diff options
Diffstat (limited to 'drivers/md/dm-bufio.c')
| -rw-r--r-- | drivers/md/dm-bufio.c | 2199 |
1 files changed, 1483 insertions, 716 deletions
diff --git a/drivers/md/dm-bufio.c b/drivers/md/dm-bufio.c index 50f3e673729c..e6d28be11c5c 100644 --- a/drivers/md/dm-bufio.c +++ b/drivers/md/dm-bufio.c @@ -1,3 +1,4 @@ +// SPDX-License-Identifier: GPL-2.0-only /* * Copyright (C) 2009-2011 Red Hat, Inc. * @@ -18,6 +19,9 @@ #include <linux/module.h> #include <linux/rbtree.h> #include <linux/stacktrace.h> +#include <linux/jump_label.h> + +#include "dm.h" #define DM_MSG_PREFIX "bufio" @@ -37,16 +41,6 @@ #define DM_BUFIO_LOW_WATERMARK_RATIO 16 /* - * Check buffer ages in this interval (seconds) - */ -#define DM_BUFIO_WORK_TIMER_SECS 30 - -/* - * Free buffers when they are older than this (seconds) - */ -#define DM_BUFIO_DEFAULT_AGE_SECS 300 - -/* * The nr of bytes of cached data to keep around. */ #define DM_BUFIO_DEFAULT_RETAIN_BYTES (256 * 1024) @@ -64,56 +58,244 @@ #define LIST_DIRTY 1 #define LIST_SIZE 2 +#define SCAN_RESCHED_CYCLE 16 + +/*--------------------------------------------------------------*/ + /* - * Linking of buffers: - * All buffers are linked to buffer_tree with their node field. - * - * Clean buffers that are not being written (B_WRITING not set) - * are linked to lru[LIST_CLEAN] with their lru_list field. - * - * Dirty and clean buffers that are being written are linked to - * lru[LIST_DIRTY] with their lru_list field. When the write - * finishes, the buffer cannot be relinked immediately (because we - * are in an interrupt context and relinking requires process - * context), so some clean-not-writing buffers can be held on - * dirty_lru too. They are later added to lru in the process - * context. + * Rather than use an LRU list, we use a clock algorithm where entries + * are held in a circular list. When an entry is 'hit' a reference bit + * is set. The least recently used entry is approximated by running a + * cursor around the list selecting unreferenced entries. Referenced + * entries have their reference bit cleared as the cursor passes them. */ -struct dm_bufio_client { - struct mutex lock; +struct lru_entry { + struct list_head list; + atomic_t referenced; +}; - struct list_head lru[LIST_SIZE]; - unsigned long n_buffers[LIST_SIZE]; +struct lru_iter { + struct lru *lru; + struct list_head list; + struct lru_entry *stop; + struct lru_entry *e; +}; - struct block_device *bdev; - unsigned block_size; - s8 sectors_per_block_bits; - void (*alloc_callback)(struct dm_buffer *); - void (*write_callback)(struct dm_buffer *); +struct lru { + struct list_head *cursor; + unsigned long count; - struct kmem_cache *slab_buffer; - struct kmem_cache *slab_cache; - struct dm_io_client *dm_io; + struct list_head iterators; +}; - struct list_head reserved_buffers; - unsigned need_reserved_buffers; +/*--------------*/ - unsigned minimum_buffers; +static void lru_init(struct lru *lru) +{ + lru->cursor = NULL; + lru->count = 0; + INIT_LIST_HEAD(&lru->iterators); +} - struct rb_root buffer_tree; - wait_queue_head_t free_buffer_wait; +static void lru_destroy(struct lru *lru) +{ + WARN_ON_ONCE(lru->cursor); + WARN_ON_ONCE(!list_empty(&lru->iterators)); +} - sector_t start; +/* + * Insert a new entry into the lru. + */ +static void lru_insert(struct lru *lru, struct lru_entry *le) +{ + /* + * Don't be tempted to set to 1, makes the lru aspect + * perform poorly. + */ + atomic_set(&le->referenced, 0); - int async_write_error; + if (lru->cursor) { + list_add_tail(&le->list, lru->cursor); + } else { + INIT_LIST_HEAD(&le->list); + lru->cursor = &le->list; + } + lru->count++; +} - struct list_head client_list; +/*--------------*/ - struct shrinker shrinker; - struct work_struct shrink_work; - atomic_long_t need_shrink; +/* + * Convert a list_head pointer to an lru_entry pointer. + */ +static inline struct lru_entry *to_le(struct list_head *l) +{ + return container_of(l, struct lru_entry, list); +} + +/* + * Initialize an lru_iter and add it to the list of cursors in the lru. + */ +static void lru_iter_begin(struct lru *lru, struct lru_iter *it) +{ + it->lru = lru; + it->stop = lru->cursor ? to_le(lru->cursor->prev) : NULL; + it->e = lru->cursor ? to_le(lru->cursor) : NULL; + list_add(&it->list, &lru->iterators); +} + +/* + * Remove an lru_iter from the list of cursors in the lru. + */ +static inline void lru_iter_end(struct lru_iter *it) +{ + list_del(&it->list); +} + +/* Predicate function type to be used with lru_iter_next */ +typedef bool (*iter_predicate)(struct lru_entry *le, void *context); + +/* + * Advance the cursor to the next entry that passes the + * predicate, and return that entry. Returns NULL if the + * iteration is complete. + */ +static struct lru_entry *lru_iter_next(struct lru_iter *it, + iter_predicate pred, void *context) +{ + struct lru_entry *e; + + while (it->e) { + e = it->e; + + /* advance the cursor */ + if (it->e == it->stop) + it->e = NULL; + else + it->e = to_le(it->e->list.next); + + if (pred(e, context)) + return e; + } + + return NULL; +} + +/* + * Invalidate a specific lru_entry and update all cursors in + * the lru accordingly. + */ +static void lru_iter_invalidate(struct lru *lru, struct lru_entry *e) +{ + struct lru_iter *it; + + list_for_each_entry(it, &lru->iterators, list) { + /* Move c->e forwards if necc. */ + if (it->e == e) { + it->e = to_le(it->e->list.next); + if (it->e == e) + it->e = NULL; + } + + /* Move it->stop backwards if necc. */ + if (it->stop == e) { + it->stop = to_le(it->stop->list.prev); + if (it->stop == e) + it->stop = NULL; + } + } +} + +/*--------------*/ + +/* + * Remove a specific entry from the lru. + */ +static void lru_remove(struct lru *lru, struct lru_entry *le) +{ + lru_iter_invalidate(lru, le); + if (lru->count == 1) { + lru->cursor = NULL; + } else { + if (lru->cursor == &le->list) + lru->cursor = lru->cursor->next; + list_del(&le->list); + } + lru->count--; +} + +/* + * Mark as referenced. + */ +static inline void lru_reference(struct lru_entry *le) +{ + atomic_set(&le->referenced, 1); +} + +/*--------------*/ + +/* + * Remove the least recently used entry (approx), that passes the predicate. + * Returns NULL on failure. + */ +enum evict_result { + ER_EVICT, + ER_DONT_EVICT, + ER_STOP, /* stop looking for something to evict */ }; +typedef enum evict_result (*le_predicate)(struct lru_entry *le, void *context); + +static struct lru_entry *lru_evict(struct lru *lru, le_predicate pred, void *context, bool no_sleep) +{ + unsigned long tested = 0; + struct list_head *h = lru->cursor; + struct lru_entry *le; + + if (!h) + return NULL; + /* + * In the worst case we have to loop around twice. Once to clear + * the reference flags, and then again to discover the predicate + * fails for all entries. + */ + while (tested < lru->count) { + le = container_of(h, struct lru_entry, list); + + if (atomic_read(&le->referenced)) { + atomic_set(&le->referenced, 0); + } else { + tested++; + switch (pred(le, context)) { + case ER_EVICT: + /* + * Adjust the cursor, so we start the next + * search from here. + */ + lru->cursor = le->list.next; + lru_remove(lru, le); + return le; + + case ER_DONT_EVICT: + break; + + case ER_STOP: + lru->cursor = le->list.next; + return NULL; + } + } + + h = h->next; + + if (!no_sleep) + cond_resched(); + } + + return NULL; +} + +/*--------------------------------------------------------------*/ + /* * Buffer state bits. */ @@ -128,32 +310,44 @@ struct dm_bufio_client { */ enum data_mode { DATA_MODE_SLAB = 0, - DATA_MODE_GET_FREE_PAGES = 1, - DATA_MODE_VMALLOC = 2, - DATA_MODE_LIMIT = 3 + DATA_MODE_KMALLOC = 1, + DATA_MODE_GET_FREE_PAGES = 2, + DATA_MODE_VMALLOC = 3, + DATA_MODE_LIMIT = 4 }; struct dm_buffer { + /* protected by the locks in dm_buffer_cache */ struct rb_node node; - struct list_head lru_list; - struct list_head global_list; + + /* immutable, so don't need protecting */ sector_t block; void *data; unsigned char data_mode; /* DATA_MODE_* */ + + /* + * These two fields are used in isolation, so do not need + * a surrounding lock. + */ + atomic_t hold_count; + unsigned long last_accessed; + + /* + * Everything else is protected by the mutex in + * dm_bufio_client + */ + unsigned long state; + struct lru_entry lru; unsigned char list_mode; /* LIST_* */ blk_status_t read_error; blk_status_t write_error; - unsigned accessed; - unsigned hold_count; - unsigned long state; - unsigned long last_accessed; - unsigned dirty_start; - unsigned dirty_end; - unsigned write_start; - unsigned write_end; - struct dm_bufio_client *c; + unsigned int dirty_start; + unsigned int dirty_end; + unsigned int write_start; + unsigned int write_end; struct list_head write_list; - void (*end_io)(struct dm_buffer *, blk_status_t); + struct dm_bufio_client *c; + void (*end_io)(struct dm_buffer *b, blk_status_t bs); #ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING #define MAX_STACK 10 unsigned int stack_len; @@ -161,113 +355,540 @@ struct dm_buffer { #endif }; -/*----------------------------------------------------------------*/ +/*--------------------------------------------------------------*/ -#define dm_bufio_in_request() (!!current->bio_list) +/* + * The buffer cache manages buffers, particularly: + * - inc/dec of holder count + * - setting the last_accessed field + * - maintains clean/dirty state along with lru + * - selecting buffers that match predicates + * + * It does *not* handle: + * - allocation/freeing of buffers. + * - IO + * - Eviction or cache sizing. + * + * cache_get() and cache_put() are threadsafe, you do not need to + * protect these calls with a surrounding mutex. All the other + * methods are not threadsafe; they do use locking primitives, but + * only enough to ensure get/put are threadsafe. + */ -static void dm_bufio_lock(struct dm_bufio_client *c) +struct buffer_tree { + union { + struct rw_semaphore lock; + rwlock_t spinlock; + } u; + struct rb_root root; +} ____cacheline_aligned_in_smp; + +struct dm_buffer_cache { + struct lru lru[LIST_SIZE]; + /* + * We spread entries across multiple trees to reduce contention + * on the locks. + */ + unsigned int num_locks; + bool no_sleep; + struct buffer_tree trees[]; +}; + +static DEFINE_STATIC_KEY_FALSE(no_sleep_enabled); + +static inline unsigned int cache_index(sector_t block, unsigned int num_locks) { - mutex_lock_nested(&c->lock, dm_bufio_in_request()); + return dm_hash_locks_index(block, num_locks); } -static int dm_bufio_trylock(struct dm_bufio_client *c) +static inline void cache_read_lock(struct dm_buffer_cache *bc, sector_t block) { - return mutex_trylock(&c->lock); + if (static_branch_unlikely(&no_sleep_enabled) && bc->no_sleep) + read_lock_bh(&bc->trees[cache_index(block, bc->num_locks)].u.spinlock); + else + down_read(&bc->trees[cache_index(block, bc->num_locks)].u.lock); } -static void dm_bufio_unlock(struct dm_bufio_client *c) +static inline void cache_read_unlock(struct dm_buffer_cache *bc, sector_t block) { - mutex_unlock(&c->lock); + if (static_branch_unlikely(&no_sleep_enabled) && bc->no_sleep) + read_unlock_bh(&bc->trees[cache_index(block, bc->num_locks)].u.spinlock); + else + up_read(&bc->trees[cache_index(block, bc->num_locks)].u.lock); } -/*----------------------------------------------------------------*/ +static inline void cache_write_lock(struct dm_buffer_cache *bc, sector_t block) +{ + if (static_branch_unlikely(&no_sleep_enabled) && bc->no_sleep) + write_lock_bh(&bc->trees[cache_index(block, bc->num_locks)].u.spinlock); + else + down_write(&bc->trees[cache_index(block, bc->num_locks)].u.lock); +} + +static inline void cache_write_unlock(struct dm_buffer_cache *bc, sector_t block) +{ + if (static_branch_unlikely(&no_sleep_enabled) && bc->no_sleep) + write_unlock_bh(&bc->trees[cache_index(block, bc->num_locks)].u.spinlock); + else + up_write(&bc->trees[cache_index(block, bc->num_locks)].u.lock); +} /* - * Default cache size: available memory divided by the ratio. + * Sometimes we want to repeatedly get and drop locks as part of an iteration. + * This struct helps avoid redundant drop and gets of the same lock. */ -static unsigned long dm_bufio_default_cache_size; +struct lock_history { + struct dm_buffer_cache *cache; + bool write; + unsigned int previous; + unsigned int no_previous; +}; + +static void lh_init(struct lock_history *lh, struct dm_buffer_cache *cache, bool write) +{ + lh->cache = cache; + lh->write = write; + lh->no_previous = cache->num_locks; + lh->previous = lh->no_previous; +} + +static void __lh_lock(struct lock_history *lh, unsigned int index) +{ + if (lh->write) { + if (static_branch_unlikely(&no_sleep_enabled) && lh->cache->no_sleep) + write_lock_bh(&lh->cache->trees[index].u.spinlock); + else + down_write(&lh->cache->trees[index].u.lock); + } else { + if (static_branch_unlikely(&no_sleep_enabled) && lh->cache->no_sleep) + read_lock_bh(&lh->cache->trees[index].u.spinlock); + else + down_read(&lh->cache->trees[index].u.lock); + } +} + +static void __lh_unlock(struct lock_history *lh, unsigned int index) +{ + if (lh->write) { + if (static_branch_unlikely(&no_sleep_enabled) && lh->cache->no_sleep) + write_unlock_bh(&lh->cache->trees[index].u.spinlock); + else + up_write(&lh->cache->trees[index].u.lock); + } else { + if (static_branch_unlikely(&no_sleep_enabled) && lh->cache->no_sleep) + read_unlock_bh(&lh->cache->trees[index].u.spinlock); + else + up_read(&lh->cache->trees[index].u.lock); + } +} /* - * Total cache size set by the user. + * Make sure you call this since it will unlock the final lock. */ -static unsigned long dm_bufio_cache_size; +static void lh_exit(struct lock_history *lh) +{ + if (lh->previous != lh->no_previous) { + __lh_unlock(lh, lh->previous); + lh->previous = lh->no_previous; + } +} /* - * A copy of dm_bufio_cache_size because dm_bufio_cache_size can change - * at any time. If it disagrees, the user has changed cache size. + * Named 'next' because there is no corresponding + * 'up/unlock' call since it's done automatically. */ -static unsigned long dm_bufio_cache_size_latch; +static void lh_next(struct lock_history *lh, sector_t b) +{ + unsigned int index = cache_index(b, lh->no_previous); /* no_previous is num_locks */ -static DEFINE_SPINLOCK(global_spinlock); + if (lh->previous != lh->no_previous) { + if (lh->previous != index) { + __lh_unlock(lh, lh->previous); + __lh_lock(lh, index); + lh->previous = index; + } + } else { + __lh_lock(lh, index); + lh->previous = index; + } +} + +static inline struct dm_buffer *le_to_buffer(struct lru_entry *le) +{ + return container_of(le, struct dm_buffer, lru); +} + +static struct dm_buffer *list_to_buffer(struct list_head *l) +{ + struct lru_entry *le = list_entry(l, struct lru_entry, list); + + return le_to_buffer(le); +} -static LIST_HEAD(global_queue); +static void cache_init(struct dm_buffer_cache *bc, unsigned int num_locks, bool no_sleep) +{ + unsigned int i; + + bc->num_locks = num_locks; + bc->no_sleep = no_sleep; + + for (i = 0; i < bc->num_locks; i++) { + if (no_sleep) + rwlock_init(&bc->trees[i].u.spinlock); + else + init_rwsem(&bc->trees[i].u.lock); + bc->trees[i].root = RB_ROOT; + } + + lru_init(&bc->lru[LIST_CLEAN]); + lru_init(&bc->lru[LIST_DIRTY]); +} + +static void cache_destroy(struct dm_buffer_cache *bc) +{ + unsigned int i; + + for (i = 0; i < bc->num_locks; i++) + WARN_ON_ONCE(!RB_EMPTY_ROOT(&bc->trees[i].root)); + + lru_destroy(&bc->lru[LIST_CLEAN]); + lru_destroy(&bc->lru[LIST_DIRTY]); +} -static unsigned long global_num = 0; +/*--------------*/ /* - * Buffers are freed after this timeout + * not threadsafe, or racey depending how you look at it */ -static unsigned dm_bufio_max_age = DM_BUFIO_DEFAULT_AGE_SECS; -static unsigned long dm_bufio_retain_bytes = DM_BUFIO_DEFAULT_RETAIN_BYTES; +static inline unsigned long cache_count(struct dm_buffer_cache *bc, int list_mode) +{ + return bc->lru[list_mode].count; +} -static unsigned long dm_bufio_peak_allocated; -static unsigned long dm_bufio_allocated_kmem_cache; -static unsigned long dm_bufio_allocated_get_free_pages; -static unsigned long dm_bufio_allocated_vmalloc; -static unsigned long dm_bufio_current_allocated; +static inline unsigned long cache_total(struct dm_buffer_cache *bc) +{ + return cache_count(bc, LIST_CLEAN) + cache_count(bc, LIST_DIRTY); +} -/*----------------------------------------------------------------*/ +/*--------------*/ /* - * The current number of clients. + * Gets a specific buffer, indexed by block. + * If the buffer is found then its holder count will be incremented and + * lru_reference will be called. + * + * threadsafe */ -static int dm_bufio_client_count; +static struct dm_buffer *__cache_get(const struct rb_root *root, sector_t block) +{ + struct rb_node *n = root->rb_node; + struct dm_buffer *b; + + while (n) { + b = container_of(n, struct dm_buffer, node); + + if (b->block == block) + return b; + + n = block < b->block ? n->rb_left : n->rb_right; + } + + return NULL; +} + +static void __cache_inc_buffer(struct dm_buffer *b) +{ + atomic_inc(&b->hold_count); + WRITE_ONCE(b->last_accessed, jiffies); +} + +static struct dm_buffer *cache_get(struct dm_buffer_cache *bc, sector_t block) +{ + struct dm_buffer *b; + + cache_read_lock(bc, block); + b = __cache_get(&bc->trees[cache_index(block, bc->num_locks)].root, block); + if (b) { + lru_reference(&b->lru); + __cache_inc_buffer(b); + } + cache_read_unlock(bc, block); + + return b; +} + +/*--------------*/ /* - * The list of all clients. + * Returns true if the hold count hits zero. + * threadsafe */ -static LIST_HEAD(dm_bufio_all_clients); +static bool cache_put(struct dm_buffer_cache *bc, struct dm_buffer *b) +{ + bool r; + + cache_read_lock(bc, b->block); + BUG_ON(!atomic_read(&b->hold_count)); + r = atomic_dec_and_test(&b->hold_count); + cache_read_unlock(bc, b->block); + + return r; +} + +/*--------------*/ + +typedef enum evict_result (*b_predicate)(struct dm_buffer *, void *); /* - * This mutex protects dm_bufio_cache_size_latch and dm_bufio_client_count + * Evicts a buffer based on a predicate. The oldest buffer that + * matches the predicate will be selected. In addition to the + * predicate the hold_count of the selected buffer will be zero. */ -static DEFINE_MUTEX(dm_bufio_clients_lock); +struct evict_wrapper { + struct lock_history *lh; + b_predicate pred; + void *context; +}; -static struct workqueue_struct *dm_bufio_wq; -static struct delayed_work dm_bufio_cleanup_old_work; -static struct work_struct dm_bufio_replacement_work; +/* + * Wraps the buffer predicate turning it into an lru predicate. Adds + * extra test for hold_count. + */ +static enum evict_result __evict_pred(struct lru_entry *le, void *context) +{ + struct evict_wrapper *w = context; + struct dm_buffer *b = le_to_buffer(le); + lh_next(w->lh, b->block); -#ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING -static void buffer_record_stack(struct dm_buffer *b) + if (atomic_read(&b->hold_count)) + return ER_DONT_EVICT; + + return w->pred(b, w->context); +} + +static struct dm_buffer *__cache_evict(struct dm_buffer_cache *bc, int list_mode, + b_predicate pred, void *context, + struct lock_history *lh) { - b->stack_len = stack_trace_save(b->stack_entries, MAX_STACK, 2); + struct evict_wrapper w = {.lh = lh, .pred = pred, .context = context}; + struct lru_entry *le; + struct dm_buffer *b; + + le = lru_evict(&bc->lru[list_mode], __evict_pred, &w, bc->no_sleep); + if (!le) + return NULL; + + b = le_to_buffer(le); + /* __evict_pred will have locked the appropriate tree. */ + rb_erase(&b->node, &bc->trees[cache_index(b->block, bc->num_locks)].root); + + return b; } -#endif -/*---------------------------------------------------------------- - * A red/black tree acts as an index for all the buffers. - *--------------------------------------------------------------*/ -static struct dm_buffer *__find(struct dm_bufio_client *c, sector_t block) +static struct dm_buffer *cache_evict(struct dm_buffer_cache *bc, int list_mode, + b_predicate pred, void *context) { - struct rb_node *n = c->buffer_tree.rb_node; struct dm_buffer *b; + struct lock_history lh; - while (n) { - b = container_of(n, struct dm_buffer, node); + lh_init(&lh, bc, true); + b = __cache_evict(bc, list_mode, pred, context, &lh); + lh_exit(&lh); - if (b->block == block) - return b; + return b; +} - n = block < b->block ? n->rb_left : n->rb_right; +/*--------------*/ + +/* + * Mark a buffer as clean or dirty. Not threadsafe. + */ +static void cache_mark(struct dm_buffer_cache *bc, struct dm_buffer *b, int list_mode) +{ + cache_write_lock(bc, b->block); + if (list_mode != b->list_mode) { + lru_remove(&bc->lru[b->list_mode], &b->lru); + b->list_mode = list_mode; + lru_insert(&bc->lru[b->list_mode], &b->lru); } + cache_write_unlock(bc, b->block); +} - return NULL; +/*--------------*/ + +/* + * Runs through the lru associated with 'old_mode', if the predicate matches then + * it moves them to 'new_mode'. Not threadsafe. + */ +static void __cache_mark_many(struct dm_buffer_cache *bc, int old_mode, int new_mode, + b_predicate pred, void *context, struct lock_history *lh) +{ + struct lru_entry *le; + struct dm_buffer *b; + struct evict_wrapper w = {.lh = lh, .pred = pred, .context = context}; + + while (true) { + le = lru_evict(&bc->lru[old_mode], __evict_pred, &w, bc->no_sleep); + if (!le) + break; + + b = le_to_buffer(le); + b->list_mode = new_mode; + lru_insert(&bc->lru[b->list_mode], &b->lru); + } +} + +static void cache_mark_many(struct dm_buffer_cache *bc, int old_mode, int