ ino_t inuma, inumb;
+ struct ubifs_info *c = priv;
+ struct ubifs_scan_node *sa, *sb;
+
+ cond_resched();
+ sa = list_entry(a, struct ubifs_scan_node, list);
+ sb = list_entry(b, struct ubifs_scan_node, list);
+ ubifs_assert(key_type(c, &sa->key) == UBIFS_DATA_KEY);
+ ubifs_assert(key_type(c, &sb->key) == UBIFS_DATA_KEY);
+
+ inuma = key_inum(c, &sa->key);
+ inumb = key_inum(c, &sb->key);
+
+ if (inuma == inumb) {
+ unsigned int blka = key_block(c, &sa->key);
+ unsigned int blkb = key_block(c, &sb->key);
+
+ if (blka <= blkb)
+ return -1;
+ } else if (inuma <= inumb)
+ return -1;
+
+ return 1;
+}
+
+/*
+ * nondata_nodes_cmp - compare 2 non-data nodes.
+ * @priv: UBIFS file-system description object
+ * @a: first node
+ * @a: second node
+ *
+ * This function compares nodes @a and @b. It makes sure that inode nodes go
+ * first and sorted by length in descending order. Directory entry nodes go
+ * after inode nodes and are sorted in ascending hash valuer order.
+ */
+int nondata_nodes_cmp(void *priv, struct list_head *a, struct list_head *b)
+{
+ int typea, typeb;
+ ino_t inuma, inumb;
+ struct ubifs_info *c = priv;
+ struct ubifs_scan_node *sa, *sb;
+
+ cond_resched();
+ sa = list_entry(a, struct ubifs_scan_node, list);
+ sb = list_entry(b, struct ubifs_scan_node, list);
+ typea = key_type(c, &sa->key);
+ typeb = key_type(c, &sb->key);
+ ubifs_assert(typea != UBIFS_DATA_KEY && typeb != UBIFS_DATA_KEY);
+
+ /* Inodes go before directory entries */
+ if (typea == UBIFS_INO_KEY) {
+ if (typeb == UBIFS_INO_KEY)
+ return sb->len - sa->len;
+ return -1;
+ }
+ if (typeb == UBIFS_INO_KEY)
+ return 1;
+
+ ubifs_assert(typea == UBIFS_DENT_KEY && typeb == UBIFS_DENT_KEY);
+ inuma = key_inum(c, &sa->key);
+ inumb = key_inum(c, &sb->key);
+
+ if (inuma == inumb) {
+ uint32_t hasha = key_hash(c, &sa->key);
+ uint32_t hashb = key_hash(c, &sb->key);
+
+ if (hasha <= hashb)
+ return -1;
+ } else if (inuma <= inumb)
+ return -1;
+
+ return 1;
+}
+
+/**
+ * sort_nodes - sort nodes for GC.
* @c: UBIFS file-system description object
- * @sleb: describes nodes to move
+ * @sleb: describes nodes to sort and contains the result on exit
+ * @nondata: contains non-data nodes on exit
+ * @min: minimum node size is returned here
*
- * This function moves valid nodes from data LEB described by @sleb to the GC
- * journal head. The obsolete nodes are dropped.
+ * This function sorts the list of inodes to garbage collect. First of all, it
+ * kills obsolete nodes and separates data and non-data nodes to the
+ * @sleb->nodes and @nondata lists correspondingly.
+ *
+ * Data nodes are then sorted in block number order - this is important for
+ * bulk-read; data nodes with lower inode number go before data nodes with
+ * higher inode number, and data nodes with lower block number go before data
+ * nodes with higher block number;
*
- * When moving nodes we have to deal with classical bin-packing problem: the
- * space in the current GC journal head LEB and in @c->gc_lnum are the "bins",
- * where the nodes in the @sleb->nodes list are the elements which should be
- * fit optimally to the bins. This function uses the "first fit decreasing"
- * strategy, although it does not really sort the nodes but just split them on
- * 3 classes - large, medium, and small, so they are roughly sorted.
+ * Non-data nodes are sorted as follows.
+ * o First go inode nodes - they are sorted in descending length order.
+ * o Then go directory entry nodes - they are sorted in hash order, which
+ * should supposedly optimize 'readdir()'. Direntry nodes with lower parent
+ * inode number go before direntry nodes with higher parent inode number,
+ * and direntry nodes with lower name hash values go before direntry nodes
+ * with higher name hash values.
