/**
* reiser4_attach_fsdata - try to attach fsdata
* @file:
* @inode:
*
* Finds or creates cursor for readdir-over-nfs.
*/
int reiser4_attach_fsdata(struct file *file, struct inode *inode)
{
loff_t pos;
int result;
dir_cursor *cursor;
/*
* we are serialized by inode->i_mutex
*/
if (!file_is_stateless(file))
return 0;
pos = file->f_pos;
result = 0;
if (pos == 0) {
/*
* first call to readdir (or rewind to the beginning of
* directory)
*/
cursor = kmem_cache_alloc(d_cursor_cache,
reiser4_ctx_gfp_mask_get());
if (cursor != NULL)
result = insert_cursor(cursor, file, inode);
else
result = RETERR(-ENOMEM);
} else {
/* try to find existing cursor */
struct d_cursor_key key;
key.cid = pos >> CID_SHIFT;
key.oid = get_inode_oid(inode);
spin_lock(&d_lock);
cursor = d_cursor_hash_find(&d_info(inode)->table, &key);
if (cursor != NULL) {
/* cursor was found */
if (cursor->ref == 0) {
/* move it from unused list */
list_del_init(&cursor->alist);
--d_cursor_unused;
}
++cursor->ref;
}
spin_unlock(&d_lock);
if (cursor != NULL) {
spin_lock_inode(inode);
assert("nikita-3556", cursor->fsdata->back == NULL);
clean_fsdata(file);
free_file_fsdata_nolock(file);
file->private_data = cursor->fsdata;
spin_unlock_inode(inode);
}
}
return result;
}
/* this is common implementation of key_by_inode method of file plugin
*/
int
key_by_inode_and_offset_common(struct inode *inode, loff_t off,
reiser4_key * key)
{
reiser4_key_init(key);
set_key_locality(key, reiser4_inode_data(inode)->locality_id);
set_key_ordering(key, get_inode_ordering(inode));
set_key_objectid(key, get_inode_oid(inode)); /*FIXME: inode->i_ino */
set_key_type(key, KEY_BODY_MINOR);
set_key_offset(key, (__u64) off);
return 0;
}
/**
* insert_cursor - allocate file_fsdata, insert cursor to tree and hash table
* @cursor:
* @file:
* @inode:
*
* Allocates reiser4_file_fsdata, attaches it to @cursor, inserts cursor to
* reiser4 super block's hash table and radix tree.
add detachable readdir
* state to the @f
*/
static int insert_cursor(dir_cursor *cursor, struct file *file,
struct inode *inode)
{
int result;
reiser4_file_fsdata *fsdata;
memset(cursor, 0, sizeof *cursor);
/* this is either first call to readdir, or rewind. Anyway, create new
* cursor. */
fsdata = create_fsdata(NULL);
if (fsdata != NULL) {
result = radix_tree_preload(reiser4_ctx_gfp_mask_get());
if (result == 0) {
struct d_cursor_info *info;
oid_t oid;
info = d_info(inode);
oid = get_inode_oid(inode);
/* cid occupies higher 12 bits of f->f_pos. Don't
* allow it to become negative: this confuses
* nfsd_readdir() */
cursor->key.cid = (++cid_counter) & 0x7ff;
cursor->key.oid = oid;
cursor->fsdata = fsdata;
cursor->info = info;
cursor->ref = 1;
spin_lock_inode(inode);
/* install cursor as @f's private_data, discarding old
* one if necessary */
#if REISER4_DEBUG
if (file->private_data)
warning("", "file has fsdata already");
#endif
clean_fsdata(file);
free_file_fsdata_nolock(file);
file->private_data = fsdata;
fsdata->cursor = cursor;
spin_unlock_inode(inode);
spin_lock(&d_lock);
/* insert cursor into hash table */
d_cursor_hash_insert(&info->table, cursor);
/* and chain it into radix-tree */
bind_cursor(cursor, (unsigned long)oid);
spin_unlock(&d_lock);
radix_tree_preload_end();
file->f_pos = ((__u64) cursor->key.cid) << CID_SHIFT;
}
} else
result = RETERR(-ENOMEM);
return result;
}
