From 5f98fe60c291075702ef61f711ba89f724d2e75e Mon Sep 17 00:00:00 2001 From: Raghavendra G Date: Tue, 19 Jul 2011 16:38:44 +0530 Subject: performance/io-cache: hold lock on ioc_inode whereever a page is being accessed/modified Change-Id: I77e94166d4ad7a9ac87127de840add7278992bf3 BUG: 2939 Reviewed-on: http://review.gluster.com/24 Tested-by: Gluster Build System Reviewed-by: Anand Avati --- xlators/performance/io-cache/src/io-cache.c | 1552 +++++++++++++------------- xlators/performance/io-cache/src/io-cache.h | 15 +- xlators/performance/io-cache/src/ioc-inode.c | 26 +- xlators/performance/io-cache/src/page.c | 203 +++- 4 files changed, 950 insertions(+), 846 deletions(-) diff --git a/xlators/performance/io-cache/src/io-cache.c b/xlators/performance/io-cache/src/io-cache.c index da6c556b95c..1b16342d607 100644 --- a/xlators/performance/io-cache/src/io-cache.c +++ b/xlators/performance/io-cache/src/io-cache.c @@ -54,50 +54,55 @@ ioc_hashfn (void *data, int len) inline ioc_inode_t * ioc_inode_reupdate (ioc_inode_t *ioc_inode) { - ioc_table_t *table = ioc_inode->table; + ioc_table_t *table = NULL; + + table = ioc_inode->table; - list_add_tail (&ioc_inode->inode_lru, - &table->inode_lru[ioc_inode->weight]); + list_add_tail (&ioc_inode->inode_lru, + &table->inode_lru[ioc_inode->weight]); - return ioc_inode; + return ioc_inode; } inline ioc_inode_t * ioc_get_inode (dict_t *dict, char *name) { - ioc_inode_t *ioc_inode = NULL; - data_t *ioc_inode_data = dict_get (dict, name); - ioc_table_t *table = NULL; - - if (ioc_inode_data) { - ioc_inode = data_to_ptr (ioc_inode_data); - table = ioc_inode->table; - - ioc_table_lock (table); - { - if (list_empty (&ioc_inode->inode_lru)) { - ioc_inode = ioc_inode_reupdate (ioc_inode); - } - } - ioc_table_unlock (table); - } - - return ioc_inode; + ioc_inode_t *ioc_inode = NULL; + data_t *ioc_inode_data = NULL; + ioc_table_t *table = NULL; + + ioc_inode_data = dict_get (dict, name); + if (ioc_inode_data) { + ioc_inode = data_to_ptr (ioc_inode_data); + table = ioc_inode->table; + + ioc_table_lock (table); + { + if (list_empty (&ioc_inode->inode_lru)) { + ioc_inode = ioc_inode_reupdate (ioc_inode); + } + } + ioc_table_unlock (table); + } + + return ioc_inode; } int32_t ioc_inode_need_revalidate (ioc_inode_t *ioc_inode) { - int8_t need_revalidate = 0; - struct timeval tv = {0,}; - ioc_table_t *table = ioc_inode->table; + int8_t need_revalidate = 0; + struct timeval tv = {0,}; + ioc_table_t *table = NULL; + + table = ioc_inode->table; - gettimeofday (&tv, NULL); + gettimeofday (&tv, NULL); - if (time_elapsed (&tv, &ioc_inode->cache.tv) >= table->cache_timeout) - need_revalidate = 1; + if (time_elapsed (&tv, &ioc_inode->cache.tv) >= table->cache_timeout) + need_revalidate = 1; - return need_revalidate; + return need_revalidate; } /* @@ -110,41 +115,41 @@ ioc_inode_need_revalidate (ioc_inode_t *ioc_inode) int64_t __ioc_inode_flush (ioc_inode_t *ioc_inode) { - ioc_page_t *curr = NULL, *next = NULL; - int64_t destroy_size = 0; - int64_t ret = 0; + ioc_page_t *curr = NULL, *next = NULL; + int64_t destroy_size = 0; + int64_t ret = 0; - list_for_each_entry_safe (curr, next, &ioc_inode->cache.page_lru, + list_for_each_entry_safe (curr, next, &ioc_inode->cache.page_lru, page_lru) { - ret = ioc_page_destroy (curr); + ret = __ioc_page_destroy (curr); - if (ret != -1) - destroy_size += ret; - } + if (ret != -1) + destroy_size += ret; + } - return destroy_size; + return destroy_size; } void ioc_inode_flush (ioc_inode_t *ioc_inode) { - int64_t destroy_size = 0; - - ioc_inode_lock (ioc_inode); - { - destroy_size = __ioc_inode_flush (ioc_inode); - } - ioc_inode_unlock (ioc_inode); - - if (destroy_size) { - ioc_table_lock (ioc_inode->table); - { - ioc_inode->table->cache_used -= destroy_size; - } - ioc_table_unlock (ioc_inode->table); - } - - return; + int64_t destroy_size = 0; + + ioc_inode_lock (ioc_inode); + { + destroy_size = __ioc_inode_flush (ioc_inode); + } + ioc_inode_unlock (ioc_inode); + + if (destroy_size) { + ioc_table_lock (ioc_inode->table); + { + ioc_inode->table->cache_used -= destroy_size; + } + ioc_table_unlock (ioc_inode->table); + } + + return; } int32_t @@ -152,44 +157,44 @@ ioc_setattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iatt *preop, struct iatt *postop) { - STACK_UNWIND_STRICT (setattr, frame, op_ret, op_errno, preop, postop); - return 0; + STACK_UNWIND_STRICT (setattr, frame, op_ret, op_errno, preop, postop); + return 0; } int32_t ioc_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *stbuf, int32_t valid) { - uint64_t ioc_inode = 0; + uint64_t ioc_inode = 0; - inode_ctx_get (loc->inode, this, &ioc_inode); + inode_ctx_get (loc->inode, this, &ioc_inode); - if (ioc_inode + if (ioc_inode && ((valid & GF_SET_ATTR_ATIME) || (valid & GF_SET_ATTR_MTIME))) - ioc_inode_flush ((ioc_inode_t *)(long)ioc_inode); + ioc_inode_flush ((ioc_inode_t *)(long)ioc_inode); - STACK_WIND (frame, ioc_setattr_cbk, FIRST_CHILD (this), + STACK_WIND (frame, ioc_setattr_cbk, FIRST_CHILD (this), FIRST_CHILD (this)->fops->setattr, loc, stbuf, valid); - return 0; + return 0; } int32_t ioc_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, inode_t *inode, - struct iatt *stbuf, dict_t *dict, struct iatt *postparent) + int32_t op_ret, int32_t op_errno, inode_t *inode, + struct iatt *stbuf, dict_t *dict, struct iatt *postparent) { - ioc_inode_t *ioc_inode = NULL; - ioc_table_t *table = NULL; - uint8_t cache_still_valid = 0; - uint64_t tmp_ioc_inode = 0; - uint32_t weight = 0xffffffff; - const char *path = NULL; - ioc_local_t *local = NULL; - - if (op_ret != 0) - goto out; + ioc_inode_t *ioc_inode = NULL; + ioc_table_t *table = NULL; + uint8_t cache_still_valid = 0; + uint64_t tmp_ioc_inode = 0; + uint32_t weight = 0xffffffff; + const char *path = NULL; + ioc_local_t *local = NULL; + + if (op_ret != 0) + goto out; local = frame->local; if (local == NULL) { @@ -256,16 +261,16 @@ out: loc_wipe (&local->file_loc); } - STACK_UNWIND_STRICT (lookup, frame, op_ret, op_errno, inode, stbuf, + STACK_UNWIND_STRICT (lookup, frame, op_ret, op_errno, inode, stbuf, dict, postparent); - return 0; + return 0; } int32_t ioc_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc, - dict_t *xattr_req) + dict_t *xattr_req) { - ioc_local_t *local = NULL; + ioc_local_t *local = NULL; int32_t op_errno = -1, ret = -1; local = GF_CALLOC (1, sizeof (*local), @@ -285,16 +290,16 @@ ioc_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc, frame->local = local; - STACK_WIND (frame, ioc_lookup_cbk, FIRST_CHILD (this), - FIRST_CHILD (this)->fops->lookup, loc, xattr_req); + STACK_WIND (frame, ioc_lookup_cbk, FIRST_CHILD (this), + FIRST_CHILD (this)->fops->lookup, loc, xattr_req); return 0; unwind: - STACK_UNWIND_STRICT (lookup, frame, -1, op_errno, NULL, NULL, + STACK_UNWIND_STRICT (lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL); - return 0; + return 0; } /* @@ -308,14 +313,14 @@ unwind: int32_t ioc_forget (xlator_t *this, inode_t *inode) { - uint64_t ioc_inode = 0; + uint64_t ioc_inode = 0; - inode_ctx_get (inode, this, &ioc_inode); + inode_ctx_get (inode, this, &ioc_inode); - if (ioc_inode) - ioc_inode_destroy ((ioc_inode_t *)(long)ioc_inode); + if (ioc_inode) + ioc_inode_destroy ((ioc_inode_t *)(long)ioc_inode); - return 0; + return 0; } @@ -332,86 +337,87 @@ ioc_forget (xlator_t *this, inode_t *inode) */ int32_t ioc_cache_validate_cbk (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, struct iatt *stbuf) + int32_t op_ret, int32_t op_errno, struct iatt *stbuf) { - ioc_local_t *local = NULL; - ioc_inode_t *ioc_inode = NULL; - size_t destroy_size = 0; - struct iatt *local_stbuf = NULL; + ioc_local_t *local = NULL; + ioc_inode_t *ioc_inode = NULL; + size_t destroy_size = 0; + struct iatt *local_stbuf = NULL; local = frame->local; - ioc_inode = local->inode; + ioc_inode = local->inode; local_stbuf = stbuf; - if ((op_ret == -1) || - ((op_ret >= 0) && !ioc_cache_still_valid(ioc_inode, stbuf))) { - gf_log (ioc_inode->table->xl->name, GF_LOG_DEBUG, - "cache for inode(%p) is invalid. flushing all pages", - ioc_inode); - /* NOTE: only pages with no waiting frames are flushed by - * ioc_inode_flush. page_fault will be generated for all - * the pages which have waiting frames by ioc_inode_wakeup() - */ - ioc_inode_lock (ioc_inode); - { - destroy_size = __ioc_inode_flush (ioc_inode); - if (op_ret >= 0) { - ioc_inode->cache.mtime = stbuf->ia_mtime; - ioc_inode->cache.mtime_nsec = stbuf->ia_mtime_nsec; + if ((op_ret == -1) || + ((op_ret >= 0) && !ioc_cache_still_valid(ioc_inode, stbuf))) { + gf_log (ioc_inode->table->xl->name, GF_LOG_DEBUG, + "cache for inode(%p) is invalid. flushing all pages", + ioc_inode); + /* NOTE: only pages with no waiting frames are flushed by + * ioc_inode_flush. page_fault will be generated for all + * the pages which have waiting frames by ioc_inode_wakeup() + */ + ioc_inode_lock (ioc_inode); + { + destroy_size = __ioc_inode_flush (ioc_inode); + if (op_ret >= 0) { + ioc_inode->cache.mtime = stbuf->ia_mtime; + ioc_inode->cache.mtime_nsec + = stbuf->ia_mtime_nsec; } - } - ioc_inode_unlock (ioc_inode); - local_stbuf = NULL; - } - - if (destroy_size) { - ioc_table_lock (ioc_inode->table); - { - ioc_inode->table->cache_used -= destroy_size; - } - ioc_table_unlock (ioc_inode->table); - } - - if (op_ret < 0) - local_stbuf = NULL; - - ioc_inode_lock (ioc_inode); - { - gettimeofday (&ioc_inode->cache.tv, NULL); - } - ioc_inode_unlock (ioc_inode); - - ioc_inode_wakeup (frame, ioc_inode, local_stbuf); - - /* any page-fault initiated by ioc_inode_wakeup() will have its own - * fd_ref on fd, safe to unref validate frame's private copy - */ - fd_unref (local->fd); - - STACK_DESTROY (frame->root); - - return 0; + } + ioc_inode_unlock (ioc_inode); + local_stbuf = NULL; + } + + if (destroy_size) { + ioc_table_lock (ioc_inode->table); + { + ioc_inode->table->cache_used -= destroy_size; + } + ioc_table_unlock (ioc_inode->table); + } + + if (op_ret < 0) + local_stbuf = NULL; + + ioc_inode_lock (ioc_inode); + { + gettimeofday (&ioc_inode->cache.tv, NULL); + } + ioc_inode_unlock (ioc_inode); + + ioc_inode_wakeup (frame, ioc_inode, local_stbuf); + + /* any page-fault initiated by ioc_inode_wakeup() will have its own + * fd_ref on fd, safe to unref validate frame's private copy + */ + fd_unref (local->fd); + + STACK_DESTROY (frame->root); + + return 0; } int32_t ioc_wait_on_inode (ioc_inode_t *ioc_inode, ioc_page_t *page) { - ioc_waitq_t *waiter = NULL, *trav = NULL; - uint32_t page_found = 0; - int32_t ret = 0; - - trav = ioc_inode->waitq; - - while (trav) { - if (trav->data == page) { - page_found = 1; - break; - } - trav = trav->next; - } - - if (!page_found) { - waiter = GF_CALLOC (1, sizeof (ioc_waitq_t), + ioc_waitq_t *waiter = NULL, *trav = NULL; + uint32_t page_found = 0; + int32_t ret = 0; + + trav = ioc_inode->waitq; + + while (trav) { + if (trav->data == page) { + page_found = 1; + break; + } + trav = trav->next; + } + + if (!page_found) { + waiter = GF_CALLOC (1, sizeof (ioc_waitq_t), gf_ioc_mt_ioc_waitq_t); if (waiter == NULL) { gf_log (ioc_inode->table->xl->name, GF_LOG_ERROR, @@ -420,13 +426,13 @@ ioc_wait_on_inode (ioc_inode_t *ioc_inode, ioc_page_t *page) goto out; } - waiter->data = page; - waiter->next = ioc_inode->waitq; - ioc_inode->waitq = waiter; - } + waiter->data = page; + waiter->next = ioc_inode->waitq; + ioc_inode->waitq = waiter; + } out: - return ret; + return ret; } /* @@ -439,15 +445,15 @@ out: */ int32_t ioc_cache_validate (call_frame_t *frame, ioc_inode_t *ioc_inode, fd_t *fd, - ioc_page_t *page) + ioc_page_t *page) { - call_frame_t *validate_frame = NULL; - ioc_local_t *validate_local = NULL; - ioc_local_t *local = NULL; - int32_t ret = 0; + call_frame_t *validate_frame = NULL; + ioc_local_t *validate_local = NULL; + ioc_local_t *local = NULL; + int32_t ret = 0; local = frame->local; - validate_local = GF_CALLOC (1, sizeof (ioc_local_t), + validate_local = GF_CALLOC (1, sizeof (ioc_local_t), gf_ioc_mt_ioc_local_t); if (validate_local == NULL) { ret = -1; @@ -458,7 +464,7 @@ ioc_cache_validate (call_frame_t *frame, ioc_inode_t *ioc_inode, fd_t *fd, goto out; } - validate_frame = copy_frame (frame); + validate_frame = copy_frame (frame); if (validate_frame == NULL) { ret = -1; local->op_ret = -1; @@ -469,44 +475,44 @@ ioc_cache_validate (call_frame_t *frame, ioc_inode_t *ioc_inode, fd_t *fd, goto out; } - validate_local->fd = fd_ref (fd); - validate_local->inode = ioc_inode; - validate_frame->local = validate_local; + validate_local->fd = fd_ref (fd); + validate_local->inode = ioc_inode; + validate_frame->local = validate_local; - STACK_WIND (validate_frame, ioc_cache_validate_cbk, + STACK_WIND (validate_frame, ioc_cache_validate_cbk, FIRST_CHILD (frame->this), FIRST_CHILD (frame->this)->fops->fstat, fd); out: - return ret; + return ret; } inline uint32_t is_match (const char *path, const char *pattern) { - int32_t ret = 0; + int32_t ret = 0; - ret = fnmatch (pattern, path, FNM_NOESCAPE); + ret = fnmatch (pattern, path, FNM_NOESCAPE); - return (ret == 0); + return (ret == 0); } uint32_t ioc_get_priority (ioc_table_t *table, const char *path) { - uint32_t priority = 1; - struct ioc_priority *curr = NULL; + uint32_t priority = 1; + struct ioc_priority *curr = NULL; - if (list_empty(&table->priority_list)) - return priority; + if (list_empty(&table->priority_list)) + return priority; - priority = 0; - list_for_each_entry (curr, &table->priority_list, list) { - if (is_match (path, curr->pattern)) - priority = curr->priority; - } + priority = 0; + list_for_each_entry (curr, &table->priority_list, list) { + if (is_match (path, curr->pattern)) + priority = curr->priority; + } - return priority; + return priority; } /* @@ -522,13 +528,13 @@ ioc_get_priority (ioc_table_t *table, const char *path) */ int32_t ioc_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, - int32_t op_errno, fd_t *fd) + int32_t op_errno, fd_t *fd) { - uint64_t tmp_ioc_inode = 0; - ioc_local_t *local = NULL; - ioc_table_t *table = NULL; - ioc_inode_t *ioc_inode = NULL; - uint32_t weight = 0xffffffff; + uint64_t tmp_ioc_inode = 0; + ioc_local_t *local = NULL; + ioc_table_t *table = NULL; + ioc_inode_t *ioc_inode = NULL; + uint32_t weight = 0xffffffff; local = frame->local; if (!this || !this->private) { @@ -565,7 +571,7 @@ ioc_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, /* O_DIRECT is only for one fd, not the inode * as a whole */ - fd_ctx_set (fd, this, 1); + fd_ctx_set (fd, this, 1); } if ((local->wbflags & GF_OPEN_NOWB) != 0) { /* disable caching as asked by NFS */ @@ -575,7 +581,7 @@ ioc_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, /* weight = 0, we disable caching on it */ if (weight == 0) { /* we allow a pattern-matched cache disable this way - */ + */ fd_ctx_set (fd, this, 1); } } @@ -586,7 +592,7 @@ out: STACK_UNWIND_STRICT (open, frame, op_ret, op_errno, fd); - return 0; + return 0; } /* @@ -605,15 +611,15 @@ out: int32_t ioc_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, fd_t *fd, - inode_t *inode, struct iatt *buf, struct iatt *preparent, + inode_t *inode, struct iatt *buf, struct iatt *preparent, struct iatt *postparent) { - ioc_local_t *local = NULL; - ioc_table_t *table = NULL; - ioc_inode_t *ioc_inode = NULL; - uint32_t weight = 0xffffffff; - const char *path = NULL; - int ret = -1; + ioc_local_t *local = NULL; + ioc_table_t *table = NULL; + ioc_inode_t *ioc_inode = NULL; + uint32_t weight = 0xffffffff; + const char *path = NULL; + int ret = -1; local = frame->local; if (!this || !this->private) { @@ -625,7 +631,7 @@ ioc_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this, table = this->private; path = local->file_loc.path; - if (op_ret != -1) { + if (op_ret != -1) { /* assign weight */ weight = ioc_get_priority (table, path); @@ -639,7 +645,8 @@ ioc_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this, if ((table->min_file_size > ioc_inode->ia_size) || ((table->max_file_size >= 0) - && (table->max_file_size < ioc_inode->ia_size))) { + && (table->max_file_size + < ioc_inode->ia_size))) { ret = fd_ctx_set (fd, this, 1); } } @@ -648,28 +655,28 @@ ioc_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this, inode_ctx_put (fd->inode, this, (uint64_t)(long)ioc_inode); - /* If O_DIRECT open, we disable caching on it */ - if (local->flags & O_DIRECT) - /* + /* If O_DIRECT open, we disable caching on it */ + if (local->flags & O_DIRECT) + /* * O_DIRECT is only for one fd, not the inode - * as a whole */ - ret = fd_ctx_set (fd, this, 1); + * as a whole */ + ret = fd_ctx_set (fd, this, 1); /* if weight == 0, we disable caching on it */ - if (!weight) - /* we allow a pattern-matched cache disable this way */ - ret = fd_ctx_set (fd, this, 1); + if (!weight) + /* we allow a pattern-matched cache disable this way */ + ret = fd_ctx_set (fd, this, 1); - } + } out: - frame->local = NULL; - GF_FREE (local); + frame->local = NULL; + GF_FREE (local); STACK_UNWIND_STRICT (create, frame, op_ret, op_errno, fd, inode, buf, preparent, postparent); - return 0; + return 0; } @@ -680,10 +687,10 @@ ioc_mknod_cbk (call_frame_t *frame, void *cookie, xlator_t *this, struct iatt *postparent) { ioc_local_t *local = NULL; - ioc_table_t *table = NULL; - ioc_inode_t *ioc_inode = NULL; - uint32_t weight = 0xffffffff; - const char *path = NULL; + ioc_table_t *table = NULL; + ioc_inode_t *ioc_inode = NULL; + uint32_t weight = 0xffffffff; + const char *path = NULL; local = frame->local; if (!this || !this->private) { @@ -695,7 +702,7 @@ ioc_mknod_cbk (call_frame_t *frame, void *cookie, xlator_t *this, table = this->private; path = local->file_loc.path; - if (op_ret != -1) { + if (op_ret != -1) { /* assign weight */ weight = ioc_get_priority (table, path); @@ -711,17 +718,17 @@ ioc_mknod_cbk (call_frame_t *frame, void *cookie, xlator_t *this, inode_ctx_put (inode, this, (uint64_t)(long)ioc_inode); - } + } out: - frame->local = NULL; + frame->local = NULL; loc_wipe (&local->file_loc); - GF_FREE (local); + GF_FREE (local); - STACK_UNWIND_STRICT (mknod, frame, op_ret, op_errno, inode, buf, + STACK_UNWIND_STRICT (mknod, frame, op_ret, op_errno, inode, buf, preparent, postparent); - return 0; + return 0; } @@ -729,7 +736,7 @@ int ioc_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode, dev_t rdev, dict_t *params) { - ioc_local_t *local = NULL; + ioc_local_t *local = NULL; int32_t op_errno = -1, ret = -1; local = GF_CALLOC (1, sizeof (*local), @@ -749,10 +756,10 @@ ioc_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode, frame->local = local; - STACK_WIND (frame, ioc_mknod_cbk, - FIRST_CHILD(this), - FIRST_CHILD(this)->fops->mknod, - loc, mode, rdev, params); + STACK_WIND (frame, ioc_mknod_cbk, + FIRST_CHILD(this), + FIRST_CHILD(this)->fops->mknod, + loc, mode, rdev, params); return 0; unwind: @@ -761,10 +768,10 @@ unwind: GF_FREE (local); } - STACK_UNWIND_STRICT (mknod, frame, -1, op_errno, NULL, NULL, + STACK_UNWIND_STRICT (mknod, frame, -1, op_errno, NULL, NULL, NULL, NULL); - return 0; + return 0; } @@ -778,10 +785,10 @@ unwind: */ int32_t ioc_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, - fd_t *fd, int32_t wbflags) + fd_t *fd, int32_t wbflags) { - ioc_local_t *local = NULL; + ioc_local_t *local = NULL; local = GF_CALLOC (1, sizeof (ioc_local_t), gf_ioc_mt_ioc_local_t); if (local == NULL) { @@ -790,17 +797,17 @@ ioc_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, return 0; } - local->flags = flags; - local->file_loc.path = loc->path; - local->file_loc.inode = loc->inode; + local->flags = flags; + local->file_loc.path = loc->path; + local->file_loc.inode = loc->inode; local->wbflags = wbflags; - frame->local = local; + frame->local = local; - STACK_WIND (frame, ioc_open_cbk, FIRST_CHILD(this), + STACK_WIND (frame, ioc_open_cbk, FIRST_CHILD(this), FIRST_CHILD(this)->fops->open, loc, flags, fd, wbflags); - return 0; + return 0; } /* @@ -815,9 +822,9 @@ ioc_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, */ int32_t ioc_create (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, - mode_t mode, fd_t *fd, dict_t *params) + mode_t mode, fd_t *fd, dict_t *params) { - ioc_local_t *local = NULL; + ioc_local_t *local = NULL; local = GF_CALLOC (1, sizeof (ioc_local_t), gf_ioc_mt_ioc_local_t); if (local == NULL) { @@ -827,15 +834,15 @@ ioc_create (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, return 0; } - local->flags = flags; - local->file_loc.path = loc->path; - frame->local = local; + local->flags = flags; + local->file_loc.path = loc->path; + frame->local = local; - STACK_WIND (frame, ioc_create_cbk, FIRST_CHILD(this), - FIRST_CHILD(this)->fops->create, loc, flags, mode, + STACK_WIND (frame, ioc_create_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->create, loc, flags, mode, fd, params); - return 0; + return 0; } @@ -852,7 +859,7 @@ ioc_create (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags, int32_t ioc_release (xlator_t *this, fd_t *fd) { - return 0; + return 0; } /* @@ -869,30 +876,30 @@ ioc_release (xlator_t *this, fd_t *fd) int32_t ioc_readv_disabled_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, int32_t op_errno, struct iovec *vector, - int32_t count, struct iatt *stbuf, + int32_t count, struct iatt *stbuf, struct iobref *iobref) { - STACK_UNWIND_STRICT (readv, frame, op_ret, op_errno, vector, count, + STACK_UNWIND_STRICT (readv, frame, op_ret, op_errno, vector, count, stbuf, iobref); - return 0; + return 0; } int32_t ioc_need_prune (ioc_table_t *table) { - int64_t cache_difference = 0; - - ioc_table_lock (table); - { - cache_difference = table->cache_used - table->cache_size; - } - ioc_table_unlock (table); - - if (cache_difference > 0) - return 1; - else - return 0; + int64_t cache_difference = 0; + + ioc_table_lock (table); + { + cache_difference = table->cache_used - table->cache_size; + } + ioc_table_unlock (table); + + if (cache_difference > 0) + return 1; + else + return 0; } /* @@ -907,122 +914,129 @@ void ioc_dispatch_requests (call_frame_t *frame, ioc_inode_t *ioc_inode, fd_t *fd, off_t offset, size_t size) { - ioc_local_t *local = NULL; - ioc_table_t *table = NULL; - ioc_page_t *trav = NULL; - ioc_waitq_t *waitq = NULL; - off_t rounded_offset = 0; - off_t rounded_end = 0; - off_t trav_offset = 0; - int32_t fault = 0; - size_t trav_size = 0; - off_t local_offset = 0; - int32_t ret = -1; - int8_t need_validate = 0; - int8_t might_need_validate = 0; /* - * if a page exists, do we need - * to validate it? - */ + ioc_local_t *local = NULL; + ioc_table_t *table = NULL; + ioc_page_t *trav = NULL; + ioc_waitq_t *waitq = NULL; + off_t rounded_offset = 0; + off_t rounded_end = 0; + off_t trav_offset = 0; + int32_t fault = 0; + size_t trav_size = 0; + off_t local_offset = 0; + int32_t ret = -1; + int8_t need_validate = 0; + int8_t might_need_validate = 0; /* + * if a page exists, do we need + * to validate it? + */ local = frame->local; table = ioc_inode->table; - rounded_offset = floor (offset, table->page_size); - rounded_end = roof (offset + size, table->page_size); - trav_offset = rounded_offset; - - /* once a frame does read, it should be waiting on something */ - local->wait_count++; - - /* Requested region can fall in three different pages, - * 1. Ready - region is already in cache, we just have to serve it. - * 2. In-transit - page fault has been generated on this page, we need - * to wait till the page is ready - * 3. Fault - page is not in cache, we have to generate a page fault - */ - - might_need_validate = ioc_inode_need_revalidate (ioc_inode); - - while (trav_offset < rounded_end) { - ioc_inode_lock (ioc_inode); - //{ - - /* look for requested region in the cache */ - trav = ioc_page_get (ioc_inode, trav_offset); - - local_offset = max (trav_offset, offset); - trav_size = min (((offset+size) - local_offset), - table->page_size); - - if (!trav) { - /* page not in cache, we need to generate page fault */ - trav = ioc_page_create (ioc_inode, trav_offset); - fault = 1; - if (!trav) { - gf_log (frame->this->name, GF_LOG_CRITICAL, - "out of memory"); - local->op_ret = -1; - local->op_errno = ENOMEM; - goto out; - } - } - - ioc_wait_on_page (trav, frame, local_offset, trav_size); - - if (trav->ready) { - /* page found in cache */ - if (!might_need_validate && !ioc_inode->waitq) { - /* fresh enough */ - gf_log (frame->this->name, GF_LOG_TRACE, - "cache hit for trav_offset=%"PRId64"" - "/local_offset=%"PRId64"", - trav_offset, local_offset); - waitq = ioc_page_wakeup (trav); - } else { - /* if waitq already exists, fstat revalidate is - already on the way */ - if (!ioc_inode->waitq) { - need_validate = 1; - } - - ret = ioc_wait_on_inode (ioc_inode, trav); - if (ret < 0) { - local->op_ret = -1; - local->op_errno = -ret; - need_validate = 0; + rounded_offset = floor (offset, table->page_size); + rounded_end = roof (offset + size, table->page_size); + trav_offset = rounded_offset; + + /* once a frame does read, it should be waiting on something */ + local->wait_count++; - waitq = ioc_page_wakeup (trav); - ioc_inode_unlock (ioc_inode); + /* Requested region can fall in three different pages, + * 1. Ready - region is already in cache, we just have to serve it. + * 2. In-transit - page fault has been generated on this page, we need + * to wait till the page is ready + * 3. Fault - page is not in cache, we have to generate