1. Fix up the problem: Reiser4 steps to the journal code of other journalling file systems: EXT3 complaints: "called recursively, non-PF_MEMALLOC" Call Trace: [<c0535873>] ext3_write_inode+0x1e/0x3a [<c04b16bc>] __writeback_single_inode+0x193/0x2ad [<c0508d86>] ? flush_some_atom+0x427/0x44d [<c04b1bb5>] ? generic_sync_sb_inodes+0x27c/0x338 [<c04b1b68>] generic_sync_sb_inodes+0x22f/0x338 [<c04b1c8e>] sync_sb_inodes+0x1d/0x20 [<c04b1e1e>] writeback_inodes+0x79/0xb9 [<c047c76a>] balance_dirty_pages_ratelimited_nr+0x119/0x21d [<c0504803>] reiser4_exit_context+0x5f/0xf6 [<c051e5a3>] reiser4_write_careful+0x3ba/0x3cc [<c049bc3c>] ? do_sync_read+0xab/0xe9 [<c0613dae>] ? selinux_file_permission+0x44/0x48 [<c060fea6>] ? security_file_permission+0xf/0x11 [<c051e1e9>] ? reiser4_write_careful+0x0/0x3cc [<c049c494>] vfs_write+0x84/0xdf [<c049c588>] sys_write+0x3b/0x60 [<c0403178>] sysenter_do_call+0x12/0x2d Solution: Drop current->journal_info before calling balance_dirty_pages_ratelimited(). 2. Update comments in plugin_header.h Signed-off-by: Edward Shishkin <edward.shishkin@xxxxxxxxx> --- fs/reiser4/context.c | 13 ++++----- fs/reiser4/plugin/file/cryptcompress.c | 3 -- fs/reiser4/plugin/file/file.c | 9 ++---- fs/reiser4/plugin/file/tail_conversion.c | 14 +++++++--- fs/reiser4/plugin/item/tail.h | 2 - fs/reiser4/plugin/plugin_header.h | 42 ++++++++++++------------------- fs/reiser4/txnmgr.c | 3 +- fs/reiser4/vfs_ops.c | 12 ++++++-- fs/reiser4/vfs_ops.h | 2 - 9 files changed, 49 insertions(+), 51 deletions(-) --- mmotm.orig/fs/reiser4/context.c +++ mmotm/fs/reiser4/context.c @@ -37,8 +37,8 @@ #include "debug.h" #include "super.h" #include "context.h" +#include "vfs_ops.h" /* for reiser4_throttle_write() */ -#include <linux/writeback.h> /* balance_dirty_pages() */ #include <linux/hardirq.h> static void _reiser4_init_context(reiser4_context * context, @@ -139,7 +139,7 @@ int is_in_reiser4_context(void) * because some important lock (like ->i_mutex on the parent directory) is * held. To achieve this, ->nobalance flag can be set in the current context. */ -static void balance_dirty_pages_at(reiser4_context *context) +static void reiser4_throttle_write_at(reiser4_context *context) { reiser4_super_info_data *sbinfo = get_super_private(context->super); @@ -152,7 +152,8 @@ static void balance_dirty_pages_at(reise if (sbinfo != NULL && sbinfo->fake != NULL && context->nr_marked_dirty != 0 && !(current->flags & PF_MEMALLOC)) - balance_dirty_pages_ratelimited(sbinfo->fake->i_mapping); + /* FIXME-EDWARD: throttle with nr_marked_dirty? */ + reiser4_throttle_write(sbinfo->fake, 1); } /* release resources associated with context. @@ -225,10 +226,8 @@ void reiser4_exit_context(reiser4_contex assert("nikita-3021", reiser4_schedulable()); if (context->nr_children == 0) { - if (!context->nobalance) { - reiser4_txn_restart(context); - balance_dirty_pages_at(context); - } + if (!context->nobalance) + reiser4_throttle_write_at(context); /* if filesystem is mounted with -o sync or -o dirsync - commit transaction. FIXME: TXNH_DONT_COMMIT is used to avoid --- mmotm.orig/fs/reiser4/plugin/file/cryptcompress.c +++ mmotm/fs/reiser4/plugin/file/cryptcompress.c @@ -1973,8 +1973,7 @@ static int balance_dirty_page_cluster(st info = cryptcompress_inode_data(inode); mutex_unlock(&info->checkin_mutex); - reiser4_txn_restart_current(); - balance_dirty_pages_ratelimited_nr(inode->i_mapping, nr_dirtied); + reiser4_throttle_write(inode, nr_dirtied); mutex_lock(&info->checkin_mutex); return 0; } --- mmotm.orig/fs/reiser4/plugin/file/file.c +++ mmotm/fs/reiser4/plugin/file/file.c @@ -2227,16 +2227,13 @@ ssize_t write_unix_file(struct file *fil } drop_access(uf_info); ea = NEITHER_OBTAINED; - reiser4_txn_restart(ctx); - current->journal_info = NULL; + /* * tell VM how many pages were dirtied. Maybe number of pages * which were dirty already should not be counted */ - balance_dirty_pages_ratelimited_nr(inode->i_mapping, - (written + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE); - current->journal_info = ctx; - + reiser4_throttle_write(inode, + (written + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE); left -= written; buf += written; *pos += written; --- mmotm.orig/fs/reiser4/plugin/file/tail_conversion.c +++ mmotm/fs/reiser4/plugin/file/tail_conversion.c @@ -486,8 +486,10 @@ int tail2extent(struct unix_file_info *u * on partially converted files. */ drop_exclusive_access(uf_info); - /* throttle the conversion */ - reiser4_throttle_write(inode); + /* throttle the conversion + FIXME-EDWARD: Pass the precise number of pages + that was dirtied */ + reiser4_throttle_write(inode, 1); get_exclusive_access(uf_info); /* @@ -685,8 +687,12 @@ int extent2tail(struct file * file, stru page_cache_release(page); drop_exclusive_access(uf_info); - /* throttle the conversion */ - reiser4_throttle_write(inode); + /* + * 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 --- mmotm.orig/fs/reiser4/plugin/item/tail.h +++ mmotm/fs/reiser4/plugin/item/tail.h @@ -40,8 +40,6 @@ int readpage_tail(void *vp, struct page reiser4_key *append_key_tail(const coord_t *, reiser4_key *); void init_coord_extension_tail(uf_coord_t *, loff_t offset); int get_block_address_tail(const coord_t *, sector_t, sector_t *); -int item_balance_dirty_pages(struct address_space *, const flow_t *, - hint_t *, int back_to_dirty, int set_hint); /* __REISER4_TAIL_H__ */ #endif --- mmotm.orig/fs/reiser4/txnmgr.c +++ mmotm/fs/reiser4/txnmgr.c @@ -2328,7 +2328,8 @@ static void do_jnode_make_dirty(jnode * JF_SET(node, JNODE_DIRTY); - get_current_context()->nr_marked_dirty++; + if (!JF_ISSET(node, JNODE_CLUSTER_PAGE)) + get_current_context()->nr_marked_dirty++; /* We grab2flush_reserve one additional block only if node was not CREATED and jnode_flush did not sort it into neither --- mmotm.orig/fs/reiser4/vfs_ops.c +++ mmotm/fs/reiser4/vfs_ops.c @@ -201,10 +201,16 @@ void reiser4_writeout(struct super_block } while (wbc->nr_to_write > 0); } -void reiser4_throttle_write(struct inode *inode) +/* tell VM how many pages were dirtied */ +void reiser4_throttle_write(struct inode *inode, int nrpages) { - reiser4_txn_restart_current(); - balance_dirty_pages_ratelimited(inode->i_mapping); + reiser4_context *ctx; + + ctx = get_current_context(); + reiser4_txn_restart(ctx); + current->journal_info = NULL; + balance_dirty_pages_ratelimited_nr(inode->i_mapping, nrpages); + current->journal_info = ctx; } const char *REISER4_SUPER_MAGIC_STRING = "ReIsEr4"; --- mmotm.orig/fs/reiser4/vfs_ops.h +++ mmotm/fs/reiser4/vfs_ops.h @@ -30,7 +30,7 @@ extern int reiser4_add_nlink(struct inod extern int reiser4_del_nlink(struct inode *, struct inode *, int); extern int reiser4_start_up_io(struct page *page); -extern void reiser4_throttle_write(struct inode *); +extern void reiser4_throttle_write(struct inode *, int nrpages); extern int jnode_is_releasable(jnode *); #define CAPTURE_APAGE_BURST (1024l) --- mmotm.orig/fs/reiser4/plugin/plugin_header.h +++ mmotm/fs/reiser4/plugin/plugin_header.h @@ -10,32 +10,24 @@ #include "../debug.h" #include "../dformat.h" -/* Every plugin type can be considered as a class of virtual objects - {(type, i) | i = 0, 1, ...}, which has one the following categories - of virtualization: - A - no virtualization; - F - per-file virtualization; - S - per-superblock virtualization; - FIXME-EDWARD: Define every such category */ - -/* Supported plugin types: (id, (virtualization category), short description) */ +/* The list of Reiser4 interfaces */ typedef enum { - REISER4_FILE_PLUGIN_TYPE, /* (F) service VFS enry-points */ - REISER4_DIR_PLUGIN_TYPE, /* (F) service VFS enry-points */ - REISER4_ITEM_PLUGIN_TYPE, /* (F) manage items */ - REISER4_NODE_PLUGIN_TYPE, /* (S) manage formatted nodes */ - REISER4_HASH_PLUGIN_TYPE, /* (F) compute hash */ - REISER4_FIBRATION_PLUGIN_TYPE, /* (F) directory fibrations */ - REISER4_FORMATTING_PLUGIN_TYPE, /* (F) tail-packing policy */ - REISER4_PERM_PLUGIN_TYPE, /* stub (vacancy) */ - REISER4_SD_EXT_PLUGIN_TYPE, /* (A) stat-data extensions */ - REISER4_FORMAT_PLUGIN_TYPE, /* (S) specify disk format */ - REISER4_JNODE_PLUGIN_TYPE, /* (A) in-memory node headers */ - REISER4_CIPHER_PLUGIN_TYPE, /* (F) cipher transform algs */ - REISER4_DIGEST_PLUGIN_TYPE, /* (F) digest transform algs */ - REISER4_COMPRESSION_PLUGIN_TYPE, /* (F) compression tfm algs */ - REISER4_COMPRESSION_MODE_PLUGIN_TYPE, /* (F) compression heuristic */ - REISER4_CLUSTER_PLUGIN_TYPE, /* (F) size of logical cluster */ + REISER4_FILE_PLUGIN_TYPE, /* manage VFS objects */ + REISER4_DIR_PLUGIN_TYPE, /* manage directories */ + REISER4_ITEM_PLUGIN_TYPE, /* manage items */ + REISER4_NODE_PLUGIN_TYPE, /* manage formatted nodes */ + REISER4_HASH_PLUGIN_TYPE, /* hash methods */ + REISER4_FIBRATION_PLUGIN_TYPE, /* directory fibrations */ + REISER4_FORMATTING_PLUGIN_TYPE, /* dispatching policy */ + REISER4_PERM_PLUGIN_TYPE, /* stub (vacancy) */ + REISER4_SD_EXT_PLUGIN_TYPE, /* manage stat-data extensions */ + REISER4_FORMAT_PLUGIN_TYPE, /* disk format specifications */ + REISER4_JNODE_PLUGIN_TYPE, /* manage in-memory headers */ + REISER4_CIPHER_PLUGIN_TYPE, /* cipher transform methods */ + REISER4_DIGEST_PLUGIN_TYPE, /* digest transform methods */ + REISER4_COMPRESSION_PLUGIN_TYPE, /* compression methods */ + REISER4_COMPRESSION_MODE_PLUGIN_TYPE, /* dispatching policies */ + REISER4_CLUSTER_PLUGIN_TYPE, /* manage logical clusters */ REISER4_PLUGIN_TYPES } reiser4_plugin_type;