*
- * This function returns zero in case of success, %-EAGAIN if commit is
- * required, and other negative error codes in case of other failures.
+ * This function returns zero in case of success and a negative error code in
+ * case of failure.
*/
-static int move_nodes(struct ubifs_info *c, struct ubifs_scan_leb *sleb)
+static int sort_nodes(struct ubifs_info *c, struct ubifs_scan_leb *sleb,
+ struct list_head *nondata, int *min)
{
struct ubifs_scan_node *snod, *tmp;
- struct list_head data, large, medium, small;
- struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf;
- int avail, err, min = INT_MAX;
- unsigned int blk = 0;
- ino_t inum = 0;
- INIT_LIST_HEAD(&data);
- INIT_LIST_HEAD(&large);
- INIT_LIST_HEAD(&medium);
- INIT_LIST_HEAD(&small);
+ *min = INT_MAX;
- while (!list_empty(&sleb->nodes)) {
- struct list_head *lst = sleb->nodes.next;
-
- snod = list_entry(lst, struct ubifs_scan_node, list);
+ /* Separate data nodes and non-data nodes */
+ list_for_each_entry_safe(snod, tmp, &sleb->nodes, list) {
+ int err;
ubifs_assert(snod->type != UBIFS_IDX_NODE);
ubifs_assert(snod->type != UBIFS_REF_NODE);
err = ubifs_tnc_has_node(c, &snod->key, 0, sleb->lnum,
snod->offs, 0);
if (err < 0)
- goto out;
+ return err;
- list_del(lst);
if (!err) {
/* The node is obsolete, remove it from the list */
+ list_del(&snod->list);
kfree(snod);
continue;
}
- /*
- * Sort the list of nodes so that data nodes go first, large
- * nodes go second, and small nodes go last.
- */
- if (key_type(c, &snod->key) == UBIFS_DATA_KEY) {
- if (inum != key_inum(c, &snod->key)) {
- if (inum) {
- /*
- * Try to move data nodes from the same
- * inode together.
- */
- err = joinup(c, sleb, inum, blk, &data);
- if (err)
- goto out;
- }
- inum = key_inum(c, &snod->key);
- blk = key_block(c, &snod->key);
- }
- list_add_tail(lst, &data);
- } else if (snod->len > MEDIUM_NODE_WM)
- list_add_tail(lst, &large);
- else if (snod->len > SMALL_NODE_WM)
- list_add_tail(lst, &medium);
- else
- list_add_tail(lst, &small);
-
- /* And find the smallest node */
- if (snod->len < min)
- min = snod->len;
+ if (snod->len < *min)
+ *min = snod->len;
+
+ if (key_type(c, &snod->key) != UBIFS_DATA_KEY)
+ list_move_tail(&snod->list, nondata);
}
- /*
- * Join the tree lists so that we'd have one roughly sorted list
- * ('large' will be the head of the joined list).
- */
- list_splice(&data, &large);
- list_splice(&medium, large.prev);
- list_splice(&small, large.prev);
+ /* Sort data and non-data nodes */
+ list_sort(c, &sleb->nodes, &data_nodes_cmp);
+ list_sort(c, nondata, &nondata_nodes_cmp);
+ return 0;
+}
+
+/**
+ * move_node - move a node.
+ * @c: UBIFS file-system description object
+ * @sleb: describes the LEB to move nodes from
+ * @snod: the mode to move
+ * @wbuf: write-buffer to move node to
+ *
+ * This function moves node @snod to @wbuf, changes TNC correspondingly, and
+ * destroys @snod. Returns zero in case of success and a negative error code in
+ * case of failure.
+ */
+static int move_node(struct ubifs_info *c, struct ubifs_scan_leb *sleb,
+ struct ubifs_scan_node *snod, struct ubifs_wbuf *wbuf)
+{
+ int err, new_lnum = wbuf->lnum, new_offs = wbuf->offs + wbuf->used;
+
+ cond_resched();
+ err = ubifs_wbuf_write_nolock(wbuf, snod->node, snod->len);
+ if (err)
+ return err;
+
+ err = ubifs_tnc_replace(c, &snod->key, sleb->lnum,
+ snod->offs, new_lnum, new_offs,
+ snod->len);
+ list_del(&snod->list);
+ kfree(snod);
+ return err;
+}
+
+/**
+ * move_nodes - move nodes.