a page fault + */ - ioc_waitq_return (waitq); - waitq = NULL; + might_need_validate = ioc_inode_need_revalidate (ioc_inode); + + while (trav_offset < rounded_end) { + ioc_inode_lock (ioc_inode); + { + /* look for requested region in the cache */ + trav = __ioc_page_get (ioc_inode, trav_offset); + + local_offset = max (trav_offset, offset); + trav_size = min (((offset+size) - local_offset), + table->page_size); + + if (!trav) { + /* page not in cache, we need to generate page + * fault + */ + trav = __ioc_page_create (ioc_inode, + trav_offset); + fault = 1; + if (!trav) { + gf_log (frame->this->name, + GF_LOG_CRITICAL, + "out of memory"); + local->op_ret = -1; + local->op_errno = ENOMEM; goto out; } - } - } - - //} - ioc_inode_unlock (ioc_inode); - - ioc_waitq_return (waitq); - waitq = NULL; - - if (fault) { - fault = 0; - /* new page created, increase the table->cache_used */ - ioc_page_fault (ioc_inode, frame, fd, trav_offset); - } - - if (need_validate) { - need_validate = 0; - gf_log (frame->this->name, GF_LOG_TRACE, - "sending validate request for " - "inode(%"PRId64") at offset=%"PRId64"", - fd->inode->ino, trav_offset); - ret = ioc_cache_validate (frame, ioc_inode, fd, trav); + } + + __ioc_wait_on_page (trav, frame, local_offset, + trav_size); + + if (trav->ready) { + /* page found in cache */ + if (!might_need_validate && !ioc_inode->waitq) { + /* fresh enough */ + gf_log (frame->this->name, GF_LOG_TRACE, + "cache hit for trav_offset=%" + PRId64"/local_offset=%"PRId64"", + trav_offset, local_offset); + waitq = __ioc_page_wakeup (trav); + } else { + /* if waitq already exists, fstat + * revalidate is + * already on the way + */ + if (!ioc_inode->waitq) { + need_validate = 1; + } + + ret = ioc_wait_on_inode (ioc_inode, + trav); + if (ret < 0) { + local->op_ret = -1; + local->op_errno = -ret; + need_validate = 0; + + waitq = __ioc_page_wakeup (trav); + ioc_inode_unlock (ioc_inode); + + ioc_waitq_return (waitq); + waitq = NULL; + goto out; + } + } + } + + } + ioc_inode_unlock (ioc_inode); + + ioc_waitq_return (waitq); + waitq = NULL; + + if (fault) { + fault = 0; + /* new page created, increase the table->cache_used */ + ioc_page_fault (ioc_inode, frame, fd, trav_offset); + } + + if (need_validate) { + need_validate = 0; + gf_log (frame->this->name, GF_LOG_TRACE, + "sending validate request for " + "inode(%"PRId64") at offset=%"PRId64"", + fd->inode->ino, trav_offset); + ret = ioc_cache_validate (frame, ioc_inode, fd, trav); if (ret == -1) { ioc_inode_lock (ioc_inode); { - waitq = ioc_page_wakeup (trav); + waitq = __ioc_page_wakeup (trav); } ioc_inode_unlock (ioc_inode); @@ -1030,19 +1044,19 @@ ioc_dispatch_requests (call_frame_t *frame, ioc_inode_t *ioc_inode, fd_t *fd, waitq = NULL; goto out; } - } + } - trav_offset += table->page_size; - } + trav_offset += table->page_size; + } out: - ioc_frame_return (frame); + ioc_frame_return (frame); - if (ioc_need_prune (ioc_inode->table)) { - ioc_prune (ioc_inode->table); - } + if (ioc_need_prune (ioc_inode->table)) { + ioc_prune (ioc_inode->table); + } - return; + return; } @@ -1058,30 +1072,30 @@ out: */ int32_t ioc_readv (call_frame_t *frame, xlator_t *this, fd_t *fd, - size_t size, off_t offset) + size_t size, off_t offset) { - uint64_t tmp_ioc_inode = 0; - ioc_inode_t *ioc_inode = NULL; - ioc_local_t *local = NULL; - uint32_t weight = 0; - ioc_table_t *table = NULL; - uint32_t num_pages = 0; - int32_t op_errno = -1; + uint64_t tmp_ioc_inode = 0; + ioc_inode_t *ioc_inode = NULL; + ioc_local_t *local = NULL; + uint32_t weight = 0; + ioc_table_t *table = NULL; + uint32_t num_pages = 0; + int32_t op_errno = -1; if (!this) { goto out; } - inode_ctx_get (fd->inode, this, &tmp_ioc_inode); - ioc_inode = (ioc_inode_t *)(long)tmp_ioc_inode; - if (!ioc_inode) { - /* caching disabled, go ahead with normal readv */ - STACK_WIND (frame, ioc_readv_disabled_cbk, + inode_ctx_get (fd->inode, this, &tmp_ioc_inode); + ioc_inode = (ioc_inode_t *)(long)tmp_ioc_inode; + if (!ioc_inode) { + /* caching disabled, go ahead with normal readv */ + STACK_WIND (frame, ioc_readv_disabled_cbk, FIRST_CHILD (frame->this), - FIRST_CHILD (frame->this)->fops->readv, fd, size, + FIRST_CHILD (frame->this)->fops->readv, fd, size, offset); - return 0; - } + return 0; + } table = this->private; @@ -1099,7 +1113,7 @@ ioc_readv (call_frame_t *frame, xlator_t *this, fd_t *fd, num_pages = (table->cache_size / table->page_size) + ((table->cache_size % table->page_size) - ? 1 : 0); + ? 1 : 0); table->mem_pool = mem_pool_new (rbthash_entry_t, num_pages); @@ -1120,9 +1134,9 @@ ioc_readv (call_frame_t *frame, xlator_t *this, fd_t *fd, if (!ioc_inode->cache.page_table) { ioc_inode->cache.page_table = rbthash_table_init - (IOC_PAGE_TABLE_BUCKET_COUNT, - ioc_hashfn, NULL, 0, - table->mem_pool); + (IOC_PAGE_TABLE_BUCKET_COUNT, + ioc_hashfn, NULL, 0, + table->mem_pool); if (ioc_inode->cache.page_table == NULL) { op_errno = ENOMEM; @@ -1133,47 +1147,47 @@ ioc_readv (call_frame_t *frame, xlator_t *this, fd_t *fd, } ioc_inode_unlock (ioc_inode); - if (!fd_ctx_get (fd, this, NULL)) { - /* disable caching for this fd, go ahead with normal readv */ - STACK_WIND (frame, ioc_readv_disabled_cbk, + if (!fd_ctx_get (fd, this, NULL)) { + /* disable caching for this fd, go ahead with normal readv */ + STACK_WIND (frame, ioc_readv_disabled_cbk, FIRST_CHILD (frame->this), - FIRST_CHILD (frame->this)->fops->readv, fd, size, + FIRST_CHILD (frame->this)->fops->readv, fd, size, offset); - return 0; - } + return 0; + } - local = (ioc_local_t *) GF_CALLOC (1, sizeof (ioc_local_t), - gf_ioc_mt_ioc_local_t); + local = (ioc_local_t *) GF_CALLOC (1, sizeof (ioc_local_t), + gf_ioc_mt_ioc_local_t); if (local == NULL) { gf_log (this->name, GF_LOG_ERROR, "out of memory"); op_errno = ENOMEM; goto out; } - INIT_LIST_HEAD (&local->fill_list); + INIT_LIST_HEAD (&local->fill_list); - frame->local = local; - local->pending_offset = offset; - local->pending_size = size; - local->offset = offset; - local->size = size; - local->inode = ioc_inode; + frame->local = local; + local->pending_offset = offset; + local->pending_size = size; + local->offset = offset; + local->size = size; + local->inode = ioc_inode; - gf_log (this->name, GF_LOG_TRACE, - "NEW REQ (%p) offset = %"PRId64" && size = %"GF_PRI_SIZET"", - frame, offset, size); + gf_log (this->name, GF_LOG_TRACE, + "NEW REQ (%p) offset = %"PRId64" && size = %"GF_PRI_SIZET"", + frame, offset, size); - weight = ioc_inode->weight; + weight = ioc_inode->weight; - ioc_table_lock (ioc_inode->table); - { - list_move_tail (&ioc_inode->inode_lru, - &ioc_inode->table->inode_lru[weight]); - } - ioc_table_unlock (ioc_inode->table); + ioc_table_lock (ioc_inode->table); + { + list_move_tail (&ioc_inode->inode_lru, + &ioc_inode->table->inode_lru[weight]); + } + ioc_table_unlock (ioc_inode->table); - ioc_dispatch_requests (frame, ioc_inode, fd, offset, size); - return 0; + ioc_dispatch_requests (frame, ioc_inode, fd, offset, size); + return 0; out: STACK_UNWIND_STRICT (readv, frame, -1, op_errno, NULL, 0, NULL, NULL); @@ -1192,20 +1206,20 @@ out: */ int32_t ioc_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, struct iatt *prebuf, + int32_t op_ret, int32_t op_errno, struct iatt *prebuf, struct iatt *postbuf) { - ioc_local_t *local = NULL; - uint64_t ioc_inode = 0; + ioc_local_t *local = NULL; + uint64_t ioc_inode = 0; local = frame->local; - inode_ctx_get (local->fd->inode, this, &ioc_inode); + inode_ctx_get (local->fd->inode, this, &ioc_inode); - if (ioc_inode) - ioc_inode_flush ((ioc_inode_t *)(long)ioc_inode); + if (ioc_inode) + ioc_inode_flush ((ioc_inode_t *)(long)ioc_inode); - STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, prebuf, postbuf); - return 0; + STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, prebuf, postbuf); + return 0; } /* @@ -1221,13 +1235,13 @@ ioc_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this, */ int32_t ioc_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, - struct iovec *vector, int32_t count, off_t offset, + struct iovec *vector, int32_t count, off_t offset, struct iobref *iobref) { - ioc_local_t *local = NULL; - uint64_t ioc_inode = 0; + ioc_local_t *local = NULL; + uint64_t ioc_inode = 0; - local = GF_CALLOC (1, sizeof (ioc_local_t), gf_ioc_mt_ioc_local_t); + local = GF_CALLOC (1, sizeof (ioc_local_t), gf_ioc_mt_ioc_local_t); if (local == NULL) { gf_log (this->name, GF_LOG_ERROR, "out of memory"); @@ -1235,19 +1249,19 @@ ioc_writev (call_frame_t *frame, xlator_t *this, fd_t *fd, return 0; } - /* TODO: why is it not fd_ref'ed */ - local->fd = fd; - frame->local = local; + /* TODO: why is it not fd_ref'ed */ + local->fd = fd; + frame->local = local; - inode_ctx_get (fd->inode, this, &ioc_inode); - if (ioc_inode) - ioc_inode_flush ((ioc_inode_t *)(long)ioc_inode); + inode_ctx_get (fd->inode, this, &ioc_inode); + if (ioc_inode) + ioc_inode_flush ((ioc_inode_t *)(long)ioc_inode); - STACK_WIND (frame, ioc_writev_cbk, FIRST_CHILD(this), - FIRST_CHILD(this)->fops->writev, fd, vector, count, offset, + STACK_WIND (frame, ioc_writev_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->writev, fd, vector, count, offset, iobref); - return 0; + return 0; } /* @@ -1267,9 +1281,9 @@ ioc_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this, struct iatt *postbuf) { - STACK_UNWIND_STRICT (truncate, frame, op_ret, op_errno, prebuf, + STACK_UNWIND_STRICT (truncate, frame, op_ret, op_errno, prebuf, postbuf); - return 0; + return 0; } @@ -1286,13 +1300,13 @@ ioc_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this, */ int32_t ioc_ftruncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, struct iatt *prebuf, - struct iatt *postbuf) + int32_t op_ret, int32_t op_errno, struct iatt *prebuf, + struct iatt *postbuf) { - STACK_UNWIND_STRICT (ftruncate, frame, op_ret, op_errno, prebuf, + STACK_UNWIND_STRICT (ftruncate, frame, op_ret, op_errno, prebuf, postbuf); - return 0; + return 0; } @@ -1308,15 +1322,16 @@ ioc_ftruncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t ioc_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset) { - uint64_t ioc_inode = 0; - inode_ctx_get (loc->inode, this, &ioc_inode); + uint64_t ioc_inode = 0; - if (ioc_inode) - ioc_inode_flush ((ioc_inode_t *)(long)ioc_inode); + inode_ctx_get (loc->inode, this, &ioc_inode); - STACK_WIND (frame, ioc_truncate_cbk, FIRST_CHILD(this), - FIRST_CHILD(this)->fops->truncate, loc, offset); - return 0; + if (ioc_inode) + ioc_inode_flush ((ioc_inode_t *)(long)ioc_inode); + + STACK_WIND (frame, ioc_truncate_cbk, FIRST_CHILD(this), + FIRST_CHILD(this)->fops->truncate, loc, offset); + return 0; } /* @@ -1331,66 +1346,67 @@ ioc_truncate (call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset) int32_t ioc_ftruncate (call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset) { - uint64_t ioc_inode = 0; - inode_ctx_get (fd->inode, this, &ioc_inode); + uint64_t ioc_inode = 0; + + inode_ctx_get (fd->inode, this, &ioc_inode); - if (ioc_inode) - ioc_inode_flush ((ioc_inode_t *)(long)ioc_inode); + if (ioc_inode) + ioc_inode_flush ((ioc_inode_t *)(long)ioc_inode); - STACK_WIND (frame, ioc_ftruncate_cbk, FIRST_CHILD(this), + STACK_WIND (frame, ioc_ftruncate_cbk, FIRST_CHILD(this), FIRST_CHILD(this)->fops->ftruncate, fd, offset); - return 0; + return 0; } int32_t ioc_lk_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret, - int32_t op_errno, struct gf_flock *lock) + int32_t op_errno, struct gf_flock *lock) { - STACK_UNWIND_STRICT (lk, frame, op_ret, op_errno, lock); - return 0; + STACK_UNWIND_STRICT (lk, frame, op_ret, op_errno, lock); + return 0; } int32_t ioc_lk (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd, struct gf_flock *lock) { - ioc_inode_t *ioc_inode = NULL; - uint64_t tmp_inode = 0; - - inode_ctx_get (fd->inode, this, &tmp_inode); - ioc_inode = (ioc_inode_t *)(long)tmp_inode; - if (!ioc_inode) { - gf_log (this->name, GF_LOG_DEBUG, - "inode context is NULL: returning EBADFD"); - STACK_UNWIND_STRICT (lk, frame, -1, EBADFD, NULL); - return 0; - } - - ioc_inode_lock (ioc_inode); - { - gettimeofday (&ioc_inode->cache.tv, NULL); - } - ioc_inode_unlock (ioc_inode); - - STACK_WIND (frame, ioc_lk_cbk, FIRST_CHILD (this), - FIRST_CHILD (this)->fops->lk, fd, cmd, lock); - - return 0; + ioc_inode_t *ioc_inode = NULL; + uint64_t tmp_inode = 0; + + inode_ctx_get (fd->inode, this, &tmp_inode); + ioc_inode = (ioc_inode_t *)(long)tmp_inode; + if (!ioc_inode) { + gf_log (this->name, GF_LOG_DEBUG, + "inode context is NULL: returning EBADFD"); + STACK_UNWIND_STRICT (lk, frame, -1, EBADFD, NULL); + return 0; + } + + ioc_inode_lock (ioc_inode); + { + gettimeofday (&ioc_inode->cache.tv, NULL); + } + ioc_inode_unlock (ioc_inode); + + STACK_WIND (frame, ioc_lk_cbk, FIRST_CHILD (this), + FIRST_CHILD (this)->fops->lk, fd, cmd, lock); + + return 0; } int32_t ioc_get_priority_list (const char *opt_str, struct list_head *first) { - int32_t max_pri = 1; - char *tmp_str = NULL; - char *tmp_str1 = NULL; - char *tmp_str2 = NULL; - char *dup_str = NULL; - char *stripe_str = NULL; - char *pattern = NULL; - char *priority = NULL; - char *string = NULL; - struct ioc_priority *curr = NULL, *tmp = NULL; + int32_t max_pri = 1; + char *tmp_str = NULL; + char *tmp_str1 = NULL; + char *tmp_str2 = NULL; + char *dup_str = NULL; + char *stripe_str = NULL; + char *pattern = NULL; + char *priority = NULL; + char *string = NULL; + struct ioc_priority *curr = NULL, *tmp = NULL; string = gf_strdup (opt_str); if (string == NULL) { @@ -1398,65 +1414,65 @@ ioc_get_priority_list (const char *opt_str, struct list_head *first) goto out; } - /* Get the pattern for cache priority. - * "option priority *.jpg:1,abc*:2" etc - */ - /* TODO: inode_lru in table is statically hard-coded to 5, - * should be changed to run-time configuration - */ - stripe_str = strtok_r (string, ",", &tmp_str); - while (stripe_str) { - curr = GF_CALLOC (1, sizeof (struct ioc_priority), + /* Get the pattern for cache priority. + * "option priority *.jpg:1,abc*:2" etc + */ + /* TODO: inode_lru in table is statically hard-coded to 5, + * should be changed to run-time configuration + */ + stripe_str = strtok_r (string, ",", &tmp_str); + while (stripe_str) { + curr = GF_CALLOC (1, sizeof (struct ioc_priority), gf_ioc_mt_ioc_priority); if (curr == NULL) { max_pri = -1; goto out; } - list_add_tail (&curr->list, first); + list_add_tail (&curr->list, first); - dup_str = gf_strdup (stripe_str); + dup_str = gf_strdup (stripe_str); if (dup_str == NULL) { max_pri = -1; goto out; } - pattern = strtok_r (dup_str, ":", &tmp_str1); - if (!pattern) { + pattern = strtok_r (dup_str, ":", &tmp_str1); + if (!pattern) { max_pri = -1; goto out; } - priority = strtok_r (NULL, ":", &tmp_str1); - if (!priority) { + priority = strtok_r (NULL, ":", &tmp_str1); + if (!priority) { max_pri = -1; goto out; } - gf_log ("io-cache", GF_LOG_TRACE, - "ioc priority : pattern %s : priority %s", - pattern, - priority); + gf_log ("io-cache", GF_LOG_TRACE, + "ioc priority : pattern %s : priority %s", + pattern, + priority); - curr->pattern = gf_strdup (pattern); + curr->pattern = gf_strdup (pattern); if (curr->pattern == NULL) { max_pri = -1; goto out; } - curr->priority = strtol (priority, &tmp_str2, 0); - if (tmp_str2 && (*tmp_str2)) { + curr->priority = strtol (priority, &tmp_str2, 0); + if (tmp_str2 && (*tmp_str2)) { max_pri = -1; goto out; } else { - max_pri = max (max_pri, curr->priority); + max_pri = max (max_pri, curr->priority); } GF_FREE (dup_str); dup_str = NULL; - stripe_str = strtok_r (NULL, ",", &tmp_str); - } + stripe_str = strtok_r (NULL, ",", &tmp_str); + } out: if (string != NULL) { GF_FREE (string); @@ -1474,7 +1490,7 @@ out: } } - return max_pri; + return max_pri; } int32_t @@ -1489,7 +1505,7 @@ mem_acct_init (xlator_t *this) if (ret != 0) { gf_log (this->name, GF_LOG_ERROR, "Memory accounting init" - "failed"); + "failed"); return ret; } @@ -1499,9 +1515,9 @@ mem_acct_init (xlator_t *this) int validate_options (xlator_t *this, char **op_errstr) { - int ret = 0; - volume_opt_list_t *vol_opt = NULL; - volume_opt_list_t *tmp; + int ret = 0; + volume_opt_list_t *vol_opt = NULL; + volume_opt_list_t *tmp; if (!this) { gf_log (this->name, GF_LOG_DEBUG, "'this' not a valid ptr"); @@ -1513,8 +1529,8 @@ validate_options (xlator_t *this, char **op_errstr) goto out; vol_opt = list_entry (this->volume_options.next, - volume_opt_list_t, list); - list_for_each_entry_safe (vol_opt, tmp, &this->volume_options, list) { + volume_opt_list_t, list); + list_for_each_entry_safe (vol_opt, tmp, &this->volume_options, list) { ret = validate_xlator_volume_options_attacherr (this, vol_opt->given_opt, op_errstr); @@ -1528,169 +1544,169 @@ out: int reconfigure (xlator_t *this, dict_t *options) { - ioc_table_t *table = NULL; - int32_t cache_timeout; - int64_t min_file_size = 0; - int64_t max_file_size = 0; - char *tmp = NULL; - uint64_t