new_mode, + b_predicate pred, void *context) +{ + struct lock_history lh; + + lh_init(&lh, bc, true); + __cache_mark_many(bc, old_mode, new_mode, pred, context, &lh); + lh_exit(&lh); +} + +/*--------------*/ + +/* + * Iterates through all clean or dirty entries calling a function for each + * entry. The callback may terminate the iteration early. Not threadsafe. + */ + +/* + * Iterator functions should return one of these actions to indicate + * how the iteration should proceed. + */ +enum it_action { + IT_NEXT, + IT_COMPLETE, +}; + +typedef enum it_action (*iter_fn)(struct dm_buffer *b, void *context); + +static void __cache_iterate(struct dm_buffer_cache *bc, int list_mode, + iter_fn fn, void *context, struct lock_history *lh) +{ + struct lru *lru = &bc->lru[list_mode]; + struct lru_entry *le, *first; + + if (!lru->cursor) + return; + + first = le = to_le(lru->cursor); + do { + struct dm_buffer *b = le_to_buffer(le); + + lh_next(lh, b->block); + + switch (fn(b, context)) { + case IT_NEXT: + break; + + case IT_COMPLETE: + return; + } + cond_resched(); + + le = to_le(le->list.next); + } while (le != first); +} + +static void cache_iterate(struct dm_buffer_cache *bc, int list_mode, + iter_fn fn, void *context) +{ + struct lock_history lh; + + lh_init(&lh, bc, false); + __cache_iterate(bc, list_mode, fn, context, &lh); + lh_exit(&lh); +} + +/*--------------*/ + +/* + * Passes ownership of the buffer to the cache. Returns false if the + * buffer was already present (in which case ownership does not pass). + * eg, a race with another thread. + * + * Holder count should be 1 on insertion. + * + * Not threadsafe. + */ +static bool __cache_insert(struct rb_root *root, struct dm_buffer *b) +{ + struct rb_node **new = &root->rb_node, *parent = NULL; + struct dm_buffer *found; + + while (*new) { + found = container_of(*new, struct dm_buffer, node); + + if (found->block == b->block) + return false; + + parent = *new; + new = b->block < found->block ? + &found->node.rb_left : &found->node.rb_right; + } + + rb_link_node(&b->node, parent, new); + rb_insert_color(&b->node, root); + + return true; +} + +static bool cache_insert(struct dm_buffer_cache *bc, struct dm_buffer *b) +{ + bool r; + + if (WARN_ON_ONCE(b->list_mode >= LIST_SIZE)) + return false; + + cache_write_lock(bc, b->block); + BUG_ON(atomic_read(&b->hold_count) != 1); + r = __cache_insert(&bc->trees[cache_index(b->block, bc->num_locks)].root, b); + if (r) + lru_insert(&bc->lru[b->list_mode], &b->lru); + cache_write_unlock(bc, b->block); + + return r; +} + +/*--------------*/ + +/* + * Removes buffer from cache, ownership of the buffer passes back to the caller. + * Fails if the hold_count is not one (ie. the caller holds the only reference). + * + * Not threadsafe. + */ +static bool cache_remove(struct dm_buffer_cache *bc, struct dm_buffer *b) +{ + bool r; + + cache_write_lock(bc, b->block); + + if (atomic_read(&b->hold_count) != 1) { + r = false; + } else { + r = true; + rb_erase(&b->node, &bc->trees[cache_index(b->block, bc->num_locks)].root); + lru_remove(&bc->lru[b->list_mode], &b->lru); + } + + cache_write_unlock(bc, b->block); + + return r; } -static struct dm_buffer *__find_next(struct dm_bufio_client *c, sector_t block) +/*--------------*/ + +typedef void (*b_release)(struct dm_buffer *); + +static struct dm_buffer *__find_next(struct rb_root *root, sector_t block) { - struct rb_node *n = c->buffer_tree.rb_node; + struct rb_node *n = root->rb_node; struct dm_buffer *b; struct dm_buffer *best = NULL; @@ -288,32 +909,182 @@ static struct dm_buffer *__find_next(struct dm_bufio_client *c, sector_t block) return best; } -static void __insert(struct dm_bufio_client *c, struct dm_buffer *b) +static void __remove_range(struct dm_buffer_cache *bc, + struct rb_root *root, + sector_t begin, sector_t end, + b_predicate pred, b_release release) { - struct rb_node **new = &c->buffer_tree.rb_node, *parent = NULL; - struct dm_buffer *found; + struct dm_buffer *b; - while (*new) { - found = container_of(*new, struct dm_buffer, node); + while (true) { + cond_resched(); - if (found->block == b->block) { - BUG_ON(found != b); - return; + b = __find_next(root, begin); + if (!b || (b->block >= end)) + break; + + begin = b->block + 1; + + if (atomic_read(&b->hold_count)) + continue; + + if (pred(b, NULL) == ER_EVICT) { + rb_erase(&b->node, root); + lru_remove(&bc->lru[b->list_mode], &b->lru); + release(b); } + } +} - parent = *new; - new = b->block < found->block ? - &found->node.rb_left : &found->node.rb_right; +static void cache_remove_range(struct dm_buffer_cache *bc, + sector_t begin, sector_t end, + b_predicate pred, b_release release) +{ + unsigned int i; + + BUG_ON(bc->no_sleep); + for (i = 0; i < bc->num_locks; i++) { + down_write(&bc->trees[i].u.lock); + __remove_range(bc, &bc->trees[i].root, begin, end, pred, release); + up_write(&bc->trees[i].u.lock); } +} - rb_link_node(&b->node, parent, new); - rb_insert_color(&b->node, &c->buffer_tree); +/*----------------------------------------------------------------*/ + +/* + * Linking of buffers: + * All buffers are linked to buffer_cache with their node field. + * + * Clean buffers that are not being written (B_WRITING not set) + * are linked to lru[LIST_CLEAN] with their lru_list field. + * + * Dirty and clean buffers that are being written are linked to + * lru[LIST_DIRTY] with their lru_list field. When the write + * finishes, the buffer cannot be relinked immediately (because we + * are in an interrupt context and relinking requires process + * context), so some clean-not-writing buffers can be held on + * dirty_lru too. They are later added to lru in the process + * context. + */ +struct dm_bufio_client { + struct block_device *bdev; + unsigned int block_size; + s8 sectors_per_block_bits; + + bool no_sleep; + struct mutex lock; + spinlock_t spinlock; + + int async_write_error; + + void (*alloc_callback)(struct dm_buffer *buf); + void (*write_callback)(struct dm_buffer *buf); + struct kmem_cache *slab_buffer; + struct kmem_cache *slab_cache; + struct dm_io_client *dm_io; + + struct list_head reserved_buffers; + unsigned int need_reserved_buffers; + + unsigned int minimum_buffers; + + sector_t start; + + struct shrinker *shrinker; + struct work_struct shrink_work; + atomic_long_t need_shrink; + + wait_queue_head_t free_buffer_wait; + + struct list_head client_list; + + /* + * Used by global_cleanup to sort the clients list. + */ + unsigned long oldest_buffer; + + struct dm_buffer_cache cache; /* must be last member */ +}; + +/*----------------------------------------------------------------*/ + +#define dm_bufio_in_request() (!!current->bio_list) + +static void dm_bufio_lock(struct dm_bufio_client *c) +{ + if (static_branch_unlikely(&no_sleep_enabled) && c->no_sleep) + spin_lock_bh(&c->spinlock); + else + mutex_lock_nested(&c->lock, dm_bufio_in_request()); } -static void __remove(struct dm_bufio_client *c, struct dm_buffer *b) +static void dm_bufio_unlock(struct dm_bufio_client *c) +{ + if (static_branch_unlikely(&no_sleep_enabled) && c->no_sleep) + spin_unlock_bh(&c->spinlock); + else + mutex_unlock(&c->lock); +} + +/*----------------------------------------------------------------*/ + +/* + * Default cache size: available memory divided by the ratio. + */ +static unsigned long dm_bufio_default_cache_size; + +/* + * Total cache size set by the user. + */ +static unsigned long dm_bufio_cache_size; + +/* + * A copy of dm_bufio_cache_size because dm_bufio_cache_size can change + * at any time. If it disagrees, the user has changed cache size. + */ +static unsigned long dm_bufio_cache_size_latch; + +static DEFINE_SPINLOCK(global_spinlock); + +static unsigned int dm_bufio_max_age; /* No longer does anything */ + +static unsigned long dm_bufio_retain_bytes = DM_BUFIO_DEFAULT_RETAIN_BYTES; + +static unsigned long dm_bufio_peak_allocated; +static unsigned long dm_bufio_allocated_kmem_cache; +static unsigned long dm_bufio_allocated_kmalloc; +static unsigned long dm_bufio_allocated_get_free_pages; +static unsigned long dm_bufio_allocated_vmalloc; +static unsigned long dm_bufio_current_allocated; + +/*----------------------------------------------------------------*/ + +/* + * The current number of clients. + */ +static int dm_bufio_client_count; + +/* + * The list of all clients. + */ +static LIST_HEAD(dm_bufio_all_clients); + +/* + * This mutex protects dm_bufio_cache_size_latch and dm_bufio_client_count + */ +static DEFINE_MUTEX(dm_bufio_clients_lock); + +static struct workqueue_struct *dm_bufio_wq; +static struct work_struct dm_bufio_replacement_work; + + +#ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING +static void buffer_record_stack(struct dm_buffer *b) { - rb_erase(&b->node, &c->buffer_tree); + b->stack_len = stack_trace_save(b->stack_entries, MAX_STACK, 2); } +#endif /*----------------------------------------------------------------*/ @@ -324,6 +1095,7 @@ static void adjust_total_allocated(struct dm_buffer *b, bool unlink) static unsigned long * const class_ptr[DATA_MODE_LIMIT] = { &dm_bufio_allocated_kmem_cache, + &dm_bufio_allocated_kmalloc, &dm_bufio_allocated_get_free_pages, &dm_bufio_allocated_vmalloc, }; @@ -342,16 +1114,9 @@ static void adjust_total_allocated(struct dm_buffer *b, bool unlink) if (dm_bufio_current_allocated > dm_bufio_peak_allocated) dm_bufio_peak_allocated = dm_bufio_current_allocated; - b->accessed = 1; - if (!unlink) { - list_add(&b->global_list, &global_queue); - global_num++; if (dm_bufio_current_allocated > dm_bufio_cache_size) queue_work(dm_bufio_wq, &dm_bufio_replacement_work); - } else { - list_del(&b->global_list); - global_num--; } spin_unlock(&global_spinlock); @@ -362,8 +1127,10 @@ static void adjust_total_allocated(struct dm_buffer *b, bool unlink) */ static void __cache_size_refresh(void) { - BUG_ON(!mutex_is_locked(&dm_bufio_clients_lock)); - BUG_ON(dm_bufio_client_count < 0); + if (WARN_ON(!mutex_is_locked(&dm_bufio_clients_lock))) + return; + if (WARN_ON(dm_bufio_client_count < 0)) + return; dm_bufio_cache_size_latch = READ_ONCE(dm_bufio_cache_size); @@ -392,7 +1159,7 @@ static void __cache_size_refresh(void) * If the allocation may fail we use __get_free_pages. Memory fragmentation * won't have a fatal effect here, but it just causes flushes of some other * buffers and more I/O will be performed. Don't use __get_free_pages if it - * always fails (i.e. order >= MAX_ORDER). + * always fails (i.e. order > MAX_PAGE_ORDER). * * If the allocation shouldn't fail we use __vmalloc. This is only for the * initial reserve allocation, so there's no risk of wasting all vmalloc @@ -406,6 +1173,11 @@ static void *alloc_buffer_data(struct dm_bufio_client *c, gfp_t gfp_mask, return kmem_cache_alloc(c->slab_cache, gfp_mask); } + if (unlikely(c->block_size < PAGE_SIZE)) { + *data_mode = DATA_MODE_KMALLOC; + return kmalloc(c->block_size, gfp_mask | __GFP_RECLAIMABLE); + } + if (c->block_size <= KMALLOC_MAX_SIZE && gfp_mask & __GFP_NORETRY) { *data_mode = DATA_MODE_GET_FREE_PAGES; @@ -415,23 +1187,6 @@ static void *alloc_buffer_data(struct dm_bufio_client *c, gfp_t gfp_mask, *data_mode = DATA_MODE_VMALLOC; - /* - * __vmalloc allocates the data pages and auxiliary structures with - * gfp_flags that were specified, but pagetables are always allocated - * with GFP_KERNEL, no matter what was specified as gfp_mask. - * - * Consequently, we must set per-process flag PF_MEMALLOC_NOIO so that - * all allocations done by this process (including pagetables) are done - * as if GFP_NOIO was specified. - */ - if (gfp_mask & __GFP_NORETRY) { - unsigned noio_flag = memalloc_noio_save(); - void *ptr = __vmalloc(c->block_size, gfp_mask); - - memalloc_noio_restore(noio_flag); - return ptr; - } - return __vmalloc(c->block_size, gfp_mask); } @@ -446,6 +1201,10 @@ static void free_buffer_data(struct dm_bufio_client *c, kmem_cache_free(c->slab_cache, data); break; + case DATA_MODE_KMALLOC: + kfree(data); + break; + case DATA_MODE_GET_FREE_PAGES: free_pages((unsigned long)data, c->sectors_per_block_bits - (PAGE_SHIFT - SECTOR_SHIFT)); @@ -479,6 +1238,7 @@ static struct dm_buffer *alloc_buffer(struct dm_bufio_client *c, gfp_t gfp_mask) kmem_cache_free(c->slab_buffer, b); return NULL; } + adjust_total_allocated(b, false); #ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING b->stack_len = 0; @@ -493,62 +1253,13 @@ static void free_buffer(struct dm_buffer *b) { struct dm_bufio_client *c = b->c; + adjust_total_allocated(b, true); free_buffer_data(c, b->data, b->data_mode); kmem_cache_free(c->slab_buffer, b); } /* - * Link buffer to the buffer tree and clean or dirty queue. - */ -static void __link_buffer(struct dm_buffer *b, sector_t block, int dirty) -{ - struct dm_bufio_client *c = b->c; - - c->n_buffers[dirty]++; - b->block = block; - b->list_mode = dirty; - list_add(&b->lru_list, &c->lru[dirty]); - __insert(b->c, b); - b->last_accessed = jiffies; - - adjust_total_allocated(b, false); -} - -/* - * Unlink buffer from the buffer tree and dirty or clean queue. - */ -static void __unlink_buffer(struct dm_buffer *b) -{ - struct dm_bufio_client *c = b->c; - - BUG_ON(!c->n_buffers[b->list_mode]); - - c->n_buffers[b->list_mode]--; - __remove(b->c, b); - list_del(&b->lru_list); - - adjust_total_allocated(b, true); -} - -/* - * Place the buffer to the head of dirty or clean LRU queue. - */ -static void __relink_lru(struct dm_buffer *b, int dirty) -{ - struct dm_bufio_client *c = b->c; - - b->accessed = 1; - - BUG_ON(!c->n_buffers[b->list_mode]); - - c->n_buffers[b->list_mode]--; - c->n_buffers[dirty]++; - b->list_mode = dirty; - list_move(&b->lru_list, &c->lru[dirty]); - b->last_accessed = jiffies; -} - -/*---------------------------------------------------------------- + *-------------------------------------------------------------------------- * Submit I/O on the buffer. * * Bio interface is faster but it has some problems: @@ -564,7 +1275,8 @@ static void __relink_lru(struct dm_buffer *b, int dirty) * rejects the bio because it is too large, use dm-io layer to do the I/O. * The dm-io layer splits the I/O into multiple requests, avoiding the above * shortcomings. - *--------------------------------------------------------------*/ + *-------------------------------------------------------------------------- + */ /* * dm-io completion routine. It just calls b->bio.bi_end_io, pretending @@ -577,13 +1289,13 @@ static void dmio_complete(unsigned long error, void *context) b->end_io(b, unlikely(error != 0) ? BLK_STS_IOERR : 0); } -static void use_dmio(struct dm_buffer *b, int rw, sector_t sector, - unsigned n_sectors, unsigned offset) +static void use_dmio(struct dm_buffer *b, enum req_op op, sector_t sector, + unsigned int n_sectors, unsigned int offset, + unsigned short ioprio) { int r; struct dm_io_request io_req = { - .bi_op = rw, - .bi_op_flags = 0, + .bi_opf = op, .notify.fn = dmio_complete, .notify.context = b, .client = b->c->dm_io, @@ -602,7 +1314,7 @@ static void use_dmio(struct dm_buffer *b, int rw, sector_t sector, io_req.mem.ptr.vma = (char *)b->data + offset; } - r = dm_io(&io_req, 1, ®ion, NULL); + r = dm_io(&io_req, 1, ®ion, NULL, ioprio); if (unlikely(r)) b->end_io(b, errno_to_blk_status(r)); } @@ -611,48 +1323,35 @@ static void bio_complete(struct bio *bio) { struct dm_buffer *b = bio->bi_private; blk_status_t status = bio->bi_status; - bio_put(bio); + + bio_uninit(bio); + kfree(bio); b->end_io(b, status); } -static void use_bio(struct dm_buffer *b, int rw, sector_t sector, - unsigned n_sectors, unsigned offset) +static void use_bio(struct dm_buffer *b, enum req_op op, sector_t sector, + unsigned int n_sectors, unsigned int offset, + unsigned short ioprio) { struct bio *bio; char *ptr; - unsigned vec_size, len; - - vec_size = b->c->block_size >> PAGE_SHIFT; - if (unlikely(b->c->sectors_per_block_bits < PAGE_SHIFT - SECTOR_SHIFT)) - vec_size += 2; + unsigned int len; - bio = bio_kmalloc(GFP_NOWAIT | __GFP_NORETRY | __GFP_NOWARN, vec_size); + bio = bio_kmalloc(1, GFP_NOWAIT); if (!bio) { -dmio: - use_dmio(b, rw, sector, n_sectors, offset); + use_dmio(b, op, sector, n_sectors, offset, ioprio); return; } - + bio_init_inline(bio, b->c->bdev, 1, op); bio->bi_iter.bi_sector = sector; - bio_set_dev(bio, b->c->bdev); - bio_set_op_attrs(bio, rw, 0); bio->bi_end_io = bio_complete; bio->bi_private = b; + bio->bi_ioprio = ioprio; ptr = (char *)b->data + offset; len = n_sectors << SECTOR_SHIFT; - do { - unsigned this_step = min((unsigned)(PAGE_SIZE - offset_in_page(ptr)), len); - if (!bio_add_page(bio, virt_to_page(ptr), this_step, - offset_in_page(ptr))) { - bio_put(bio); - goto dmio; - } - - len -= this_step; - ptr += this_step; - } while (len > 0); + bio_add_virt_nofail(bio, ptr, len); submit_bio(bio); } @@ -670,17 +1369,18 @@ static inline sector_t block_to_sector(struct dm_bufio_client *c, sector_t block return sector; } -static void submit_io(struct dm_buffer *b, int rw, void (*end_io)(struct dm_buffer *, blk_status_t)) +static void submit_io(struct dm_buffer *b, enum req_op op, unsigned short ioprio, + void (*end_io)(struct dm_buffer *, blk_status_t)) { - unsigned n_sectors; + unsigned int n_sectors; sector_t sector; - unsigned offset, end; + unsigned int offset, end; b->end_io = end_io; sector = block_to_sector(b->c, b->block); - if (rw != REQ_OP_WRITE) { + if (op != REQ_OP_WRITE) { n_sectors = b->c->block_size >> SECTOR_SHIFT; offset = 0; } else { @@ -699,14 +1399,16 @@ static void submit_io(struct dm_buffer *b, int rw, void (*end_io)(struct dm_buff } if (b->data_mode != DATA_MODE_VMALLOC) - use_bio(b, rw, sector, n_sectors, offset); + use_bio(b, op, sector, n_sectors, offset, ioprio); else - use_dmio(b, rw, sector, n_sectors, offset); + use_dmio(b, op, sector, n_sectors, offset, ioprio); } -/*---------------------------------------------------------------- +/* + *-------------------------------------------------------------- * Writing dirty buffers - *--------------------------------------------------------------*/ + *-------------------------------------------------------------- + */ /* * The endio routine for write. @@ -755,7 +1457,7 @@ static void __write_dirty_buffer(struct dm_buffer *b, b->write_end = b->dirty_end; if (!write_list) - submit_io(b, REQ_OP_WRITE, write_endio); + submit_io(b, REQ_OP_WRITE, IOPRIO_DEFAULT, write_endio); else list_add_tail(&b->write_list, write_list); } @@ -763,12 +1465,13 @@ static void __write_dirty_buffer(struct dm_buffer *b, static void __flush_write_list(struct list_head *write_list) { struct blk_plug plug; + blk_start_plug(&plug); while (!list_empty(write_list)) { struct dm_buffer *b = list_entry(write_list->next, struct dm_buffer, write_list); list_del(&b->write_list); - submit_io(b, REQ_OP_WRITE, write_endio); + submit_io(b, REQ_OP_WRITE, IOPRIO_DEFAULT, write_endio); cond_resched(); } blk_finish_plug(&plug); @@ -781,9 +1484,10 @@ static void __flush_write_list(struct list_head *write_list) */ static void __make_buffer_clean(struct dm_buffer *b) { - BUG_ON(b->hold_count); + BUG_ON(atomic_read(&b->hold_count)); - if (!b->state) /* fast case */ + /* smp_load_acquire() pairs with read_endio()'s smp_mb__before_atomic() */ + if (!smp_load_acquire(&b->state)) /* fast case */ return; wait_on_bit_io(&b->state, B_READING, TASK_UNINTERRUPTIBLE); @@ -791,6 +1495,36 @@ static void __make_buffer_clean(struct dm_buffer *b) wait_on_bit_io(&b->state, B_WRITING, TASK_UNINTERRUPTIBLE); } +static enum evict_result is_clean(struct dm_buffer *b, void *context) +{ + struct dm_bufio_client *c = context; + + /* These should never happen */ + if (WARN_ON_ONCE(test_bit(B_WRITING, &b->state))) + return ER_DONT_EVICT; + if (WARN_ON_ONCE(test_bit(B_DIRTY, &b->state))) + return ER_DONT_EVICT; + if (WARN_ON_ONCE(b->list_mode != LIST_CLEAN)) + return ER_DONT_EVICT; + + if (static_branch_unlikely(&no_sleep_enabled) && c->no_sleep && + unlikely(test_bit(B_READING, &b->state))) + return ER_DONT_EVICT; + + return ER_EVICT; +} + +static enum evict_result is_dirty(struct dm_buffer *b, void *context) +{ + /* These should never happen */ + if (WARN_ON_ONCE(test_bit(B_READING, &b->state))) + return ER_DONT_EVICT; + if (WARN_ON_ONCE(b->list_mode != LIST_DIRTY)) + return ER_DONT_EVICT; + + return ER_EVICT; +} + /* * Find some buffer that is not held by anybody, clean it, unlink it and * return it. @@ -799,27 +1533,20 @@ static struct dm_buffer *__get_unclaimed_buffer(struct dm_bufio_client *c) { struct dm_buffer *b; - list_for_each_entry_reverse(b, &c->lru[LIST_CLEAN], lru_list) { - BUG_ON(test_bit(B_WRITING, &b->state)); - BUG_ON(test_bit(B_DIRTY, &b->state)); - - if (!b->hold_count) { - __make_buffer_clean(b); - __unlink_buffer(b); - return b; - } - cond_resched(); + b = cache_evict(&c->cache, LIST_CLEAN, is_clean, c); + if (b) { + /* this also waits for pending reads */ + __make_buffer_clean(b); + return b; } - list_for_each_entry_reverse(b, &c->lru[LIST_DIRTY], lru_list) { - BUG_ON(test_bit(B_READING, &b->state)); + if (static_branch_unlikely(&no_sleep_enabled) && c->no_sleep) + return NULL; - if (!b->hold_count) { - __make_buffer_clean(b); - __unlink_buffer(b); - return b; - } - cond_resched(); + b = cache_evict(&c->cache, LIST_DIRTY, is_dirty, NULL); + if (b) { + __make_buffer_clean(b); + return b; } return NULL; @@ -840,7 +1567,12 @@ static void __wait_for_free_buffer(struct dm_bufio_client *c) set_current_state(TASK_UNINTERRUPTIBLE); dm_bufio_unlock(c); - io_schedule(); + /* + * It's possible to miss a wake up event since we don't always + * hold c->lock when wake_up is called. So we have a timeout