+ * @c: UBIFS file-system description object
+ * @sleb: describes the LEB to move nodes from
+ *
+ * This function moves valid nodes from data LEB described by @sleb to the GC
+ * journal head. This function returns zero in case of success, %-EAGAIN if
+ * commit is required, and other negative error codes in case of other
+ * failures.
+ */
+static int move_nodes(struct ubifs_info *c, struct ubifs_scan_leb *sleb)
+{
+ int err, min;
+ LIST_HEAD(nondata);
+ struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf;
if (wbuf->lnum == -1) {
/*
*/
err = switch_gc_head(c);
if (err)
- goto out;
+ return err;
}
+ err = sort_nodes(c, sleb, &nondata, &min);
+ if (err)
+ goto out;
+
/* Write nodes to their new location. Use the first-fit strategy */
while (1) {
- avail = c->leb_size - wbuf->offs - wbuf->used;
- list_for_each_entry_safe(snod, tmp, &large, list) {
- int new_lnum, new_offs;
+ int avail;
+ struct ubifs_scan_node *snod, *tmp;
+
+ /* Move data nodes */
+ list_for_each_entry_safe(snod, tmp, &sleb->nodes, list) {
+ avail = c->leb_size - wbuf->offs - wbuf->used;
+ if (snod->len > avail)
+ /*
+ * Do not skip data nodes in order to optimize
+ * bulk-read.
+ */
+ break;
+
+ err = move_node(c, sleb, snod, wbuf);
+ if (err)
+ goto out;
+ }
+ /* Move non-data nodes */
+ list_for_each_entry_safe(snod, tmp, &nondata, list) {
+ avail = c->leb_size - wbuf->offs - wbuf->used;
if (avail < min)
break;
- if (snod->len > avail)
- /* This node does not fit */
+ if (snod->len > avail) {
+ /*
+ * Keep going only if this is an inode with
+ * some data. Otherwise stop and switch the GC
+ * head. IOW, we assume that data-less inode
+ * nodes and direntry nodes are roughly of the
+ * same size.
+ */
+ if (key_type(c, &snod->key) == UBIFS_DENT_KEY ||
+ snod->len == UBIFS_INO_NODE_SZ)
+ break;
continue;
+ }
- cond_resched();
-
- new_lnum = wbuf->lnum;
- new_offs = wbuf->offs + wbuf->used;
- err = ubifs_wbuf_write_nolock(wbuf, snod->node,
- snod->len);
+ err = move_node(c, sleb, snod, wbuf);
if (err)
goto out;
- err = ubifs_tnc_replace(c, &snod->key, sleb->lnum,
- snod->offs, new_lnum, new_offs,
- snod->len);
- if (err)
- goto out;
-
- avail = c->leb_size - wbuf->offs - wbuf->used;
- list_del(&snod->list);
- kfree(snod);
}
- if (list_empty(&large))
+ if (list_empty(&sleb->nodes) && list_empty(&nondata))
break;
/*
return 0;
out:
- list_for_each_entry_safe(snod, tmp, &large, list) {
- list_del(&snod->list);
- kfree(snod);
- }
+ list_splice_tail(&nondata, &sleb->nodes);
return err;
}
*/
static int reserve_space(struct ubifs_info *c, int jhead, int len)
{
- int err = 0, err1, retries = 0, avail, lnum, offs, free, squeeze;
+ int err = 0, err1, retries = 0, avail, lnum, offs, squeeze;
struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
/*
* Write buffer wasn't seek'ed or there is no enough space - look for an
* LEB with some empty space.