cache_size; - char *cache_size_string = NULL; - int ret = 0; + ioc_table_t *table = NULL; + int32_t cache_timeout = 0; + int64_t min_file_size = 0; + int64_t max_file_size = 0; + char *tmp = NULL; + uint64_t cache_size = 0; + char *cache_size_string = NULL; + int ret = 0; if (!this || !this->private) goto out; - table = this->private; - - ioc_table_lock (table); - { - if (dict_get (options, "cache-timeout")) { - cache_timeout = - data_to_uint32 (dict_get (options, - "cache-timeout")); - if (cache_timeout < 0){ - gf_log (this->name, GF_LOG_WARNING, - "cache-timeout %d seconds invalid," - " has to be >=0", cache_timeout); - goto out; - } - - - if (cache_timeout > 60){ - gf_log (this->name, GF_LOG_WARNING, - "cache-timeout %d seconds invalid," - " has to be <=60", cache_timeout); - goto out; - } - - table->cache_timeout = cache_timeout; - - gf_log (this->name, GF_LOG_DEBUG, - "Reconfiguring %d seconds to" - " revalidate cache", table->cache_timeout); - } + table = this->private; + + ioc_table_lock (table); + { + if (dict_get (options, "cache-timeout")) { + cache_timeout = + data_to_uint32 (dict_get (options, + "cache-timeout")); + if (cache_timeout < 0){ + gf_log (this->name, GF_LOG_WARNING, + "cache-timeout %d seconds invalid," + " has to be >=0", cache_timeout); + goto out; + } + + + if (cache_timeout > 60){ + gf_log (this->name, GF_LOG_WARNING, + "cache-timeout %d seconds invalid," + " has to be <=60", cache_timeout); + goto out; + } + + table->cache_timeout = cache_timeout; + + gf_log (this->name, GF_LOG_DEBUG, + "Reconfiguring %d seconds to" + " revalidate cache", table->cache_timeout); + } else table->cache_timeout = 1; - if (dict_get (options, "cache-size")) - cache_size_string = data_to_str (dict_get (options, - "cache-size")); - if (cache_size_string) { - if (gf_string2bytesize (cache_size_string, - &cache_size) != 0) { - gf_log ("io-cache", GF_LOG_ERROR, - "invalid number format \"%s\" of " - "\"option cache-size\" Defaulting" - "to old value", cache_size_string); - goto out; - } + if (dict_get (options, "cache-size")) + cache_size_string = data_to_str (dict_get (options, + "cache-size")); + if (cache_size_string) { + if (gf_string2bytesize (cache_size_string, + &cache_size) != 0) { + gf_log ("io-cache", GF_LOG_ERROR, + "invalid number format \"%s\" of " + "\"option cache-size\" Defaulting" + "to old value", cache_size_string); + goto out; + } - if (cache_size < (4 * GF_UNIT_MB)) { - gf_log(this->name, GF_LOG_ERROR, + if (cache_size < (4 * GF_UNIT_MB)) { + gf_log(this->name, GF_LOG_ERROR, "Reconfiguration" "'option cache-size %s' failed , " "Max value can be 4MiB, Defaulting to " "old value (%"PRIu64")", cache_size_string, table->cache_size); - goto out; - } + goto out; + } - if (cache_size > (6 * GF_UNIT_GB)) { - gf_log (this->name, GF_LOG_ERROR, + if (cache_size > (6 * GF_UNIT_GB)) { + gf_log (this->name, GF_LOG_ERROR, "Reconfiguration" "'option cache-size %s' failed , " "Max value can be 6GiB, Defaulting to " "old value (%"PRIu64")", cache_size_string, table->cache_size); - goto out; - } + goto out; + } - gf_log (this->name, GF_LOG_DEBUG, "Reconfiguring " - " cache-size %"PRIu64"", cache_size); - table->cache_size = cache_size; - } + gf_log (this->name, GF_LOG_DEBUG, "Reconfiguring " + " cache-size %"PRIu64"", cache_size); + table->cache_size = cache_size; + } else table->cache_size = IOC_CACHE_SIZE; - if (dict_get (options, "priority")) { - char *option_list = data_to_str (dict_get (options, - "priority")); - gf_log (this->name, GF_LOG_TRACE, - "option path %s", option_list); - /* parse the list of pattern:priority */ - table->max_pri = ioc_get_priority_list (option_list, - &table->priority_list); - - if (table->max_pri == -1) { - ret = -1; - goto out; - } - table->max_pri ++; - } - - - - min_file_size = table->min_file_size; - tmp = data_to_str (dict_get (options, "min-file-size")); - if (tmp != NULL) { - if (gf_string2bytesize (tmp, - (uint64_t *)&min_file_size) - != 0) { - gf_log ("io-cache", GF_LOG_ERROR, - "invalid number format \"%s\" of " - "\"option min-file-size\"", tmp); - ret = -1; - goto out; - } - - gf_log (this->name, GF_LOG_DEBUG, - "Reconfiguring min-file-size %"PRIu64"", - table->min_file_size); - } - - max_file_size = table->max_file_size; - tmp = data_to_str (dict_get (options, "max-file-size")); - if (tmp != NULL) { - if (gf_string2bytesize (tmp, - (uint64_t *)&max_file_size) - != 0) { - gf_log ("io-cache", GF_LOG_ERROR, - "invalid number format \"%s\" of " - "\"option max-file-size\"", tmp); - ret = -1; - goto out; - } - - - gf_log (this->name, GF_LOG_DEBUG, - "Reconfiguring max-file-size %"PRIu64"", - table->max_file_size); - } + if (dict_get (options, "priority")) { + char *option_list = data_to_str (dict_get (options, + "priority")); + gf_log (this->name, GF_LOG_TRACE, + "option path %s", option_list); + /* parse the list of pattern:priority */ + table->max_pri = ioc_get_priority_list (option_list, + &table->priority_list); + + if (table->max_pri == -1) { + ret = -1; + goto out; + } + table->max_pri ++; + } + + + + min_file_size = table->min_file_size; + tmp = data_to_str (dict_get (options, "min-file-size")); + if (tmp != NULL) { + if (gf_string2bytesize (tmp, + (uint64_t *)&min_file_size) + != 0) { + gf_log ("io-cache", GF_LOG_ERROR, + "invalid number format \"%s\" of " + "\"option min-file-size\"", tmp); + ret = -1; + goto out; + } + + gf_log (this->name, GF_LOG_DEBUG, + "Reconfiguring min-file-size %"PRIu64"", + table->min_file_size); + } + + max_file_size = table->max_file_size; + tmp = data_to_str (dict_get (options, "max-file-size")); + if (tmp != NULL) { + if (gf_string2bytesize (tmp, + (uint64_t *)&max_file_size) + != 0) { + gf_log ("io-cache", GF_LOG_ERROR, + "invalid number format \"%s\" of " + "\"option max-file-size\"", tmp); + ret = -1; + goto out; + } + + + gf_log (this->name, GF_LOG_DEBUG, + "Reconfiguring max-file-size %"PRIu64"", + table->max_file_size); + } if ((max_file_size >= 0) && (min_file_size > max_file_size)) { gf_log ("io-cache", GF_LOG_ERROR, "minimum size (%" PRIu64") of a file that can be cached is " "greater than maximum size (%"PRIu64"). " - "Hence Defaulting to old value", + "Hence Defaulting to old value", table->min_file_size, table->max_file_size); - goto out; - } + goto out; + } - table->min_file_size = min_file_size; - table->max_file_size = max_file_size; + table->min_file_size = min_file_size; + table->max_file_size = max_file_size; if (!data_to_str (dict_get (options, "min-file-size"))) - table->min_file_size = 0; + table->min_file_size = 0; if (data_to_str (dict_get (options, "max-file-size"))) - table->max_file_size = 0; - } + table->max_file_size = 0; + } - ioc_table_unlock (table); + ioc_table_unlock (table); out: - return ret; + return ret; } @@ -1702,80 +1718,82 @@ out: int32_t init (xlator_t *this) { - ioc_table_t *table = NULL; - dict_t *options = this->options; - uint32_t index = 0; - char *cache_size_string = NULL, *tmp = NULL; - int32_t ret = -1; - glusterfs_ctx_t *ctx = NULL; - data_t *data = 0; - - if (!this->children || this->children->next) { - gf_log (this->name, GF_LOG_ERROR, - "FATAL: io-cache not configured with exactly " - "one child"); + ioc_table_t *table = NULL; + dict_t *options = NULL; + uint32_t index = 0; + char *cache_size_string = NULL, *tmp = NULL; + int32_t ret = -1; + glusterfs_ctx_t *ctx = NULL; + data_t *data = 0; + + options = this->options; + + if (!this->children || this->children->next) { + gf_log (this->name, GF_LOG_ERROR, + "FATAL: io-cache not configured with exactly " + "one child"); goto out; - } + } - if (!this->parents) { - gf_log (this->name, GF_LOG_WARNING, - "dangling volume. check volfile "); - } + if (!this->parents) { + gf_log (this->name, GF_LOG_WARNING, + "dangling volume. check volfile "); + } - table = (void *) GF_CALLOC (1, sizeof (*table), gf_ioc_mt_ioc_table_t); + table = (void *) GF_CALLOC (1, sizeof (*table), gf_ioc_mt_ioc_table_t); if (table == NULL) { gf_log (this->name, GF_LOG_ERROR, "out of memory"); goto out; } - table->xl = this; - table->page_size = this->ctx->page_size; - table->cache_size = IOC_CACHE_SIZE; + table->xl = this; + table->page_size = this->ctx->page_size; + table->cache_size = IOC_CACHE_SIZE; data = dict_get (options, "cache-size"); if (data) - cache_size_string = data_to_str (data); - - if (cache_size_string) { - if (gf_string2bytesize (cache_size_string, - &table->cache_size) != 0) { - gf_log ("io-cache", GF_LOG_ERROR, - "invalid number format \"%s\" of " - "\"option cache-size\"", - cache_size_string); + cache_size_string = data_to_str (data); + + if (cache_size_string) { + if (gf_string2bytesize (cache_size_string, + &table->cache_size) != 0) { + gf_log ("io-cache", GF_LOG_ERROR, + "invalid number format \"%s\" of " + "\"option cache-size\"", + cache_size_string); goto out; - } + } - gf_log (this->name, GF_LOG_TRACE, - "using cache-size %"PRIu64"", table->cache_size); - } + gf_log (this->name, GF_LOG_TRACE, + "using cache-size %"PRIu64"", table->cache_size); + } - table->cache_timeout = 1; + table->cache_timeout = 1; - data = dict_get (options, "cache-timeout"); + data = dict_get (options, "cache-timeout"); if (data) { - table->cache_timeout = data_to_uint32 (data); - gf_log (this->name, GF_LOG_TRACE, - "Using %d seconds to revalidate cache", - table->cache_timeout); - } - - INIT_LIST_HEAD (&table->priority_list); - table->max_pri = 1; + table->cache_timeout = data_to_uint32 (data); + gf_log (this->name, GF_LOG_TRACE, + "Using %d seconds to revalidate cache", + table->cache_timeout); + } + + INIT_LIST_HEAD (&table->priority_list); + table->max_pri = 1; data = dict_get (options, "priority"); if (data) { - char *option_list = data_to_str (data); - gf_log (this->name, GF_LOG_TRACE, - "option path %s", option_list); - /* parse the list of pattern:priority */ - table->max_pri = ioc_get_priority_list (option_list, - &table->priority_list); - - if (table->max_pri == -1) { + char *option_list = data_to_str (data); + gf_log (this->name, GF_LOG_TRACE, + "option path %s", option_list); + /* parse the list of pattern:priority */ + table->max_pri = ioc_get_priority_list (option_list, + &table->priority_list); + + if (table->max_pri == -1) { goto out; } - } - table->max_pri ++; + } + table->max_pri ++; table->min_file_size = 0; @@ -1784,16 +1802,17 @@ init (xlator_t *this) tmp = data_to_str (data); if (tmp != NULL) { - if (gf_string2bytesize (tmp, - (uint64_t *)&table->min_file_size) != 0) { - gf_log ("io-cache", GF_LOG_ERROR, - "invalid number format \"%s\" of " - "\"option min-file-size\"", tmp); + if (gf_string2bytesize (tmp, + (uint64_t *)&table->min_file_size) + != 0) { + gf_log ("io-cache", GF_LOG_ERROR, + "invalid number format \"%s\" of " + "\"option min-file-size\"", tmp); goto out; - } + } - gf_log (this->name, GF_LOG_TRACE, - "using min-file-size %"PRIu64"", table->min_file_size); + gf_log (this->name, GF_LOG_TRACE, + "using min-file-size %"PRIu64"", table->min_file_size); } tmp = NULL; @@ -1804,7 +1823,8 @@ init (xlator_t *this) if (tmp != NULL) { if (gf_string2bytesize (tmp, - (uint64_t *)&table->max_file_size) != 0) { + (uint64_t *)&table->max_file_size) + != 0) { gf_log ("io-cache", GF_LOG_ERROR, "invalid number format \"%s\" of " "\"option max-file-size\"", tmp); @@ -1814,29 +1834,29 @@ init (xlator_t *this) gf_log (this->name, GF_LOG_TRACE, "using max-file-size %"PRIu64"", table->max_file_size); } - INIT_LIST_HEAD (&table->inodes); + INIT_LIST_HEAD (&table->inodes); if ((table->max_file_size >= 0) && (table->min_file_size > table->max_file_size)) { - gf_log ("io-cache", GF_LOG_ERROR, "minimum size (%" - PRIu64") of a file that can be cached is " - "greater than maximum size (%"PRIu64")", - table->min_file_size, table->max_file_size); - goto out; + gf_log ("io-cache", GF_LOG_ERROR, "minimum size (%" + PRIu64") of a file that can be cached is " + "greater than maximum size (%"PRIu64")", + table->min_file_size, table->max_file_size); + goto out; } - table->inode_lru = GF_CALLOC (table->max_pri, + table->inode_lru = GF_CALLOC (table->max_pri, sizeof (struct list_head), gf_ioc_mt_list_head); if (table->inode_lru == NULL) { goto out; } - for (index = 0; index < (table->max_pri); index++) - INIT_LIST_HEAD (&table->inode_lru[index]); + for (index = 0; index < (table->max_pri); index++) + INIT_LIST_HEAD (&table->inode_lru[index]); - pthread_mutex_init (&table->table_lock, NULL); - this->private = table; + pthread_mutex_init (&table->table_lock, NULL); + this->private = table; ret = 0; ctx = this->ctx; @@ -1850,15 +1870,15 @@ out: } } - return ret; + return ret; } int ioc_priv_dump (xlator_t *this) { - ioc_table_t *priv = NULL; - char key_prefix[GF_DUMP_MAX_BUF_LEN]; - char key[GF_DUMP_MAX_BUF_LEN]; + ioc_table_t *priv = NULL; + char key_prefix[GF_DUMP_MAX_BUF_LEN] = {0, }; + char key[GF_DUMP_MAX_BUF_LEN] = {0, }; if (!this || !this->private) goto out; @@ -1890,7 +1910,7 @@ out: void fini (xlator_t *this) { - ioc_table_t *table = NULL; + ioc_table_t *table = NULL; table = this->private; @@ -1902,22 +1922,22 @@ fini (xlator_t *this) table->mem_pool = NULL; } - pthread_mutex_destroy (&table->table_lock); - GF_FREE (table); + pthread_mutex_destroy (&table->table_lock); + GF_FREE (table); - this->private = NULL; - return; + this->private = NULL; + return; } struct xlator_fops fops = { - .open = ioc_open, - .create = ioc_create, - .readv = ioc_readv, - .writev = ioc_writev, - .truncate = ioc_truncate, - .ftruncate = ioc_ftruncate, - .lookup = ioc_lookup, - .lk = ioc_lk, + .open = ioc_open, + .create = ioc_create, + .readv = ioc_readv, + .writev = ioc_writev, + .truncate = ioc_truncate, + .ftruncate = ioc_ftruncate, + .lookup = ioc_lookup, + .lk = ioc_lk, .setattr = ioc_setattr, .mknod = ioc_mknod }; @@ -1928,29 +1948,29 @@ struct xlator_dumpops dumpops = { }; struct xlator_cbks cbks = { - .forget = ioc_forget, - .release = ioc_release + .forget = ioc_forget, + .release = ioc_release }; struct volume_options options[] = { - { .key = {"priority"}, - .type = GF_OPTION_TYPE_ANY - }, - { .key = {"cache-timeout", "force-revalidate-timeout"}, - .type = GF_OPTION_TYPE_INT, - .min = 0, - .max = 60 - }, - { .key = {"cache-size"}, - .type = GF_OPTION_TYPE_SIZET, - .min = 4 * GF_UNIT_MB, - .max = 6 * GF_UNIT_GB - }, + { .key = {"priority"}, + .type = GF_OPTION_TYPE_ANY + }, + { .key = {"cache-timeout", "force-revalidate-timeout"}, + .type = GF_OPTION_TYPE_INT, + .min = 0, + .max = 60 + }, + { .key = {"cache-size"}, + .type = GF_OPTION_TYPE_SIZET, + .min = 4 * GF_UNIT_MB, + .max = 6 * GF_UNIT_GB + }, { .key = {"min-file-size"}, .type = GF_OPTION_TYPE_SIZET, }, { .key = {"max-file-size"}, .type = GF_OPTION_TYPE_SIZET, }, - { .key = {NULL} }, + { .key = {NULL} }, }; diff --git a/xlators/performance/io-cache/src/io-cache.h b/xlators/performance/io-cache/src/io-cache.h index 5d05264404a..7fefc5f6169 100644 --- a/xlators/performance/io-cache/src/io-cache.h +++ b/xlators/performance/io-cache/src/io-cache.h @@ -198,29 +198,26 @@ ioc_readv_disabled_cbk (call_frame_t *frame, void *cookie, xlator_t *this, struct iobref *iobref); ioc_page_t * -ioc_page_get (ioc_inode_t *ioc_inode, off_t offset); +__ioc_page_get (ioc_inode_t *ioc_inode, off_t offset); ioc_page_t * -ioc_page_create (ioc_inode_t *ioc_inode, off_t offset); +__ioc_page_create (ioc_inode_t *ioc_inode, off_t offset); void ioc_page_fault (ioc_inode_t *ioc_inode, call_frame_t *frame, fd_t *fd, off_t offset); void -ioc_wait_on_page (ioc_page_t *page, call_frame_t *frame, off_t offset, +__ioc_wait_on_page (ioc_page_t *page, call_frame_t *frame, off_t offset, size_t size); ioc_waitq_t * -ioc_page_wakeup (ioc_page_t *page); +__ioc_page_wakeup (ioc_page_t *page); void ioc_page_flush (ioc_page_t *page); ioc_waitq_t * -ioc_page_error (ioc_page_t *page, int32_t op_ret, int32_t op_errno); - -void -ioc_page_purge (ioc_page_t *page); +__ioc_page_error (ioc_page_t *page, int32_t op_ret, int32_t op_errno); void ioc_frame_return (call_frame_t *frame); @@ -318,7 +315,7 @@ ioc_inode_t * ioc_inode_update (ioc_table_t *table, inode_t *inode, uint32_t weight); int64_t -ioc_page_destroy (ioc_page_t *page); +__ioc_page_destroy (ioc_page_t *page); int64_t __ioc_inode_flush (ioc_inode_t *ioc_inode); diff --git a/xlators/performance/io-cache/src/ioc-inode.c b/xlators/performance/io-cache/src/ioc-inode.c index e268cac2621..a305d0316f3 100644 --- a/xlators/performance/io-cache/src/ioc-inode.c +++ b/xlators/performance/io-cache/src/ioc-inode.c @@ -36,6 +36,7 @@ void * str_to_ptr (char *string) { void *ptr = NULL; + GF_VALIDATE_OR_GOTO ("io-cache", string, out); ptr = (void *)strtoul (string, NULL, 16); @@ -122,9 +123,10 @@ ioc_inode_wakeup (call_frame_t *frame, ioc_inode_t *ioc_inode, ioc_inode_lock (ioc_inode); { page_waitq = - ioc_page_wakeup (waiter_page); + __ioc_page_wakeup (waiter_page); } ioc_inode_unlock (ioc_inode); + if (page_waitq) ioc_waitq_return (page_waitq); } else { @@ -132,17 +134,19 @@ ioc_inode_wakeup (call_frame_t *frame, ioc_inode_t *ioc_inode, * page->ready = 0, to avoid double faults */ ioc_inode_lock (ioc_inode); - - if (waiter_page->ready) { - waiter_page->ready = 0; - need_fault = 1; - } else { - gf_log (frame->this->name, GF_LOG_TRACE, - "validate frame(%p) is waiting" - "for in-transit page = %p", - frame, waiter_page); + { + if (waiter_page->ready) { + waiter_page->ready = 0; + need_fault = 1; + } else { + gf_log (frame->this->name, + GF_LOG_TRACE, + "validate frame(%p) is " + "waiting for " + "in-transit page = %p", + frame, waiter_page); + } } - ioc_inode_unlock (ioc_inode); if (need_fault) { diff --git a/xlators/performance/io-cache/src/page.c b/xlators/performance/io-cache/src/page.c index 0721f5f04a3..895775d9366 100644 --- a/xlators/performance/io-cache/src/page.c +++ b/xlators/performance/io-cache/src/page.c @@ -46,7 +46,7 @@ out: ioc_page_t * -ioc_page_get (ioc_inode_t *ioc_inode, off_t offset) +__ioc_page_get (ioc_inode_t *ioc_inode, off_t offset) { ioc_page_t *page = NULL; ioc_table_t *table = NULL; @@ -72,14 +72,34 @@ out: } +ioc_page_t * +ioc_page_get (ioc_inode_t *ioc_inode, off_t offset) +{ + ioc_page_t *page = NULL; + + if (ioc_inode == NULL) { + goto out; + } + + ioc_inode_lock (ioc_inode); + { + page = __ioc_page_get (ioc_inode, offset); + } + ioc_inode_unlock (ioc_inode); + +out: + return page; +} + + /* - * ioc_page_destroy - + * __ioc_page_destroy - * * @page: * */ int64_t -ioc_page_destroy (ioc_page_t *page) +__ioc_page_destroy (ioc_page_t *page) { int64_t page_size = 0; @@ -119,6 +139,63 @@ out: return page_size; } + +int64_t +ioc_page_destroy (ioc_page_t *page) +{ + int64_t ret = 0; + + if (page == NULL) { + goto out; + } + + ioc_inode_lock (page->inode); + { + ret = __ioc_page_destroy (page); + } + ioc_inode_unlock (page->inode); + +out: + return ret; +} + +int32_t +__ioc_inode_prune (ioc_inode_t *curr, uint64_t *size_pruned, + uint64_t size_to_prune, uint32_t index) +{ + ioc_page_t *page = NULL, *next = NULL; + int32_t ret = 0; + ioc_table_t *table = NULL; + + if (curr == NULL) { + goto out; + } + + table = curr->table; + + list_for_each_entry_safe (page, next, &curr->cache.page_lru, page_lru) { + *size_pruned += page->size; + ret = __ioc_page_destroy (page); + + if (ret != -1) + table->cache_used -= ret; + + gf_log (table->xl->name, GF_LOG_TRACE, + "index = %d && table->cache_used = %"PRIu64" && table->" + "cache_size = %"PRIu64, index, table->cache_used, + table->cache_size); + + if ((*size_pruned) >= size_to_prune) + break; + } + + if (ioc_empty (&curr->cache)) { + list_del_init (&curr->inode_lru); + } + +out: + return 0; +} /* * ioc_prune - prune the cache. we have a limit to the number of pages we * can have in-memory. @@ -130,8 +207,6 @@ int32_t ioc_prune (ioc_table_t *table) { ioc_inode_t *curr = NULL, *next_ioc_inode = NULL; - ioc_page_t *page = NULL, *next = NULL; - int32_t ret = -1; int32_t index = 0; uint64_t size_to_prune = 0; uint64_t size_pruned = 0; @@ -149,37 +224,11 @@ ioc_prune (ioc_table_t *table) /* prune page-by-page for this inode, till * we reach the equilibrium */ ioc_inode_lock (curr); - /* { */ - - list_for_each_entry_safe (page, next, - &curr->cache.page_lru, - page_lru) { - /* done with all pages, and not - * reached equilibrium yet?? - * continue with next inode in - * lru_list */ - size_pruned += page->size; - ret = ioc_page_destroy (page); - - if (ret != -1) - table->cache_used -= ret; - - gf_log (table->xl->name, GF_LOG_TRACE, - "index = %d && table->cache_" - "used = %"PRIu64" && table->" - "cache_size = %"PRIu64, - index, table->cache_used, - table->cache_size); - - if (size_pruned >= size_to_prune) - break; - } /* list_for_each_entry_safe(page...) */ - - if (ioc_empty (&curr->cache)) { - list_del_init (&curr->inode_lru); + { + __ioc_inode_prune (curr, &size_pruned, + size_to_prune, + index); } - - /* } */ ioc_inode_unlock (curr); if (size_pruned >= size_to_prune) @@ -198,14 +247,14 @@ out: } /* - * ioc_page_create - create a new page. + * __ioc_page_create - create a new page. * * @ioc_inode: * @offset: * */ ioc_page_t * -ioc_page_create (ioc_inode_t *ioc_inode, off_t offset) +__ioc_page_create (ioc_inode_t *ioc_inode, off_t offset) { ioc_table_t *table = NULL; ioc_page_t *page = NULL; @@ -258,8 +307,8 @@ out: * */ void -ioc_wait_on_page (ioc_page_t *page, call_frame_t *frame, off_t offset, - size_t size) +__ioc_wait_on_page (ioc_page_t *page, call_frame_t *frame, off_t offset, + size_t size) { ioc_waitq_t *waitq = NULL; ioc_local_t *local = NULL; @@ -414,14 +463,14 @@ ioc_fault_cbk (call_frame_t *frame, void *cookie, xlator_t *this, if (op_ret < 0) { /* error, readv returned -1 */ - page = ioc_page_get (ioc_inode, offset); + page = __ioc_page_get (ioc_inode, offset); if (page) - waitq = ioc_page_error (page, op_ret, - op_errno); + waitq = __ioc_page_error (page, op_ret, + op_errno); } else { gf_log (ioc_inode->table->xl->name, GF_LOG_TRACE, "op_ret = %d", op_ret); - page = ioc_page_get (ioc_inode, offset); + page = __ioc_page_get (ioc_inode, offset); if (!page) { /* page was flushed */ /* some serious bug ? */ @@ -439,11 +488,12 @@ ioc_fault_cbk (call_frame_t *frame, void *cookie, xlator_t *this, /* keep a copy of the page for our cache */ page->vector = iov_dup (vector, count); if (page->vector == NULL) { - page = ioc_page_get (ioc_inode, offset); + page = __ioc_page_get (ioc_inode, + offset); if (page != NULL) - waitq = ioc_page_error (page, - -1, - ENOMEM); + waitq = __ioc_page_error (page, + -1, + ENOMEM); goto unlock; } @@ -474,7 +524,7 @@ ioc_fault_cbk (call_frame_t *frame, void *cookie, xlator_t *this, /* wake up all the frames waiting on * this page, including * the frame which triggered fault */ - waitq = ioc_page_wakeup (page); + waitq = __ioc_page_wakeup (page); } /* if(page->waitq) */ } /* if(!page)...else */ } /* if(op_ret < 0)...else */ @@ -584,19 +634,24 @@ ioc_page_fault (ioc_inode_t *ioc_inode, call_frame_t *frame, fd_t *fd, return; err: - page = ioc_page_get (ioc_inode, offset); - if (page != NULL) { - waitq = ioc_page_error (page, op_ret, op_errno); - if (waitq != NULL) { - ioc_waitq_return (waitq); + ioc_inode_lock (ioc_inode); + { + page = __ioc_page_get (ioc_inode, offset); + if (page != NULL) { + waitq = __ioc_page_error (page, op_ret, op_errno); } } + ioc_inode_unlock (ioc_inode); + + if (waitq != NULL) { + ioc_waitq_return (waitq); + } } int32_t -ioc_frame_fill (ioc_page_t *page, call_frame_t *frame, off_t offset, - size_t size) +__ioc_frame_fill (ioc_page_t *page, call_frame_t *frame, off_t offset, + size_t size) { ioc_local_t *local = NULL; ioc_fill_t *fill = NULL; @@ -720,6 +775,7 @@ ioc_frame_fill (ioc_page_t *page, call_frame_t *frame, off_t offset, } } } + local->op_ret += copy_size; } @@ -867,7 +923,7 @@ ioc_frame_return (call_frame_t *frame) * to be called only when a frame is waiting on an in-transit page */ ioc_waitq_t * -ioc_page_wakeup (ioc_page_t *page) +__ioc_page_wakeup (ioc_page_t *page) { ioc_waitq_t *waitq = NULL, *trav = NULL; call_frame_t *frame = NULL; @@ -885,8 +941,8 @@ ioc_page_wakeup (ioc_page_t *page) for (trav = waitq; trav; trav = trav->next) { frame = trav->data; - ret = ioc_frame_fill (page, frame, trav->pending_offset, - trav->pending_size); + ret = __ioc_frame_fill (page, frame, trav->pending_offset, + trav->pending_size); if (ret == -1) { break; } @@ -897,6 +953,7 @@ out: } + /* * ioc_page_error - * @page: @@ -905,7 +962,7 @@ out: * */ ioc_waitq_t * -ioc_page_error (ioc_page_t *page, int32_t op_ret, int32_t op_errno) +__ioc_page_error (ioc_page_t *page, int32_t op_ret, int32_t op_errno) { ioc_waitq_t *waitq = NULL, *trav = NULL; call_frame_t *frame = NULL; @@ -937,7 +994,7 @@ ioc_page_error (ioc_page_t *page, int32_t op_ret, int32_t op_errno) } table = page->inode->table; - ret = ioc_page_destroy (page); + ret = __ioc_page_destroy (page); if (ret != -1) { table->cache_used -= ret; @@ -946,3 +1003,29 @@ ioc_page_error (ioc_page_t *page, int32_t op_ret, int32_t op_errno) out: return waitq; } + +/* + * ioc_page_error - + * @page: + * @op_ret: + * @op_errno: + * + */ +ioc_waitq_t * +ioc_page_error (ioc_page_t *page, int32_t op_ret, int32_t op_errno) +{ + ioc_waitq_t *waitq = NULL; + + if (page == NULL) { + goto out; + } + + ioc_inode_lock (page->inode); + { + waitq = __ioc_page_error (page, op_ret, op_errno); + } + ioc_inode_unlock (page->inode); + +out: + return waitq; +} -- cgit