here, + * just in case. + */ + io_schedule_timeout(5 * HZ); remove_wait_queue(&c->free_buffer_wait, &wait); @@ -869,18 +1601,18 @@ static struct dm_buffer *__alloc_buffer_wait_no_callback(struct dm_bufio_client * dm-bufio is resistant to allocation failures (it just keeps * one buffer reserved in cases all the allocations fail). * So set flags to not try too hard: - * GFP_NOWAIT: don't wait; if we need to sleep we'll release our - * mutex and wait ourselves. + * GFP_NOWAIT: don't wait and don't print a warning in case of + * failure; if we need to sleep we'll release our mutex + * and wait ourselves. * __GFP_NORETRY: don't retry and rather return failure * __GFP_NOMEMALLOC: don't use emergency reserves - * __GFP_NOWARN: don't print a warning in case of failure * * For debugging, if we set the cache size to 1, no new buffers will * be allocated. */ while (1) { if (dm_bufio_cache_size_latch != 1) { - b = alloc_buffer(c, GFP_NOWAIT | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN); + b = alloc_buffer(c, GFP_NOWAIT | __GFP_NORETRY | __GFP_NOMEMALLOC); if (b) return b; } @@ -898,9 +1630,8 @@ static struct dm_buffer *__alloc_buffer_wait_no_callback(struct dm_bufio_client } if (!list_empty(&c->reserved_buffers)) { - b = list_entry(c->reserved_buffers.next, - struct dm_buffer, lru_list); - list_del(&b->lru_list); + b = list_to_buffer(c->reserved_buffers.next); + list_del(&b->lru.list); c->need_reserved_buffers++; return b; @@ -934,36 +1665,61 @@ static void __free_buffer_wake(struct dm_buffer *b) { struct dm_bufio_client *c = b->c; + b->block = -1; if (!c->need_reserved_buffers) free_buffer(b); else { - list_add(&b->lru_list, &c->reserved_buffers); + list_add(&b->lru.list, &c->reserved_buffers); c->need_reserved_buffers--; } - wake_up(&c->free_buffer_wait); + /* + * We hold the bufio lock here, so no one can add entries to the + * wait queue anyway. + */ + if (unlikely(waitqueue_active(&c->free_buffer_wait))) + wake_up(&c->free_buffer_wait); } -static void __write_dirty_buffers_async(struct dm_bufio_client *c, int no_wait, - struct list_head *write_list) +static enum evict_result cleaned(struct dm_buffer *b, void *context) { - struct dm_buffer *b, *tmp; + if (WARN_ON_ONCE(test_bit(B_READING, &b->state))) + return ER_DONT_EVICT; /* should never happen */ - list_for_each_entry_safe_reverse(b, tmp, &c->lru[LIST_DIRTY], lru_list) { - BUG_ON(test_bit(B_READING, &b->state)); + if (test_bit(B_DIRTY, &b->state) || test_bit(B_WRITING, &b->state)) + return ER_DONT_EVICT; + else + return ER_EVICT; +} - if (!test_bit(B_DIRTY, &b->state) && - !test_bit(B_WRITING, &b->state)) { - __relink_lru(b, LIST_CLEAN); - continue; - } +static void __move_clean_buffers(struct dm_bufio_client *c) +{ + cache_mark_many(&c->cache, LIST_DIRTY, LIST_CLEAN, cleaned, NULL); +} - if (no_wait && test_bit(B_WRITING, &b->state)) - return; +struct write_context { + int no_wait; + struct list_head *write_list; +}; - __write_dirty_buffer(b, write_list); - cond_resched(); - } +static enum it_action write_one(struct dm_buffer *b, void *context) +{ + struct write_context *wc = context; + + if (wc->no_wait && test_bit(B_WRITING, &b->state)) + return IT_COMPLETE; + + __write_dirty_buffer(b, wc->write_list); + return IT_NEXT; +} + +static void __write_dirty_buffers_async(struct dm_bufio_client *c, int no_wait, + struct list_head *write_list) +{ + struct write_context wc = {.no_wait = no_wait, .write_list = write_list}; + + __move_clean_buffers(c); + cache_iterate(&c->cache, LIST_DIRTY, write_one, &wc); } /* @@ -974,14 +1730,32 @@ static void __write_dirty_buffers_async(struct dm_bufio_client *c, int no_wait, static void __check_watermark(struct dm_bufio_client *c, struct list_head *write_list) { - if (c->n_buffers[LIST_DIRTY] > c->n_buffers[LIST_CLEAN] * DM_BUFIO_WRITEBACK_RATIO) + if (cache_count(&c->cache, LIST_DIRTY) > + cache_count(&c->cache, LIST_CLEAN) * DM_BUFIO_WRITEBACK_RATIO) __write_dirty_buffers_async(c, 1, write_list); } -/*---------------------------------------------------------------- +/* + *-------------------------------------------------------------- * Getting a buffer - *--------------------------------------------------------------*/ + *-------------------------------------------------------------- + */ + +static void cache_put_and_wake(struct dm_bufio_client *c, struct dm_buffer *b) +{ + /* + * Relying on waitqueue_active() is racey, but we sleep + * with schedule_timeout anyway. + */ + if (cache_put(&c->cache, b) && + unlikely(waitqueue_active(&c->free_buffer_wait))) + wake_up(&c->free_buffer_wait); +} +/* + * This assumes you have already checked the cache to see if the buffer + * is already present (it will recheck after dropping the lock for allocation). + */ static struct dm_buffer *__bufio_new(struct dm_bufio_client *c, sector_t block, enum new_flag nf, int *need_submit, struct list_head *write_list) @@ -990,11 +1764,8 @@ static struct dm_buffer *__bufio_new(struct dm_bufio_client *c, sector_t block, *need_submit = 0; - b = __find(c, block); - if (b) - goto found_buffer; - - if (nf == NF_GET) + /* This can't be called with NF_GET */ + if (WARN_ON_ONCE(nf == NF_GET)) return NULL; new_b = __alloc_buffer_wait(c, nf); @@ -1005,7 +1776,7 @@ static struct dm_buffer *__bufio_new(struct dm_bufio_client *c, sector_t block, * We've had a period where the mutex was unlocked, so need to * recheck the buffer tree. */ - b = __find(c, block); + b = cache_get(&c->cache, block); if (b) { __free_buffer_wake(new_b); goto found_buffer; @@ -1014,24 +1785,35 @@ static struct dm_buffer *__bufio_new(struct dm_bufio_client *c, sector_t block, __check_watermark(c, write_list); b = new_b; - b->hold_count = 1; + atomic_set(&b->hold_count, 1); + WRITE_ONCE(b->last_accessed, jiffies); + b->block = block; b->read_error = 0; b->write_error = 0; - __link_buffer(b, block, LIST_CLEAN); + b->list_mode = LIST_CLEAN; - if (nf == NF_FRESH) { + if (nf == NF_FRESH) b->state = 0; - return b; + else { + b->state = 1 << B_READING; + *need_submit = 1; } - b->state = 1 << B_READING; - *need_submit = 1; + /* + * We mustn't insert into the cache until the B_READING state + * is set. Otherwise another thread could get it and use + * it before it had been read. + */ + cache_insert(&c->cache, b); return b; found_buffer: - if (nf == NF_PREFETCH) + if (nf == NF_PREFETCH) { + cache_put_and_wake(c, b); return NULL; + } + /* * Note: it is essential that we don't wait for the buffer to be * read if dm_bufio_get function is used. Both dm_bufio_get and @@ -1039,12 +1821,11 @@ found_buffer: * If the user called both dm_bufio_prefetch and dm_bufio_get on * the same buffer, it would deadlock if we waited. */ - if (nf == NF_GET && unlikely(test_bit(B_READING, &b->state))) + if (nf == NF_GET && unlikely(test_bit_acquire(B_READING, &b->state))) { + cache_put_and_wake(c, b); return NULL; + } - b->hold_count++; - __relink_lru(b, test_bit(B_DIRTY, &b->state) || - test_bit(B_WRITING, &b->state)); return b; } @@ -1072,20 +1853,53 @@ static void read_endio(struct dm_buffer *b, blk_status_t status) * and uses dm_bufio_mark_buffer_dirty to write new data back). */ static void *new_read(struct dm_bufio_client *c, sector_t block, - enum new_flag nf, struct dm_buffer **bp) + enum new_flag nf, struct dm_buffer **bp, + unsigned short ioprio) { - int need_submit; + int need_submit = 0; struct dm_buffer *b; LIST_HEAD(write_list); - dm_bufio_lock(c); - b = __bufio_new(c, block, nf, &need_submit, &write_list); + *bp = NULL; + + /* + * Fast path, hopefully the block is already in the cache. No need + * to get the client lock for this. + */ + b = cache_get(&c->cache, block); + if (b) { + if (nf == NF_PREFETCH) { + cache_put_and_wake(c, b); + return NULL; + } + + /* + * Note: it is essential that we don't wait for the buffer to be + * read if dm_bufio_get function is used. Both dm_bufio_get and + * dm_bufio_prefetch can be used in the driver request routine. + * If the user called both dm_bufio_prefetch and dm_bufio_get on + * the same buffer, it would deadlock if we waited. + */ + if (nf == NF_GET && unlikely(test_bit_acquire(B_READING, &b->state))) { + cache_put_and_wake(c, b); + return NULL; + } + } + + if (!b) { + if (nf == NF_GET) + return NULL; + + dm_bufio_lock(c); + b = __bufio_new(c, block, nf, &need_submit, &write_list); + dm_bufio_unlock(c); + } + #ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING - if (b && b->hold_count == 1) + if (b && (atomic_read(&b->hold_count) == 1)) buffer_record_stack(b); #endif - dm_bufio_unlock(c); __flush_write_list(&write_list); @@ -1093,9 +1907,10 @@ static void *new_read(struct dm_bufio_client *c, sector_t block, return NULL; if (need_submit) - submit_io(b, REQ_OP_READ, read_endio); + submit_io(b, REQ_OP_READ, ioprio, read_endio); - wait_on_bit_io(&b->state, B_READING, TASK_UNINTERRUPTIBLE); + if (nf != NF_GET) /* we already tested this condition above */ + wait_on_bit_io(&b->state, B_READING, TASK_UNINTERRUPTIBLE); if (b->read_error) { int error = blk_status_to_errno(b->read_error); @@ -1113,43 +1928,68 @@ static void *new_read(struct dm_bufio_client *c, sector_t block, void *dm_bufio_get(struct dm_bufio_client *c, sector_t block, struct dm_buffer **bp) { - return new_read(c, block, NF_GET, bp); + return new_read(c, block, NF_GET, bp, IOPRIO_DEFAULT); } EXPORT_SYMBOL_GPL(dm_bufio_get); +static void *__dm_bufio_read(struct dm_bufio_client *c, sector_t block, + struct dm_buffer **bp, unsigned short ioprio) +{ + if (WARN_ON_ONCE(dm_bufio_in_request())) + return ERR_PTR(-EINVAL); + + return new_read(c, block, NF_READ, bp, ioprio); +} + void *dm_bufio_read(struct dm_bufio_client *c, sector_t block, struct dm_buffer **bp) { - BUG_ON(dm_bufio_in_request()); - - return new_read(c, block, NF_READ, bp); + return __dm_bufio_read(c, block, bp, IOPRIO_DEFAULT); } EXPORT_SYMBOL_GPL(dm_bufio_read); +void *dm_bufio_read_with_ioprio(struct dm_bufio_client *c, sector_t block, + struct dm_buffer **bp, unsigned short ioprio) +{ + return __dm_bufio_read(c, block, bp, ioprio); +} +EXPORT_SYMBOL_GPL(dm_bufio_read_with_ioprio); + void *dm_bufio_new(struct dm_bufio_client *c, sector_t block, struct dm_buffer **bp) { - BUG_ON(dm_bufio_in_request()); + if (WARN_ON_ONCE(dm_bufio_in_request())) + return ERR_PTR(-EINVAL); - return new_read(c, block, NF_FRESH, bp); + return new_read(c, block, NF_FRESH, bp, IOPRIO_DEFAULT); } EXPORT_SYMBOL_GPL(dm_bufio_new); -void dm_bufio_prefetch(struct dm_bufio_client *c, - sector_t block, unsigned n_blocks) +static void __dm_bufio_prefetch(struct dm_bufio_client *c, + sector_t block, unsigned int n_blocks, + unsigned short ioprio) { struct blk_plug plug; LIST_HEAD(write_list); - BUG_ON(dm_bufio_in_request()); + if (WARN_ON_ONCE(dm_bufio_in_request())) + return; /* should never happen */ blk_start_plug(&plug); - dm_bufio_lock(c); for (; n_blocks--; block++) { int need_submit; struct dm_buffer *b; + + b = cache_get(&c->cache, block); + if (b) { + /* already in cache */ + cache_put_and_wake(c, b); + continue; + } + + dm_bufio_lock(c); b = __bufio_new(c, block, NF_PREFETCH, &need_submit, &write_list); if (unlikely(!list_empty(&write_list))) { @@ -1163,7 +2003,7 @@ void dm_bufio_prefetch(struct dm_bufio_client *c, dm_bufio_unlock(c); if (need_submit) - submit_io(b, REQ_OP_READ, read_endio); + submit_io(b, REQ_OP_READ, ioprio, read_endio); dm_bufio_release(b); cond_resched(); @@ -1172,47 +2012,57 @@ void dm_bufio_prefetch(struct dm_bufio_client *c, goto flush_plug; dm_bufio_lock(c); } + dm_bufio_unlock(c); } - dm_bufio_unlock(c); - flush_plug: blk_finish_plug(&plug); } + +void dm_bufio_prefetch(struct dm_bufio_client *c, sector_t block, unsigned int n_blocks) +{ + return __dm_bufio_prefetch(c, block, n_blocks, IOPRIO_DEFAULT); +} EXPORT_SYMBOL_GPL(dm_bufio_prefetch); +void dm_bufio_prefetch_with_ioprio(struct dm_bufio_client *c, sector_t block, + unsigned int n_blocks, unsigned short ioprio) +{ + return __dm_bufio_prefetch(c, block, n_blocks, ioprio); +} +EXPORT_SYMBOL_GPL(dm_bufio_prefetch_with_ioprio); + void dm_bufio_release(struct dm_buffer *b) { struct dm_bufio_client *c = b->c; - dm_bufio_lock(c); - - BUG_ON(!b->hold_count); - - b->hold_count--; - if (!b->hold_count) { - wake_up(&c->free_buffer_wait); + /* + * If there were errors on the buffer, and the buffer is not + * to be written, free the buffer. There is no point in caching + * invalid buffer. + */ + if ((b->read_error || b->write_error) && + !test_bit_acquire(B_READING, &b->state) && + !test_bit(B_WRITING, &b->state) && + !test_bit(B_DIRTY, &b->state)) { + dm_bufio_lock(c); - /* - * If there were errors on the buffer, and the buffer is not - * to be written, free the buffer. There is no point in caching - * invalid buffer. - */ - if ((b->read_error || b->write_error) && - !test_bit(B_READING, &b->state) && - !test_bit(B_WRITING, &b->state) && - !test_bit(B_DIRTY, &b->state)) { - __unlink_buffer(b); + /* cache remove can fail if there are other holders */ + if (cache_remove(&c->cache, b)) { __free_buffer_wake(b); + dm_bufio_unlock(c); + return; } + + dm_bufio_unlock(c); } - dm_bufio_unlock(c); + cache_put_and_wake(c, b); } EXPORT_SYMBOL_GPL(dm_bufio_release); void dm_bufio_mark_partial_buffer_dirty(struct dm_buffer *b, - unsigned start, unsigned end) + unsigned int start, unsigned int end) { struct dm_bufio_client *c = b->c; @@ -1226,7 +2076,7 @@ void dm_bufio_mark_partial_buffer_dirty(struct dm_buffer *b, if (!test_and_set_bit(B_DIRTY, &b->state)) { b->dirty_start = start; b->dirty_end = end; - __relink_lru(b, LIST_DIRTY); + cache_mark(&c->cache, b, LIST_DIRTY); } else { if (start < b->dirty_start) b->dirty_start = start; @@ -1248,7 +2098,8 @@ void dm_bufio_write_dirty_buffers_async(struct dm_bufio_client *c) { LIST_HEAD(write_list); - BUG_ON(dm_bufio_in_request()); + if (WARN_ON_ONCE(dm_bufio_in_request())) + return; /* should never happen */ dm_bufio_lock(c); __write_dirty_buffers_async(c, 0, &write_list); @@ -1264,11 +2115,19 @@ EXPORT_SYMBOL_GPL(dm_bufio_write_dirty_buffers_async); * * Finally, we flush hardware disk cache. */ +static bool is_writing(struct lru_entry *e, void *context) +{ + struct dm_buffer *b = le_to_buffer(e); + + return test_bit(B_WRITING, &b->state); +} + int dm_bufio_write_dirty_buffers(struct dm_bufio_client *c) { int a, f; - unsigned long buffers_processed = 0; - struct dm_buffer *b, *tmp; + unsigned long nr_buffers; + struct lru_entry *e; + struct lru_iter it; LIST_HEAD(write_list); @@ -1278,52 +2137,32 @@ int dm_bufio_write_dirty_buffers(struct dm_bufio_client *c) __flush_write_list(&write_list); dm_bufio_lock(c); -again: - list_for_each_entry_safe_reverse(b, tmp, &c->lru[LIST_DIRTY], lru_list) { - int dropped_lock = 0; - - if (buffers_processed < c->n_buffers[LIST_DIRTY]) - buffers_processed++; + nr_buffers = cache_count(&c->cache, LIST_DIRTY); + lru_iter_begin(&c->cache.lru[LIST_DIRTY], &it); + while ((e = lru_iter_next(&it, is_writing, c))) { + struct dm_buffer *b = le_to_buffer(e); + __cache_inc_buffer(b); BUG_ON(test_bit(B_READING, &b->state)); - if (test_bit(B_WRITING, &b->state)) { - if (buffers_processed < c->n_buffers[LIST_DIRTY]) { - dropped_lock = 1; - b->hold_count++; - dm_bufio_unlock(c); - wait_on_bit_io(&b->state, B_WRITING, - TASK_UNINTERRUPTIBLE); - dm_bufio_lock(c); - b->hold_count--; - } else - wait_on_bit_io(&b->state, B_WRITING, - TASK_UNINTERRUPTIBLE); + if (nr_buffers) { + nr_buffers--; + dm_bufio_unlock(c); + wait_on_bit_io(&b->state, B_WRITING, TASK_UNINTERRUPTIBLE); + dm_bufio_lock(c); + } else { + wait_on_bit_io(&b->state, B_WRITING, TASK_UNINTERRUPTIBLE); } - if (!test_bit(B_DIRTY, &b->state) && - !test_bit(B_WRITING, &b->state)) - __relink_lru(b, LIST_CLEAN); + if (!test_bit(B_DIRTY, &b->state) && !test_bit(B_WRITING, &b->state)) + cache_mark(&c->cache, b, LIST_CLEAN); - cond_resched(); + cache_put_and_wake(c, b); - /* - * If we dropped the lock, the list is no longer consistent, - * so we must restart the search. - * - * In the most common case, the buffer just processed is - * relinked to the clean list, so we won't loop scanning the - * same buffer again and again. - * - * This may livelock if there is another thread simultaneously - * dirtying buffers, so we count the number of buffers walked - * and if it exceeds the total number of buffers, it means that - * someone is doing some writes simultaneously with us. In - * this case, stop, dropping the lock. - */ - if (dropped_lock) - goto again; + cond_resched(); } + lru_iter_end(&it); + wake_up(&c->free_buffer_wait); dm_bufio_unlock(c); @@ -1342,8 +2181,7 @@ EXPORT_SYMBOL_GPL(dm_bufio_write_dirty_buffers); int dm_bufio_issue_flush(struct dm_bufio_client *c) { struct dm_io_request io_req = { - .bi_op = REQ_OP_WRITE, - .bi_op_flags = REQ_PREFLUSH | REQ_SYNC, + .bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC, .mem.type = DM_IO_KMEM, .mem.ptr.addr = NULL, .client = c->dm_io, @@ -1354,9 +2192,10 @@ int dm_bufio_issue_flush(struct dm_bufio_client *c) .count = 0, }; - BUG_ON(dm_bufio_in_request()); + if (WARN_ON_ONCE(dm_bufio_in_request())) + return -EINVAL; - return dm_io(&io_req, 1, &io_reg, NULL); + return dm_io(&io_req, 1, &io_reg, NULL, IOPRIO_DEFAULT); } EXPORT_SYMBOL_GPL(dm_bufio_issue_flush); @@ -1366,8 +2205,7 @@ EXPORT_SYMBOL_GPL(dm_bufio_issue_flush); int dm_bufio_issue_discard(struct dm_bufio_client *c, sector_t block, sector_t count) { struct dm_io_request io_req = { - .bi_op = REQ_OP_DISCARD, - .bi_op_flags = REQ_SYNC, + .bi_opf = REQ_OP_DISCARD | REQ_SYNC, .mem.type = DM_IO_KMEM, .mem.ptr.addr = NULL, .client = c->dm_io, @@ -1378,93 +2216,27 @@ int dm_bufio_issue_discard(struct dm_bufio_client *c, sector_t block, sector_t c .count = block_to_sector(c, count), }; - BUG_ON(dm_bufio_in_request()); + if (WARN_ON_ONCE(dm_bufio_in_request())) + return -EINVAL; /* discards are optional */ - return dm_io(&io_req, 1, &io_reg, NULL); + return dm_io(&io_req, 1, &io_reg, NULL, IOPRIO_DEFAULT); } EXPORT_SYMBOL_GPL(dm_bufio_issue_discard); -/* - * We first delete any other buffer that may be at that new location. - * - * Then, we write the buffer to the original location if it was dirty. - * - * Then, if we are the only one who is holding the buffer, relink the buffer - * in the buffer tree for the new location. - * - * If there was someone else holding the buffer, we write it to the new - * location but not relink it, because that other user needs to have the buffer - * at the same place. - */ -void dm_bufio_release_move(struct dm_buffer *b, sector_t new_block) +static void forget_buffer(struct dm_bufio_client *c, sector_t block) { - struct dm_bufio_client *c = b->c; - struct dm_buffer *new; - - BUG_ON(dm_bufio_in_request()); - - dm_bufio_lock(c); + struct dm_buffer *b; -retry: - new = __find(c, new_block); - if (new) { - if (new->hold_count) { - __wait_for_free_buffer(c); - goto retry; + b = cache_get(&c->cache, block); + if (b) { + if (likely(!smp_load_acquire(&b->state))) { + if (cache_remove(&c->cache, b)) + __free_buffer_wake(b); + else + cache_put_and_wake(c, b); + } else { + cache_put_and_wake(c, b); } - - /* - * FIXME: Is there any point waiting for a write that's going - * to be overwritten in a bit? - */ - __make_buffer_clean(new); - __unlink_buffer(new); - __free_buffer_wake(new); - } - - BUG_ON(!b->hold_count); - BUG_ON(test_bit(B_READING, &b->state)); - - __write_dirty_buffer(b, NULL); - if (b->hold_count == 1) { - wait_on_bit_io(&b->state, B_WRITING, - TASK_UNINTERRUPTIBLE); - set_bit(B_DIRTY, &b->state); - b->dirty_start = 0; - b->dirty_end = c->block_size; - __unlink_buffer(b); - __link_buffer(b, new_block, LIST_DIRTY); - } else { - sector_t old_block; - wait_on_bit_lock_io(&b->state, B_WRITING, - TASK_UNINTERRUPTIBLE); - /* - * Relink buffer to "new_block" so that write_callback - * sees "new_block" as a block number. - * After the write, link the buffer back to old_block. - * All this must be done in bufio lock, so that block number - * change isn't visible to other threads. - */ - old_block = b->block; - __unlink_buffer(b); - __link_buffer(b, new_block, b->list_mode); - submit_io(b, REQ_OP_WRITE, write_endio); - wait_on_bit_io(&b->state, B_WRITING, - TASK_UNINTERRUPTIBLE); - __unlink_buffer(b); - __link_buffer(b, old_block, b->list_mode); - } - - dm_bufio_unlock(c); - dm_bufio_release(b); -} -EXPORT_SYMBOL_GPL(dm_bufio_release_move); - -static void forget_buffer_locked(struct dm_buffer *b) -{ - if (likely(!b->hold_count) && likely(!b->state)) { - __unlink_buffer(b); - __free_buffer_wake(b); } } @@ -1476,48 +2248,32 @@ static void forget_buffer_locked(struct dm_buffer *b) */ void dm_bufio_forget(struct dm_bufio_client *c, sector_t block) { - struct dm_buffer *b; - dm_bufio_lock(c); - - b = __find(c, block); - if (b) - forget_buffer_locked(b); - + forget_buffer(c, block); dm_bufio_unlock(c); } EXPORT_SYMBOL_GPL(dm_bufio_forget); -void dm_bufio_forget_buffers(struct dm_bufio_client *c, sector_t block, sector_t n_blocks) +static enum evict_result idle(struct dm_buffer *b, void *context) { - struct dm_buffer *b; - sector_t end_block = block + n_blocks; - - while (block < end_block) { - dm_bufio_lock(c); - - b = __find_next(c, block); - if (b) { - block = b->block + 1; - forget_buffer_locked(b); - } - - dm_bufio_unlock(c); - - if (!b) - break; - } + return b->state ? ER_DONT_EVICT : ER_EVICT; +} +void dm_bufio_forget_buffers(struct dm_bufio_client *c, sector_t block, sector_t n_blocks) +{ + dm_bufio_lock(c); + cache_remove_range(&c->cache, block, block + n_blocks, idle, __free_buffer_wake); + dm_bufio_unlock(c); } EXPORT_SYMBOL_GPL(dm_bufio_forget_buffers); -void dm_bufio_set_minimum_buffers(struct dm_bufio_client *c, unsigned n) +void dm_bufio_set_minimum_buffers(struct dm_bufio_client *c, unsigned int n) { c->minimum_buffers = n; } EXPORT_SYMBOL_GPL(dm_bufio_set_minimum_buffers); -unsigned dm_bufio_get_block_size(struct dm_bufio_client *c) +unsigned int dm_bufio_get_block_size(struct dm_bufio_client *c) { return c->block_size; } @@ -1525,7 +2281,8 @@ EXPORT_SYMBOL_GPL(dm_bufio_get_block_size); sector_t dm_bufio_get_device_size(struct dm_bufio_client *c) { - sector_t s = i_size_read(c->bdev->bd_inode) >> SECTOR_SHIFT; + sector_t s = bdev_nr_sectors(c->bdev); + if (s >= c->start) s -= c->start; else @@ -1568,13 +2325,29 @@ struct dm_bufio_client *dm_bufio_get_client(struct dm_buffer *b) } EXPORT_SYMBOL_GPL(dm_bufio_get_client); +static enum it_action warn_leak(struct dm_buffer *b, void *context) +{ + bool *warned = context; + + WARN_ON(!