*/
- lnum = ubifs_find_free_space(c, len, &free, squeeze);
+ lnum = ubifs_find_free_space(c, len, &offs, squeeze);
if (lnum >= 0) {
/* Found an LEB, add it to the journal head */
- offs = c->leb_size - free;
err = ubifs_add_bud_to_log(c, jhead, lnum, offs);
if (err)
goto out_return;
* @host: host inode
*
* This function writes the updated version of an extended attribute inode and
- * the host inode tho the journal (to the base head). The host inode is written
+ * the host inode to the journal (to the base head). The host inode is written
* after the extended attribute inode in order to guarantee that the extended
* attribute will be flushed when the inode is synchronized by 'fsync()' and
* consequently, the write-buffer is synchronized. This function returns zero
* @c: UBIFS file-system description object
* @key: the key to get hash from
*/
-static inline int key_hash(const struct ubifs_info *c,
- const union ubifs_key *key)
+static inline uint32_t key_hash(const struct ubifs_info *c,
+ const union ubifs_key *key)
{
return key->u32[1] & UBIFS_S_KEY_HASH_MASK;
}
* @c: UBIFS file-system description object
* @k: the key to get hash from
*/
-static inline int key_hash_flash(const struct ubifs_info *c, const void *k)
+static inline uint32_t key_hash_flash(const struct ubifs_info *c, const void *k)
{
const union ubifs_key *key = k;
}
/*
- * Make sure the the amount of space in buds will not exceed
+ * Make sure the amount of space in buds will not exceed the
* 'c->max_bud_bytes' limit, because we want to guarantee mount time
* limits.
*
bud->jhead, c->leb_size - bud->start,
c->cmt_bud_bytes);
rb_erase(p1, &c->buds);
- list_del(&bud->list);
/*
* If the commit does not finish, the recovery will need
* to replay the journal, in which case the old buds
* commit i.e. do not allow them to be garbage
* collected.
*/
- list_add(&bud->list, &c->old_buds);
+ list_move(&bud->list, &c->old_buds);
}
}
spin_unlock(&c->buds_lock);
while (offs + len > c->leb_size) {
alen = ALIGN(offs, c->min_io_size);
upd_ltab(c, lnum, c->leb_size - alen, alen - offs);
- dbg_chk_lpt_sz(c, 2, alen - offs);
+ dbg_chk_lpt_sz(c, 2, c->leb_size - offs);
err = alloc_lpt_leb(c, &lnum);
if (err)
goto no_space;
if (offs + c->lsave_sz > c->leb_size) {
alen = ALIGN(offs, c->min_io_size);
upd_ltab(c, lnum, c->leb_size - alen, alen - offs);
- dbg_chk_lpt_sz(c, 2, alen - offs);
+ dbg_chk_lpt_sz(c, 2, c->leb_size - offs);
err = alloc_lpt_leb(c, &lnum);
if (err)
goto no_space;
if (offs + c->ltab_sz > c->leb_size) {
alen = ALIGN(offs, c->min_io_size);
upd_ltab(c, lnum, c->leb_size - alen, alen - offs);
- dbg_chk_lpt_sz(c, 2, alen - offs);
+ dbg_chk_lpt_sz(c, 2, c->leb_size - offs);
err = alloc_lpt_leb(c, &lnum);
if (err)
goto no_space;
alen, UBI_SHORTTERM);
if (err)
return err;
- dbg_chk_lpt_sz(c, 4, alen - wlen);
}
- dbg_chk_lpt_sz(c, 2, 0);
+ dbg_chk_lpt_sz(c, 2, c->leb_size - offs);
err = realloc_lpt_leb(c, &lnum);
if (err)
goto no_space;
- offs = 0;
- from = 0;
+ offs = from = 0;
ubifs_assert(lnum >= c->lpt_first &&
lnum <= c->lpt_last);
err = ubifs_leb_unmap(c, lnum);
UBI_SHORTTERM);
if (err)
return err;
- dbg_chk_lpt_sz(c, 2, alen - wlen);
+ dbg_chk_lpt_sz(c, 2, c->leb_size - offs);
err = realloc_lpt_leb(c, &lnum);
if (err)
goto no_space;
- offs = 0;
+ offs = from = 0;
ubifs_assert(lnum >= c->lpt_first &&
lnum <= c->lpt_last);
err = ubifs_leb_unmap(c, lnum);
UBI_SHORTTERM);
if (err)
return err;
- dbg_chk_lpt_sz(c, 2, alen - wlen);
+ dbg_chk_lpt_sz(c, 2, c->leb_size - offs);
err = realloc_lpt_leb(c, &lnum);
if (err)
goto no_space;
- offs = 0;
+ offs = from = 0;
ubifs_assert(lnum >= c->lpt_first &&
lnum <= c->lpt_last);
err = ubifs_leb_unmap(c, lnum);
/**
* dbg_chk_lpt_sz - check LPT does not write more than LPT size.