/* clear stat data's flag indicating that conversion is being converted */
static int complete_conversion(struct inode *inode)
{
int result;
grab_space_enable();
result =
reiser4_grab_space(inode_file_plugin(inode)->estimate.update(inode),
BA_CAN_COMMIT);
if (result == 0) {
reiser4_inode_clr_flag(inode, REISER4_PART_MIXED);
result = reiser4_update_sd(inode);
}
if (result)
warning("vs-1696", "Failed to clear converting bit of %llu: %i",
(unsigned long long)get_inode_oid(inode), result);
return 0;
}
/* this is common implementation of write_sd_by_inode method of file plugin
either insert stat data or update it
*/
int write_sd_by_inode_common(struct inode *inode/* object to save */)
{
int result;
assert("nikita-730", inode != NULL);
if (reiser4_inode_get_flag(inode, REISER4_NO_SD))
/* object doesn't have stat-data yet */
result = insert_new_sd(inode);
else
result = update_sd(inode);
if (result != 0 && result != -ENAMETOOLONG && result != -ENOMEM)
/* Don't issue warnings about "name is too long" */
warning("nikita-2221", "Failed to save sd for %llu: %i",
(unsigned long long)get_inode_oid(inode), result);
return result;
}
/* for every page of file: read page, cut part of extent pointing to this page,
put data of page tree by tail item */
int extent2tail(struct file * file, struct unix_file_info *uf_info)
{
int result;
struct inode *inode;
struct page *page;
unsigned long num_pages, i;
unsigned long start_page;
reiser4_key from;
reiser4_key to;
unsigned count;
__u64 offset;
assert("nikita-3362", ea_obtained(uf_info));
inode = unix_file_info_to_inode(uf_info);
assert("nikita-3412", !IS_RDONLY(inode));
assert("vs-1649", uf_info->container != UF_CONTAINER_TAILS);
assert("", !reiser4_inode_get_flag(inode, REISER4_PART_IN_CONV));
offset = 0;
if (reiser4_inode_get_flag(inode, REISER4_PART_MIXED)) {
/*
* file is marked on disk as there was a conversion which did
* not complete due to either crash or some error. Find which
* offset tail conversion stopped at
*/
result = find_start(inode, EXTENT_POINTER_ID, &offset);
if (result == -ENOENT) {
/* no extent found, everything is converted */
uf_info->container = UF_CONTAINER_TAILS;
complete_conversion(inode);
return 0;
} else if (result != 0)
/* some other error */
return result;
}
reiser4_inode_set_flag(inode, REISER4_PART_IN_CONV);
/* number of pages in the file */
num_pages =
(inode->i_size + - offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
start_page = offset >> PAGE_CACHE_SHIFT;
inode_file_plugin(inode)->key_by_inode(inode, offset, &from);
to = from;
result = 0;
for (i = 0; i < num_pages; i++) {
__u64 start_byte;
result = reserve_extent2tail_iteration(inode);
if (result != 0)
break;
if (i == 0 && offset == 0) {
reiser4_inode_set_flag(inode, REISER4_PART_MIXED);
reiser4_update_sd(inode);
}
page = read_mapping_page(inode->i_mapping,
(unsigned)(i + start_page), NULL);
if (IS_ERR(page)) {
result = PTR_ERR(page);
break;
}
wait_on_page_locked(page);
if (!PageUptodate(page)) {
page_cache_release(page);
result = RETERR(-EIO);
break;
}
/* cut part of file we have read */
start_byte = (__u64) ((i + start_page) << PAGE_CACHE_SHIFT);
set_key_offset(&from, start_byte);
set_key_offset(&to, start_byte + PAGE_CACHE_SIZE - 1);
/*
* reiser4_cut_tree_object() returns -E_REPEAT to allow atom
* commits during over-long truncates. But
* extent->tail conversion should be performed in one
* transaction.