(*warned)); + *warned = true; + DMERR("leaked buffer %llx, hold count %u, list %d", + (unsigned long long)b->block, atomic_read(&b->hold_count), b->list_mode); +#ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING + stack_trace_print(b->stack_entries, b->stack_len, 1); + /* mark unclaimed to avoid WARN_ON at end of drop_buffers() */ + atomic_set(&b->hold_count, 0); +#endif + return IT_NEXT; +} + static void drop_buffers(struct dm_bufio_client *c) { - struct dm_buffer *b; int i; - bool warned = false; + struct dm_buffer *b; - BUG_ON(dm_bufio_in_request()); + if (WARN_ON(dm_bufio_in_request())) + return; /* should never happen */ /* * An optimization so that the buffers are not written one-by-one. @@ -1586,18 +2359,11 @@ static void drop_buffers(struct dm_bufio_client *c) while ((b = __get_unclaimed_buffer(c))) __free_buffer_wake(b); - for (i = 0; i < LIST_SIZE; i++) - list_for_each_entry(b, &c->lru[i], lru_list) { - WARN_ON(!warned); - warned = true; - DMERR("leaked buffer %llx, hold count %u, list %d", - (unsigned long long)b->block, b->hold_count, i); -#ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING - stack_trace_print(b->stack_entries, b->stack_len, 1); - /* mark unclaimed to avoid BUG_ON below */ - b->hold_count = 0; -#endif - } + for (i = 0; i < LIST_SIZE; i++) { + bool warned = false; + + cache_iterate(&c->cache, i, warn_leak, &warned); + } #ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING while ((b = __get_unclaimed_buffer(c))) @@ -1605,68 +2371,54 @@ static void drop_buffers(struct dm_bufio_client *c) #endif for (i = 0; i < LIST_SIZE; i++) - BUG_ON(!list_empty(&c->lru[i])); + WARN_ON(cache_count(&c->cache, i)); dm_bufio_unlock(c); } -/* - * We may not be able to evict this buffer if IO pending or the client - * is still using it. Caller is expected to know buffer is too old. - * - * And if GFP_NOFS is used, we must not do any I/O because we hold - * dm_bufio_clients_lock and we would risk deadlock if the I/O gets - * rerouted to different bufio client. - */ -static bool __try_evict_buffer(struct dm_buffer *b, gfp_t gfp) -{ - if (!(gfp & __GFP_FS)) { - if (test_bit(B_READING, &b->state) || - test_bit(B_WRITING, &b->state) || - test_bit(B_DIRTY, &b->state)) - return false; - } - - if (b->hold_count) - return false; - - __make_buffer_clean(b); - __unlink_buffer(b); - __free_buffer_wake(b); - - return true; -} - static unsigned long get_retain_buffers(struct dm_bufio_client *c) { unsigned long retain_bytes = READ_ONCE(dm_bufio_retain_bytes); + if (likely(c->sectors_per_block_bits >= 0)) retain_bytes >>= c->sectors_per_block_bits + SECTOR_SHIFT; else retain_bytes /= c->block_size; + return retain_bytes; } static void __scan(struct dm_bufio_client *c) { int l; - struct dm_buffer *b, *tmp; + struct dm_buffer *b; unsigned long freed = 0; - unsigned long count = c->n_buffers[LIST_CLEAN] + - c->n_buffers[LIST_DIRTY]; unsigned long retain_target = get_retain_buffers(c); + unsigned long count = cache_total(&c->cache); for (l = 0; l < LIST_SIZE; l++) { - list_for_each_entry_safe_reverse(b, tmp, &c->lru[l], lru_list) { + while (true) { if (count - freed <= retain_target) atomic_long_set(&c->need_shrink, 0); if (!atomic_long_read(&c->need_shrink)) - return; - if (__try_evict_buffer(b, GFP_KERNEL)) { - atomic_long_dec(&c->need_shrink); - freed++; + break; + + b = cache_evict(&c->cache, l, + l == LIST_CLEAN ? is_clean : is_dirty, c); + if (!b) + break; + + __make_buffer_clean(b); + __free_buffer_wake(b); + + atomic_long_dec(&c->need_shrink); + freed++; + + if (unlikely(freed % SCAN_RESCHED_CYCLE == 0)) { + dm_bufio_unlock(c); + cond_resched(); + dm_bufio_lock(c); } - cond_resched(); } } } @@ -1684,7 +2436,7 @@ static unsigned long dm_bufio_shrink_scan(struct shrinker *shrink, struct shrink { struct dm_bufio_client *c; - c = container_of(shrink, struct dm_bufio_client, shrinker); + c = shrink->private_data; atomic_long_add(sc->nr_to_scan, &c->need_shrink); queue_work(dm_bufio_wq, &c->shrink_work); @@ -1693,9 +2445,8 @@ static unsigned long dm_bufio_shrink_scan(struct shrinker *shrink, struct shrink static unsigned long dm_bufio_shrink_count(struct shrinker *shrink, struct shrink_control *sc) { - struct dm_bufio_client *c = container_of(shrink, struct dm_bufio_client, shrinker); - unsigned long count = READ_ONCE(c->n_buffers[LIST_CLEAN]) + - READ_ONCE(c->n_buffers[LIST_DIRTY]); + struct dm_bufio_client *c = shrink->private_data; + unsigned long count = cache_total(&c->cache); unsigned long retain_target = get_retain_buffers(c); unsigned long queued_for_cleanup = atomic_long_read(&c->need_shrink); @@ -1715,15 +2466,17 @@ static unsigned long dm_bufio_shrink_count(struct shrinker *shrink, struct shrin /* * Create the buffering interface */ -struct dm_bufio_client *dm_bufio_client_create(struct block_device *bdev, unsigned block_size, - unsigned reserved_buffers, unsigned aux_size, +struct dm_bufio_client *dm_bufio_client_create(struct block_device *bdev, unsigned int block_size, + unsigned int reserved_buffers, unsigned int aux_size, void (*alloc_callback)(struct dm_buffer *), - void (*write_callback)(struct dm_buffer *)) + void (*write_callback)(struct dm_buffer *), + unsigned int flags) { int r; + unsigned int num_locks; struct dm_bufio_client *c; - unsigned i; - char slab_name[27]; + char slab_name[64]; + static atomic_t seqno = ATOMIC_INIT(0); if (!block_size || block_size & ((1 << SECTOR_SHIFT) - 1)) { DMERR("%s: block size not specified or is not multiple of 512b", __func__); @@ -1731,12 +2484,13 @@ struct dm_bufio_client *dm_bufio_client_create(struct block_device *bdev, unsign goto bad_client; } - c = kzalloc(sizeof(*c), GFP_KERNEL); + num_locks = dm_num_hash_locks(); + c = kzalloc(sizeof(*c) + (num_locks * sizeof(struct buffer_tree)), GFP_KERNEL); if (!c) { r = -ENOMEM; goto bad_client; } - c->buffer_tree = RB_ROOT; + cache_init(&c->cache, num_locks, (flags & DM_BUFIO_CLIENT_NO_SLEEP) != 0); c->bdev = bdev; c->block_size = block_size; @@ -1748,12 +2502,13 @@ struct dm_bufio_client *dm_bufio_client_create(struct block_device *bdev, unsign c->alloc_callback = alloc_callback; c->write_callback = write_callback; - for (i = 0; i < LIST_SIZE; i++) { - INIT_LIST_HEAD(&c->lru[i]); - c->n_buffers[i] = 0; + if (flags & DM_BUFIO_CLIENT_NO_SLEEP) { + c->no_sleep = true; + static_branch_inc(&no_sleep_enabled); } mutex_init(&c->lock); + spin_lock_init(&c->spinlock); INIT_LIST_HEAD(&c->reserved_buffers); c->need_reserved_buffers = reserved_buffers; @@ -1768,10 +2523,11 @@ struct dm_bufio_client *dm_bufio_client_create(struct block_device *bdev, unsign goto bad_dm_io; } - if (block_size <= KMALLOC_MAX_SIZE && - (block_size < PAGE_SIZE || !is_power_of_2(block_size))) { - unsigned align = min(1U << __ffs(block_size), (unsigned)PAGE_SIZE); - snprintf(slab_name, sizeof slab_name, "dm_bufio_cache-%u", block_size); + if (block_size <= KMALLOC_MAX_SIZE && !is_power_of_2(block_size)) { + unsigned int align = min(1U << __ffs(block_size), (unsigned int)PAGE_SIZE); + + snprintf(slab_name, sizeof(slab_name), "dm_bufio_cache-%u-%u", + block_size, atomic_inc_return(&seqno)); c->slab_cache = kmem_cache_create(slab_name, block_size, align, SLAB_RECLAIM_ACCOUNT, NULL); if (!c->slab_cache) { @@ -1780,9 +2536,11 @@ struct dm_bufio_client *dm_bufio_client_create(struct block_device *bdev, unsign } } if (aux_size) - snprintf(slab_name, sizeof slab_name, "dm_bufio_buffer-%u", aux_size); + snprintf(slab_name, sizeof(slab_name), "dm_bufio_buffer-%u-%u", + aux_size, atomic_inc_return(&seqno)); else - snprintf(slab_name, sizeof slab_name, "dm_bufio_buffer"); + snprintf(slab_name, sizeof(slab_name), "dm_bufio_buffer-%u", + atomic_inc_return(&seqno)); c->slab_buffer = kmem_cache_create(slab_name, sizeof(struct dm_buffer) + aux_size, 0, SLAB_RECLAIM_ACCOUNT, NULL); if (!c->slab_buffer) { @@ -1803,13 +2561,20 @@ struct dm_bufio_client *dm_bufio_client_create(struct block_device *bdev, unsign INIT_WORK(&c->shrink_work, shrink_work); atomic_long_set(&c->need_shrink, 0); - c->shrinker.count_objects = dm_bufio_shrink_count; - c->shrinker.scan_objects = dm_bufio_shrink_scan; - c->shrinker.seeks = 1; - c->shrinker.batch = 0; - r = register_shrinker(&c->shrinker); - if (r) + c->shrinker = shrinker_alloc(0, "dm-bufio:(%u:%u)", + MAJOR(bdev->bd_dev), MINOR(bdev->bd_dev)); + if (!c->shrinker) { + r = -ENOMEM; goto bad; + } + + c->shrinker->count_objects = dm_bufio_shrink_count; + c->shrinker->scan_objects = dm_bufio_shrink_scan; + c->shrinker->seeks = 1; + c->shrinker->batch = 0; + c->shrinker->private_data = c; + + shrinker_register(c->shrinker); mutex_lock(&dm_bufio_clients_lock); dm_bufio_client_count++; @@ -1821,9 +2586,9 @@ struct dm_bufio_client *dm_bufio_client_create(struct block_device *bdev, unsign bad: while (!list_empty(&c->reserved_buffers)) { - struct dm_buffer *b = list_entry(c->reserved_buffers.next, - struct dm_buffer, lru_list); - list_del(&b->lru_list); + struct dm_buffer *b = list_to_buffer(c->reserved_buffers.next); + + list_del(&b->lru.list); free_buffer(b); } kmem_cache_destroy(c->slab_cache); @@ -1831,6 +2596,8 @@ bad: dm_io_client_destroy(c->dm_io); bad_dm_io: mutex_destroy(&c->lock); + if (c->no_sleep) + static_branch_dec(&no_sleep_enabled); kfree(c); bad_client: return ERR_PTR(r); @@ -1843,11 +2610,11 @@ EXPORT_SYMBOL_GPL(dm_bufio_client_create); */ void dm_bufio_client_destroy(struct dm_bufio_client *c) { - unsigned i; + unsigned int i; drop_buffers(c); - unregister_shrinker(&c->shrinker); + shrinker_free(c->shrinker); flush_work(&c->shrink_work); mutex_lock(&dm_bufio_clients_lock); @@ -1858,179 +2625,180 @@ void dm_bufio_client_destroy(struct dm_bufio_client *c) mutex_unlock(&dm_bufio_clients_lock); - BUG_ON(!RB_EMPTY_ROOT(&c->buffer_tree)); - BUG_ON(c->need_reserved_buffers); + WARN_ON(c->need_reserved_buffers); while (!list_empty(&c->reserved_buffers)) { - struct dm_buffer *b = list_entry(c->reserved_buffers.next, - struct dm_buffer, lru_list); - list_del(&b->lru_list); + struct dm_buffer *b = list_to_buffer(c->reserved_buffers.next); + + list_del(&b->lru.list); free_buffer(b); } for (i = 0; i < LIST_SIZE; i++) - if (c->n_buffers[i]) - DMERR("leaked buffer count %d: %ld", i, c->n_buffers[i]); + if (cache_count(&c->cache, i)) + DMERR("leaked buffer count %d: %lu", i, cache_count(&c->cache, i)); for (i = 0; i < LIST_SIZE; i++) - BUG_ON(c->n_buffers[i]); + WARN_ON(cache_count(&c->cache, i)); + cache_destroy(&c->cache); kmem_cache_destroy(c->slab_cache); kmem_cache_destroy(c->slab_buffer); dm_io_client_destroy(c->dm_io); mutex_destroy(&c->lock); + if (c->no_sleep) + static_branch_dec(&no_sleep_enabled); kfree(c); } EXPORT_SYMBOL_GPL(dm_bufio_client_destroy); +void dm_bufio_client_reset(struct dm_bufio_client *c) +{ + drop_buffers(c); + flush_work(&c->shrink_work); +} +EXPORT_SYMBOL_GPL(dm_bufio_client_reset); + void dm_bufio_set_sector_offset(struct dm_bufio_client *c, sector_t start) { c->start = start; } EXPORT_SYMBOL_GPL(dm_bufio_set_sector_offset); -static unsigned get_max_age_hz(void) -{ - unsigned max_age = READ_ONCE(dm_bufio_max_age); +/*--------------------------------------------------------------*/ - if (max_age > UINT_MAX / HZ) - max_age = UINT_MAX / HZ; +/* + * Global cleanup tries to evict the oldest buffers from across _all_ + * the clients. It does this by repeatedly evicting a few buffers from + * the client that holds the oldest buffer. It's approximate, but hopefully + * good enough. + */ +static struct dm_bufio_client *__pop_client(void) +{ + struct list_head *h; - return max_age * HZ; -} + if (list_empty(&dm_bufio_all_clients)) + return NULL; -static bool older_than(struct dm_buffer *b, unsigned long age_hz) -{ - return time_after_eq(jiffies, b->last_accessed + age_hz); + h = dm_bufio_all_clients.next; + list_del(h); + return container_of(h, struct dm_bufio_client, client_list); } -static void __evict_old_buffers(struct dm_bufio_client *c, unsigned long age_hz) +/* + * Inserts the client in the global client list based on its + * 'oldest_buffer' field. + */ +static void __insert_client(struct dm_bufio_client *new_client) { - struct dm_buffer *b, *tmp; - unsigned long retain_target = get_retain_buffers(c); - unsigned long count; - LIST_HEAD(write_list); - - dm_bufio_lock(c); - - __check_watermark(c, &write_list); - if (unlikely(!list_empty(&write_list))) { - dm_bufio_unlock(c); - __flush_write_list(&write_list); - dm_bufio_lock(c); - } - - count = c->n_buffers[LIST_CLEAN] + c->n_buffers[LIST_DIRTY]; - list_for_each_entry_safe_reverse(b, tmp, &c->lru[LIST_CLEAN], lru_list) { - if (count <= retain_target) - break; + struct dm_bufio_client *c; + struct list_head *h = dm_bufio_all_clients.next; - if (!older_than(b, age_hz)) + while (h != &dm_bufio_all_clients) { + c = container_of(h, struct dm_bufio_client, client_list); + if (time_after_eq(c->oldest_buffer, new_client->oldest_buffer)) break; + h = h->next; + } - if (__try_evict_buffer(b, 0)) - count--; + list_add_tail(&new_client->client_list, h); +} - cond_resched(); +static enum evict_result select_for_evict(struct dm_buffer *b, void *context) +{ + /* In no-sleep mode, we cannot wait on IO. */ + if (static_branch_unlikely(&no_sleep_enabled) && b->c->no_sleep) { + if (test_bit_acquire(B_READING, &b->state) || + test_bit(B_WRITING, &b->state) || + test_bit(B_DIRTY, &b->state)) + return ER_DONT_EVICT; } - - dm_bufio_unlock(c); + return ER_EVICT; } -static void do_global_cleanup(struct work_struct *w) +static unsigned long __evict_a_few(unsigned long nr_buffers) { - struct dm_bufio_client *locked_client = NULL; - struct dm_bufio_client *current_client; + struct dm_bufio_client *c; + unsigned long oldest_buffer = jiffies; + unsigned long last_accessed; + unsigned long count; struct dm_buffer *b; - unsigned spinlock_hold_count; - unsigned long threshold = dm_bufio_cache_size - - dm_bufio_cache_size / DM_BUFIO_LOW_WATERMARK_RATIO; - unsigned long loops = global_num * 2; - mutex_lock(&dm_bufio_clients_lock); - - while (1) { - cond_resched(); + c = __pop_client(); + if (!c) + return 0; - spin_lock(&global_spinlock); - if (unlikely(dm_bufio_current_allocated <= threshold)) - break; + dm_bufio_lock(c); - spinlock_hold_count = 0; -get_next: - if (!loops--) - break; - if (unlikely(list_empty(&global_queue))) + for (count = 0; count < nr_buffers; count++) { + b = cache_evict(&c->cache, LIST_CLEAN, select_for_evict, NULL); + if (!b) break; - b = list_entry(global_queue.prev, struct dm_buffer, global_list); - - if (b->accessed) { - b->accessed = 0; - list_move(&b->global_list, &global_queue); - if (likely(++spinlock_hold_count < 16)) - goto get_next; - spin_unlock(&global_spinlock); - continue; - } - - current_client = b->c; - if (unlikely(current_client != locked_client)) { - if (locked_client) - dm_bufio_unlock(locked_client); - if (!dm_bufio_trylock(current_client)) { - spin_unlock(&global_spinlock); - dm_bufio_lock(current_client); - locked_client = current_client; - continue; - } - - locked_client = current_client; - } + last_accessed = READ_ONCE(b->last_accessed); + if (time_after_eq(oldest_buffer, last_accessed)) + oldest_buffer = last_accessed; - spin_unlock(&global_spinlock); + __make_buffer_clean(b); + __free_buffer_wake(b); - if (unlikely(!__try_evict_buffer(b, GFP_KERNEL))) { - spin_lock(&global_spinlock); - list_move(&b->global_list, &global_queue); - spin_unlock(&global_spinlock); + if (need_resched()) { + dm_bufio_unlock(c); + cond_resched(); + dm_bufio_lock(c); } } - spin_unlock(&global_spinlock); + dm_bufio_unlock(c); - if (locked_client) - dm_bufio_unlock(locked_client); + if (count) + c->oldest_buffer = oldest_buffer; + __insert_client(c); - mutex_unlock(&dm_bufio_clients_lock); + return count; } -static void cleanup_old_buffers(void) +static void check_watermarks(void) { - unsigned long max_age_hz = get_max_age_hz(); + LIST_HEAD(write_list); struct dm_bufio_client *c; mutex_lock(&dm_bufio_clients_lock); + list_for_each_entry(c, &dm_bufio_all_clients, client_list) { + dm_bufio_lock(c); + __check_watermark(c, &write_list); + dm_bufio_unlock(c); + } + mutex_unlock(&dm_bufio_clients_lock); - __cache_size_refresh(); + __flush_write_list(&write_list); +} - list_for_each_entry(c, &dm_bufio_all_clients, client_list) - __evict_old_buffers(c, max_age_hz); +static void evict_old(void) +{ + unsigned long threshold = dm_bufio_cache_size - + dm_bufio_cache_size / DM_BUFIO_LOW_WATERMARK_RATIO; + mutex_lock(&dm_bufio_clients_lock); + while (dm_bufio_current_allocated > threshold) { + if (!__evict_a_few(64)) + break; + cond_resched(); + } mutex_unlock(&dm_bufio_clients_lock); } -static void work_fn(struct work_struct *w) +static void do_global_cleanup(struct work_struct *w) { - cleanup_old_buffers(); - - queue_delayed_work(dm_bufio_wq, &dm_bufio_cleanup_old_work, - DM_BUFIO_WORK_TIMER_SECS * HZ); + check_watermarks(); + evict_old(); } -/*---------------------------------------------------------------- +/* + *-------------------------------------------------------------- * Module setup - *--------------------------------------------------------------*/ + *-------------------------------------------------------------- + */ /* * This is called only once for the whole dm_bufio module. @@ -2041,6 +2809,7 @@ static int __init dm_bufio_init(void) __u64 mem; dm_bufio_allocated_kmem_cache = 0; + dm_bufio_allocated_kmalloc = 0; dm_bufio_allocated_get_free_pages = 0; dm_bufio_allocated_vmalloc = 0; dm_bufio_current_allocated = 0; @@ -2066,10 +2835,7 @@ static int __init dm_bufio_init(void) if (!dm_bufio_wq) return -ENOMEM; - INIT_DELAYED_WORK(&dm_bufio_cleanup_old_work, work_fn); INIT_WORK(&dm_bufio_replacement_work, do_global_cleanup); - queue_delayed_work(dm_bufio_wq, &dm_bufio_cleanup_old_work, - DM_BUFIO_WORK_TIMER_SECS * HZ); return 0; } @@ -2081,8 +2847,6 @@ static void __exit dm_bufio_exit(void) { int bug = 0; - cancel_delayed_work_sync(&dm_bufio_cleanup_old_work); - flush_workqueue(dm_bufio_wq); destroy_workqueue(dm_bufio_wq); if (dm_bufio_client_count) { @@ -2109,36 +2873,39 @@ static void __exit dm_bufio_exit(void) bug = 1; } - BUG_ON(bug); + WARN_ON(bug); /* leaks are not worth crashing the system */ } module_init(dm_bufio_init) module_exit(dm_bufio_exit) -module_param_named(max_cache_size_bytes, dm_bufio_cache_size, ulong, S_IRUGO | S_IWUSR); +module_param_named(max_cache_size_bytes, dm_bufio_cache_size, ulong, 0644); MODULE_PARM_DESC(max_cache_size_bytes, "Size of metadata cache"); -module_param_named(max_age_seconds, dm_bufio_max_age, uint, S_IRUGO | S_IWUSR); -MODULE_PARM_DESC(max_age_seconds, "Max age of a buffer in seconds"); +module_param_named(max_age_seconds, dm_bufio_max_age, uint, 0644); +MODULE_PARM_DESC(max_age_seconds, "No longer does anything"); -module_param_named(retain_bytes, dm_bufio_retain_bytes, ulong, S_IRUGO | S_IWUSR); +module_param_named(retain_bytes, dm_bufio_retain_bytes, ulong, 0644); MODULE_PARM_DESC(retain_bytes, "Try to keep at least this many bytes cached in memory"); -module_param_named(peak_allocated_bytes, dm_bufio_peak_allocated, ulong, S_IRUGO | S_IWUSR); +module_param_named(peak_allocated_bytes, dm_bufio_peak_allocated, ulong, 0644); MODULE_PARM_DESC(peak_allocated_bytes, "Tracks the maximum allocated memory"); -module_param_named(allocated_kmem_cache_bytes, dm_bufio_allocated_kmem_cache, ulong, S_IRUGO); +module_param_named(allocated_kmem_cache_bytes, dm_bufio_allocated_kmem_cache, ulong, 0444); MODULE_PARM_DESC(allocated_kmem_cache_bytes, "Memory allocated with kmem_cache_alloc"); -module_param_named(allocated_get_free_pages_bytes, dm_bufio_allocated_get_free_pages, ulong, S_IRUGO); +module_param_named(allocated_kmalloc_bytes, dm_bufio_allocated_kmalloc, ulong, 0444); +MODULE_PARM_DESC(allocated_kmalloc_bytes, "Memory allocated with kmalloc_alloc"); + +module_param_named(allocated_get_free_pages_bytes, dm_bufio_allocated_get_free_pages, ulong, 0444); MODULE_PARM_DESC(allocated_get_free_pages_bytes, "Memory allocated with get_free_pages"); -module_param_named(allocated_vmalloc_bytes, dm_bufio_allocated_vmalloc, ulong, S_IRUGO); +module_param_named(allocated_vmalloc_bytes, dm_bufio_allocated_vmalloc, ulong, 0444); MODULE_PARM_DESC(allocated_vmalloc_bytes, "Memory allocated with vmalloc"); -module_param_named(current_allocated_bytes, dm_bufio_current_allocated, ulong, S_IRUGO); +module_param_named(current_allocated_bytes, dm_bufio_current_allocated, ulong, 0444); MODULE_PARM_DESC(current_allocated_bytes, "Memory currently used by the cache"); -MODULE_AUTHOR("Mikulas Patocka <dm-devel@redhat.com>"); +MODULE_AUTHOR("Mikulas Patocka <dm-devel@lists.linux.dev>"); MODULE_DESCRIPTION(DM_NAME " buffered I/O library"); MODULE_LICENSE("GPL"); |