* @c: the UBIFS file-system description object
- * @action: action
+ * @action: what to do
* @len: length written
*
* This function returns %0 on success and a negative error code on failure.
+ * The @action argument may be one of:
+ * o %0 - LPT debugging checking starts, initialize debugging variables;
+ * o %1 - wrote an LPT node, increase LPT size by @len bytes;
+ * o %2 - switched to a different LEB and wasted @len bytes;
+ * o %3 - check that we've written the right number of bytes.
+ * o %4 - wasted @len bytes;
*/
int dbg_chk_lpt_sz(struct ubifs_info *c, int action, int len)
{
lnum, offs);
err = ubifs_unpack_nnode(c, buf, &nnode);
for (i = 0; i < UBIFS_LPT_FANOUT; i++) {
- printk("%d:%d", nnode.nbranch[i].lnum,
+ printk(KERN_CONT "%d:%d", nnode.nbranch[i].lnum,
nnode.nbranch[i].offs);
if (i != UBIFS_LPT_FANOUT - 1)
- printk(", ");
+ printk(KERN_CONT ", ");
}
- printk("\n");
+ printk(KERN_CONT "\n");
break;
}
case UBIFS_LPT_LTAB:
* @lnum: LEB number of the LEB from which @buf was read
* @offs: offset from which @buf was read
*
- * This function scans @buf for more nodes and returns %0 is a node is found and
- * %1 if no more nodes are found.
+ * This function ensures that the corrupted node at @offs is the last thing
+ * written to a LEB. This function returns %1 if more data is not found and
+ * %0 if more data is found.
*/
static int no_more_nodes(const struct ubifs_info *c, void *buf, int len,
int lnum, int offs)
{
- int skip, next_offs = 0;
+ struct ubifs_ch *ch = buf;
+ int skip, dlen = le32_to_cpu(ch->len);
- if (len > UBIFS_DATA_NODE_SZ) {
- struct ubifs_ch *ch = buf;
- int dlen = le32_to_cpu(ch->len);
-
- if (ch->node_type == UBIFS_DATA_NODE && dlen >= UBIFS_CH_SZ &&
- dlen <= UBIFS_MAX_DATA_NODE_SZ)
- /* The corrupt node looks like a data node */
- next_offs = ALIGN(offs + dlen, 8);
- }
-
- if (c->min_io_size == 1)
- skip = 8;
- else
- skip = ALIGN(offs + 1, c->min_io_size) - offs;
-
- offs += skip;
- buf += skip;
- len -= skip;
- while (len > 8) {
- struct ubifs_ch *ch = buf;
- uint32_t magic = le32_to_cpu(ch->magic);
- int ret;
-
- if (magic == UBIFS_NODE_MAGIC) {
- ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1);
- if (ret == SCANNED_A_NODE || ret > 0) {
- /*
- * There is a small chance this is just data in
- * a data node, so check that possibility. e.g.
- * this is part of a file that itself contains
- * a UBIFS image.
- */
- if (next_offs && offs + le32_to_cpu(ch->len) <=
- next_offs)
- continue;
- dbg_rcvry("unexpected node at %d:%d", lnum,
- offs);
- return 0;
- }
- }
- offs += 8;
- buf += 8;
- len -= 8;
+ /* Check for empty space after the corrupt node's common header */
+ skip = ALIGN(offs + UBIFS_CH_SZ, c->min_io_size) - offs;
+ if (is_empty(buf + skip, len - skip))
+ return 1;
+ /*
+ * The area after the common header size is not empty, so the common
+ * header must be intact. Check it.