*/
result = reiser4_cut_tree(reiser4_tree_by_inode(inode), &from,
&to, inode, 0);
if (result) {
page_cache_release(page);
break;
}
/* put page data into tree via tail_write */
count = PAGE_CACHE_SIZE;
if ((i == (num_pages - 1)) &&
(inode->i_size & ~PAGE_CACHE_MASK))
/* last page can be incompleted */
count = (inode->i_size & ~PAGE_CACHE_MASK);
while (count) {
loff_t pos = start_byte;
assert("edward-1537",
file != NULL && file->f_dentry != NULL);
assert("edward-1538",
file->f_dentry->d_inode == inode);
result = reiser4_write_tail(file, inode,
(char __user *)kmap(page),
count, &pos);
reiser4_free_file_fsdata(file);
if (result <= 0) {
warning("", "reiser4_write_tail failed");
page_cache_release(page);
reiser4_inode_clr_flag(inode, REISER4_PART_IN_CONV);
return result;
}
count -= result;
}
/* release page */
lock_page(page);
/* page is already detached from jnode and mapping. */
assert("vs-1086", page->mapping == NULL);
assert("nikita-2690",
(!PagePrivate(page) && jprivate(page) == 0));
/* waiting for writeback completion with page lock held is
* perfectly valid. */
wait_on_page_writeback(page);
reiser4_drop_page(page);
/* release reference taken by read_cache_page() above */
page_cache_release(page);
drop_exclusive_access(uf_info);
/*
* throttle the conversion.
* FIXME-EDWARD: Calculate and pass the precise number
* of pages that was dirtied
*/
reiser4_throttle_write(inode, 1);
get_exclusive_access(uf_info);
/*
* nobody is allowed to complete conversion but a process which
* started it
*/
assert("", reiser4_inode_get_flag(inode, REISER4_PART_MIXED));
}
reiser4_inode_clr_flag(inode, REISER4_PART_IN_CONV);
if (i == num_pages) {
/* file is converted to formatted items */
assert("vs-1698", reiser4_inode_get_flag(inode,
REISER4_PART_MIXED));
assert("vs-1260",
inode_has_no_jnodes(reiser4_inode_data(inode)));
uf_info->container = UF_CONTAINER_TAILS;
complete_conversion(inode);
return 0;
}
/*
* conversion is not complete. Inode was already marked as
* REISER4_PART_MIXED and stat-data were updated at the first
* iteration of the loop above.
*/
warning("nikita-2282",
"Partial conversion of %llu: %lu of %lu: %i",
(unsigned long long)get_inode_oid(inode), i,
num_pages, result);
/* this flag should be cleared, otherwise get_exclusive_access_careful()
will fall into infinite loop */
assert("edward-1550", !reiser4_inode_get_flag(inode,
REISER4_PART_IN_CONV));
return result;
}
/**
* tail2extent
* @uf_info:
*
*
*/
int tail2extent(struct unix_file_info *uf_info)
{
int result;
reiser4_key key; /* key of next byte to be moved to page */
char *p_data; /* data of page */
unsigned page_off = 0, /* offset within the page where to copy data */
count; /* number of bytes of item which can be
* copied to page */
struct page *pages[TAIL2EXTENT_PAGE_NUM];
struct page *page;
int done; /* set to 1 when all file is read */
char *item;
int i;
struct inode *inode;
int first_iteration;
int bytes;
__u64 offset;
assert("nikita-3362", ea_obtained(uf_info));
inode = unix_file_info_to_inode(uf_info);
assert("nikita-3412", !IS_RDONLY(inode));
assert("vs-1649", uf_info->container != UF_CONTAINER_EXTENTS);
assert("", !reiser4_inode_get_flag(inode, REISER4_PART_IN_CONV));
offset = 0;
first_iteration = 1;
result = 0;
if (reiser4_inode_get_flag(inode, REISER4_PART_MIXED)) {
/*
* file is marked on disk as there was a conversion which did
* not complete due to either crash or some error. Find which
* offset tail conversion stopped at
*/
result = find_start(inode, FORMATTING_ID, &offset);
if (result == -ENOENT) {
/* no tail items found, everything is converted */