+ */
+ if (ubifs_check_node(c, buf, lnum, offs, 1, 0) != -EUCLEAN) {
+ dbg_rcvry("unexpected bad common header at %d:%d", lnum, offs);
+ return 0;
}
- return 1;
+ /* Now we know the corrupt node's length we can skip over it */
+ skip = ALIGN(offs + dlen, c->min_io_size) - offs;
+ /* After which there should be empty space */
+ if (is_empty(buf + skip, len - skip))
+ return 1;
+ dbg_rcvry("unexpected data at %d:%d", lnum, offs + skip);
+ return 0;
}
/**
dirty -= c->leb_size - lp->free;
/*
* If the replay order was perfect the dirty space would now be
- * zero. The order is not perfect because the the journal heads
+ * zero. The order is not perfect because the journal heads
* race with each other. This is not a problem but is does mean
* that the dirty space may temporarily exceed c->leb_size
* during the replay.
if (tmp64 > DEFAULT_MAX_RP_SIZE)
tmp64 = DEFAULT_MAX_RP_SIZE;
sup->rp_size = cpu_to_le64(tmp64);
+ sup->ro_compat_version = cpu_to_le32(UBIFS_RO_COMPAT_VERSION);
err = ubifs_write_node(c, sup, UBIFS_SB_NODE_SZ, 0, 0, UBI_LONGTERM);
kfree(sup);
if (IS_ERR(sup))
return PTR_ERR(sup);
+ c->fmt_version = le32_to_cpu(sup->fmt_version);
+ c->ro_compat_version = le32_to_cpu(sup->ro_compat_version);
+
/*
* The software supports all previous versions but not future versions,
* due to the unavailability of time-travelling equipment.
*/
- c->fmt_version = le32_to_cpu(sup->fmt_version);
if (c->fmt_version > UBIFS_FORMAT_VERSION) {
- ubifs_err("on-flash format version is %d, but software only "
- "supports up to version %d", c->fmt_version,
- UBIFS_FORMAT_VERSION);
- err = -EINVAL;
- goto out;
+ struct super_block *sb = c->vfs_sb;
+ int mounting_ro = sb->s_flags & MS_RDONLY;
+
+ ubifs_assert(!c->ro_media || mounting_ro);
+ if (!mounting_ro ||
+ c->ro_compat_version > UBIFS_RO_COMPAT_VERSION) {
+ ubifs_err("on-flash format version is w%d/r%d, but "
+ "software only supports up to version "
+ "w%d/r%d", c->fmt_version,
+ c->ro_compat_version, UBIFS_FORMAT_VERSION,
+ UBIFS_RO_COMPAT_VERSION);
+ if (c->ro_compat_version <= UBIFS_RO_COMPAT_VERSION) {
+ ubifs_msg("only R/O mounting is possible");
+ err = -EROFS;
+ } else
+ err = -EINVAL;
+ goto out;
+ }
+
+ /*
+ * The FS is mounted R/O, and the media format is
+ * R/O-compatible with the UBIFS implementation, so we can
+ * mount.
+ */
+ c->rw_incompat = 1;
}
if (c->fmt_version < 3) {
c->main_lebs = c->leb_cnt - UBIFS_SB_LEBS - UBIFS_MST_LEBS;
c->main_lebs -= c->log_lebs + c->lpt_lebs + c->orph_lebs;
c->main_first = c->leb_cnt - c->main_lebs;
- c->report_rp_size = ubifs_reported_space(c, c->rp_size);
err = validate_sb(c, sup);
out:
* Move this one to the end of the list to provide some
* fairness.
*/
- list_del(&c->infos_list);
- list_add_tail(&c->infos_list, &ubifs_infos);
+ list_move_tail(&c->infos_list, &ubifs_infos);
mutex_unlock(&c->umount_mutex);
if (freed >= nr)
break;
}
if (i == 1) {
- list_del(&c->infos_list);
- list_add_tail(&c->infos_list, &ubifs_infos);
+ list_move_tail(&c->infos_list, &ubifs_infos);
spin_unlock(&ubifs_infos_lock);
ubifs_request_bg_commit(c);
seq_printf(s, ",no_chk_data_crc");
if (c->mount_opts.override_compr) {
- seq_printf(s, ",compr=");
- seq_printf(s, ubifs_compr_name(c->mount_opts.compr_type));
+ seq_printf(s, ",compr=%s",
+ ubifs_compr_name(c->mount_opts.compr_type));
}
return 0;
if (err)
return err;
+ /* Initialize effective LEB size used in budgeting calculations */
+ c->idx_leb_size = c->leb_size - c->max_idx_node_sz;
return 0;
}
long long tmp64;
c->min_idx_lebs = ubifs_calc_min_idx_lebs(c);
+ c->report_rp_size = ubifs_reported_space(c, c->rp_size);
/*
* Calculate total amount of FS blocks. This number is not used
goto out_cbuf;
/* Create background thread */
- c->bgt = kthread_create(ubifs_bg_thread, c, c->bgt_name);
+ c->bgt = kthread_create(ubifs_bg_thread, c, "%s", c->bgt_name);
if (IS_ERR(c->bgt)) {
err = PTR_ERR(c->bgt);
c->bgt = NULL;
else {
c->need_recovery = 0;
ubifs_msg("recovery completed");
- /* GC LEB has to be empty and taken at this point */
- ubifs_assert(c->lst.taken_empty_lebs == 1);
+ /*
+ * GC LEB has to be empty and taken at this point. But
+ * the journal head LEBs may also be accounted as
+ * "empty taken" if they are empty.