uf_info->container = UF_CONTAINER_EXTENTS;
complete_conversion(inode);
return 0;
} else if (result != 0)
/* some other error */
return result;
first_iteration = 0;
}
reiser4_inode_set_flag(inode, REISER4_PART_IN_CONV);
/* get key of first byte of a file */
inode_file_plugin(inode)->key_by_inode(inode, offset, &key);
done = 0;
while (done == 0) {
memset(pages, 0, sizeof(pages));
result = reserve_tail2extent_iteration(inode);
if (result != 0) {
reiser4_inode_clr_flag(inode, REISER4_PART_IN_CONV);
goto out;
}
if (first_iteration) {
reiser4_inode_set_flag(inode, REISER4_PART_MIXED);
reiser4_update_sd(inode);
first_iteration = 0;
}
bytes = 0;
for (i = 0; i < sizeof_array(pages) && done == 0; i++) {
assert("vs-598",
(get_key_offset(&key) & ~PAGE_CACHE_MASK) == 0);
page = alloc_page(reiser4_ctx_gfp_mask_get());
if (!page) {
result = RETERR(-ENOMEM);
goto error;
}
page->index =
(unsigned long)(get_key_offset(&key) >>
PAGE_CACHE_SHIFT);
/*
* usually when one is going to longterm lock znode (as
* find_file_item does, for instance) he must not hold
* locked pages. However, there is an exception for
* case tail2extent. Pages appearing here are not
* reachable to everyone else, they are clean, they do
* not have jnodes attached so keeping them locked do
* not risk deadlock appearance
*/
assert("vs-983", !PagePrivate(page));
reiser4_invalidate_pages(inode->i_mapping, page->index,
1, 0);
for (page_off = 0; page_off < PAGE_CACHE_SIZE;) {
coord_t coord;
lock_handle lh;
/* get next item */
/* FIXME: we might want to readahead here */
init_lh(&lh);
result =
find_file_item_nohint(&coord, &lh, &key,
ZNODE_READ_LOCK,
inode);
if (result != CBK_COORD_FOUND) {
/*
* error happened of not items of file
* were found
*/
done_lh(&lh);
page_cache_release(page);
goto error;
}
if (coord.between == AFTER_UNIT) {
/*
* end of file is reached. Padd page
* with zeros
*/
done_lh(&lh);
done = 1;
p_data = kmap_atomic(page, KM_USER0);
memset(p_data + page_off, 0,
PAGE_CACHE_SIZE - page_off);
kunmap_atomic(p_data, KM_USER0);
break;
}
result = zload(coord.node);
if (result) {
page_cache_release(page);
done_lh(&lh);
goto error;
}
assert("vs-856", coord.between == AT_UNIT);
item = ((char *)item_body_by_coord(&coord)) +
coord.unit_pos;
/* how many bytes to copy */
count =
item_length_by_coord(&coord) -
coord.unit_pos;
/* limit length of copy to end of page */
if (count > PAGE_CACHE_SIZE - page_off)
count = PAGE_CACHE_SIZE - page_off;
/*
* copy item (as much as will fit starting from
* the beginning of the item) into the page
*/
p_data = kmap_atomic(page, KM_USER0);
memcpy(p_data + page_off, item, count);
kunmap_atomic(p_data, KM_USER0);
page_off += count;
bytes += count;
set_key_offset(&key,
get_key_offset(&key) + count);
zrelse(coord.node);
done_lh(&lh);
} /* end of loop which fills one page by content of
* formatting items */
if (page_off) {
/* something was copied into page */
pages[i] = page;
} else {
page_cache_release(page);
assert("vs-1648", done == 1);
break;
}
} /* end of loop through pages of one conversion iteration */
if (i > 0) {
result = replace(inode, pages, i, bytes);
release_all_pages(pages, sizeof_array(pages));
if (result)
goto error;
/*
* We have to drop exclusive access to avoid deadlock
* which may happen because called by reiser4_writepages
* capture_unix_file requires to get non-exclusive
* access to a file. It is safe to drop EA in the middle
* of tail2extent conversion because write_unix_file,
* setattr_unix_file(truncate), mmap_unix_file,
* release_unix_file(extent2tail) checks if conversion
* is not in progress (see comments before
* get_exclusive_access_careful().