+ */
+ ubifs_assert(c->lst.taken_empty_lebs > 0);
}
} else
- ubifs_assert(c->lst.taken_empty_lebs == 1);
+ ubifs_assert(c->lst.taken_empty_lebs > 0);
err = dbg_check_filesystem(c);
if (err)
x = (long long)c->log_lebs * c->leb_size + c->max_bud_bytes;
ubifs_msg("journal size: %lld bytes (%lld KiB, %lld MiB, %d "
"LEBs)", x, x >> 10, x >> 20, c->log_lebs + c->max_bud_cnt);
- ubifs_msg("media format: %d (latest is %d)",
- c->fmt_version, UBIFS_FORMAT_VERSION);
+ ubifs_msg("media format: w%d/r%d (latest is w%d/r%d)",
+ c->fmt_version, c->ro_compat_version,
+ UBIFS_FORMAT_VERSION, UBIFS_RO_COMPAT_VERSION);
ubifs_msg("default compressor: %s", ubifs_compr_name(c->default_compr));
ubifs_msg("reserved for root: %llu bytes (%llu KiB)",
c->report_rp_size, c->report_rp_size >> 10);
{
int err, lnum;
+ if (c->rw_incompat) {
+ ubifs_err("the file-system is not R/W-compatible");
+ ubifs_msg("on-flash format version is w%d/r%d, but software "
+ "only supports up to version w%d/r%d", c->fmt_version,
+ c->ro_compat_version, UBIFS_FORMAT_VERSION,
+ UBIFS_RO_COMPAT_VERSION);
+ return -EROFS;
+ }
+
mutex_lock(&c->umount_mutex);
dbg_save_space_info(c);
c->remounting_rw = 1;
ubifs_create_buds_lists(c);
/* Create background thread */
- c->bgt = kthread_create(ubifs_bg_thread, c, c->bgt_name);
+ c->bgt = kthread_create(ubifs_bg_thread, c, "%s", c->bgt_name);
if (IS_ERR(c->bgt)) {
err = PTR_ERR(c->bgt);
c->bgt = NULL;
c->bu.buf = NULL;
}
- ubifs_assert(c->lst.taken_empty_lebs == 1);
+ ubifs_assert(c->lst.taken_empty_lebs > 0);
return 0;
}
* splitting in the middle of the colliding sequence. Also, when
* removing the leftmost key, we would have to correct the key of the
* parent node, which would introduce additional complications. Namely,
- * if we changed the the leftmost key of the parent znode, the garbage
+ * if we changed the leftmost key of the parent znode, the garbage
* collector would be unable to find it (GC is doing this when GC'ing
* indexing LEBs). Although we already have an additional RB-tree where
* we save such changed znodes (see 'ins_clr_old_idx_znode()') until
/* UBIFS node magic number (must not have the padding byte first or last) */
#define UBIFS_NODE_MAGIC 0x06101831
-/* UBIFS on-flash format version */
+/*
+ * UBIFS on-flash format version. This version is increased when the on-flash
+ * format is changing. If this happens, UBIFS is will support older versions as
+ * well. But older UBIFS code will not support newer formats. Format changes
+ * will be rare and only when absolutely necessary, e.g. to fix a bug or to add
+ * a new feature.
+ *
+ * UBIFS went into mainline kernel with format version 4. The older formats
+ * were development formats.
+ */
#define UBIFS_FORMAT_VERSION 4
+/*
+ * Read-only compatibility version. If the UBIFS format is changed, older UBIFS
+ * implementations will not be able to mount newer formats in read-write mode.
+ * However, depending on the change, it may be possible to mount newer formats
+ * in R/O mode. This is indicated by the R/O compatibility version which is
+ * stored in the super-block.
+ *
+ * This is needed to support boot-loaders which only need R/O mounting. With
+ * this flag it is possible to do UBIFS format changes without a need to update
+ * boot-loaders.
+ */
+#define UBIFS_RO_COMPAT_VERSION 0
+
/* Minimum logical eraseblock size in bytes */
#define UBIFS_MIN_LEB_SZ (15*1024)
/*
* If compressed data length is less than %UBIFS_MIN_COMPRESS_DIFF bytes
- * shorter than uncompressed data length, UBIFS preferes to leave this data
+ * shorter than uncompressed data length, UBIFS prefers to leave this data
* node uncompress, because it'll be read faster.
*/
#define UBIFS_MIN_COMPRESS_DIFF 64
* @padding2: reserved for future, zeroes
* @time_gran: time granularity in nanoseconds
* @uuid: UUID generated when the file system image was created
+ * @ro_compat_version: UBIFS R/O compatibility version
*/
struct ubifs_sb_node {
struct ubifs_ch ch;
__le64 rp_size;
__le32 time_gran;
__u8 uuid[16];
- __u8 padding2[3972];
+ __le32 ro_compat_version;
+ __u8 padding2[3968];
} __attribute__ ((packed));
/**
* by @commit_sem
* @cnt_lock: protects @highest_inum and @max_sqnum counters
* @fmt_version: UBIFS on-flash format version
+ * @ro_compat_version: R/O compatibility version
* @uuid: UUID from super block
*
* @lhead_lnum: log head logical eraseblock number
* recovery)
* @bulk_read: enable bulk-reads
* @default_compr: default compression algorithm (%UBIFS_COMPR_LZO, etc)
+ * @rw_incompat: the media is not R/W compatible
*
* @tnc_mutex: protects the Tree Node Cache (TNC), @zroot, @cnext, @enext, and
* @calc_idx_sz
* @min_io_shift: number of bits in @min_io_size minus one
* @leb_size: logical eraseblock size in bytes
* @half_leb_size: half LEB size
+ * @idx_leb_size: how many bytes of an LEB are effectively available when it is
+ * used to store indexing nodes (@leb_size - @max_idx_node_sz)
* @leb_cnt: count of logical eraseblocks
* @max_leb_cnt: maximum count of logical eraseblocks
* @old_leb_cnt: count of logical eraseblocks before re-size
* previous commit start
* @uncat_list: list of un-categorized LEBs
* @empty_list: list of empty LEBs
- * @freeable_list: list of freeable non-index LEBs (free + dirty == leb_size)
- * @frdi_idx_list: list of freeable index LEBs (free + dirty == leb_size)
+ * @freeable_list: list of freeable non-index LEBs (free + dirty == @leb_size)
+ * @frdi_idx_list: list of freeable index LEBs (free + dirty == @leb_size)
* @freeable_cnt: number of freeable LEBs in @freeable_list
*
* @ltab_lnum: LEB number of LPT's own lprops table
unsigned long long cmt_no;
spinlock_t cnt_lock;
int fmt_version;
+ int ro_compat_version;
unsigned char uuid[16];
int lhead_lnum;
unsigned int no_chk_data_crc:1;
unsigned int bulk_read:1;
unsigned int default_compr:2;
+ unsigned int rw_incompat:1;
struct mutex tnc_mutex;
struct ubifs_zbranch zroot;
int min_io_shift;
int leb_size;
int half_leb_size;
+ int idx_leb_size;
int leb_cnt;
int max_leb_cnt;
int old_leb_cnt;
long long ubifs_calc_available(const struct ubifs_info *c, int min_idx_lebs);
/* find.c */
-int ubifs_find_free_space(struct ubifs_info *c, int min_space, int *free,
+int ubifs_find_free_space(struct ubifs_info *c, int min_space, int *offs,
int squeeze);
int ubifs_find_free_leb_for_idx(struct ubifs_info *c);
int ubifs_find_dirty_leb(struct ubifs_info *c, struct ubifs_lprops *ret_lp,
projects / linux-2.6-omap-h63xx.git / commitdiff