* Other processes that acquire non-exclusive access
* (read_unix_file, reiser4_writepages, etc) should work
* on partially converted files.
*/
drop_exclusive_access(uf_info);
/* throttle the conversion
FIXME-EDWARD: Pass the precise number of pages
that was dirtied */
reiser4_throttle_write(inode, 1);
get_exclusive_access(uf_info);
/*
* nobody is allowed to complete conversion but a
* process which started it
*/
assert("", reiser4_inode_get_flag(inode,
REISER4_PART_MIXED));
}
}
if (result == 0) {
/* file is converted to extent items */
reiser4_inode_clr_flag(inode, REISER4_PART_IN_CONV);
assert("vs-1697", reiser4_inode_get_flag(inode,
REISER4_PART_MIXED));
uf_info->container = UF_CONTAINER_EXTENTS;
complete_conversion(inode);
} else {
/*
* conversion is not complete. Inode was already marked as
* REISER4_PART_MIXED and stat-data were updated at the first
* iteration of the loop above.
*/
error:
release_all_pages(pages, sizeof_array(pages));
reiser4_inode_clr_flag(inode, REISER4_PART_IN_CONV);
warning("edward-1548", "Partial conversion of %llu: %i",
(unsigned long long)get_inode_oid(inode), result);
}
out:
/* this flag should be cleared, otherwise get_exclusive_access_careful()
will fall into infinite loop */
assert("edward-1549", !reiser4_inode_get_flag(inode,
REISER4_PART_IN_CONV));
return result;
}
/**
* process_cursors - do action on each cursor attached to inode
* @inode:
* @act: action to do
*
* Finds all cursors of @inode in reiser4's super block radix tree of cursors
* and performs action specified by @act on each of cursors.
*/
static void process_cursors(struct inode *inode, enum cursor_action act)
{
oid_t oid;
dir_cursor *start;
struct list_head *head;
reiser4_context *ctx;
struct d_cursor_info *info;
/* this can be called by
*
* kswapd->...->prune_icache->..reiser4_destroy_inode
*
* without reiser4_context
*/
ctx = reiser4_init_context(inode->i_sb);
if (IS_ERR(ctx)) {
warning("vs-23", "failed to init context");
return;
}
assert("nikita-3558", inode != NULL);
info = d_info(inode);
oid = get_inode_oid(inode);
spin_lock_inode(inode);
head = get_readdir_list(inode);
spin_lock(&d_lock);
/* find any cursor for this oid: reference to it is hanging of radix
* tree */
start = lookup(info, (unsigned long)oid);
if (start != NULL) {
dir_cursor *scan;
reiser4_file_fsdata *fsdata;
/* process circular list of cursors for this oid */
scan = start;
do {
dir_cursor *next;
next = list_entry(scan->list.next, dir_cursor, list);
fsdata = scan->fsdata;
assert("nikita-3557", fsdata != NULL);
if (scan->key.oid == oid) {
switch (act) {
case CURSOR_DISPOSE:
list_del_init(&fsdata->dir.linkage);
break;
case CURSOR_LOAD:
list_add(&fsdata->dir.linkage, head);
break;
case CURSOR_KILL:
kill_cursor(scan);
break;
}
}
if (scan == next)
/* last cursor was just killed */
break;
scan = next;
} while (scan != start);
}
spin_unlock(&d_lock);
/* check that we killed 'em all */
assert("nikita-3568",
ergo(act == CURSOR_KILL,
list_empty_careful(get_readdir_list(inode))));
assert("nikita-3569",
ergo(act == CURSOR_KILL, lookup(info, oid) == NULL));
spin_unlock_inode(inode);
reiser4_exit_context(ctx);
}
