Merge tag 'configfs-5.11' of git://git.infradead.org/users/hch/configfs
[linux-2.6-microblaze.git] / fs / f2fs / compress.c
index 1dfb126..4bcbacf 100644 (file)
 #include "node.h"
 #include <trace/events/f2fs.h>
 
+static struct kmem_cache *cic_entry_slab;
+static struct kmem_cache *dic_entry_slab;
+
+static void *page_array_alloc(struct inode *inode, int nr)
+{
+       struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
+       unsigned int size = sizeof(struct page *) * nr;
+
+       if (likely(size <= sbi->page_array_slab_size))
+               return kmem_cache_zalloc(sbi->page_array_slab, GFP_NOFS);
+       return f2fs_kzalloc(sbi, size, GFP_NOFS);
+}
+
+static void page_array_free(struct inode *inode, void *pages, int nr)
+{
+       struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
+       unsigned int size = sizeof(struct page *) * nr;
+
+       if (!pages)
+               return;
+
+       if (likely(size <= sbi->page_array_slab_size))
+               kmem_cache_free(sbi->page_array_slab, pages);
+       else
+               kfree(pages);
+}
+
 struct f2fs_compress_ops {
        int (*init_compress_ctx)(struct compress_ctx *cc);
        void (*destroy_compress_ctx)(struct compress_ctx *cc);
@@ -130,19 +157,16 @@ struct page *f2fs_compress_control_page(struct page *page)
 
 int f2fs_init_compress_ctx(struct compress_ctx *cc)
 {
-       struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode);
-
-       if (cc->nr_rpages)
+       if (cc->rpages)
                return 0;
 
-       cc->rpages = f2fs_kzalloc(sbi, sizeof(struct page *) <<
-                                       cc->log_cluster_size, GFP_NOFS);
+       cc->rpages = page_array_alloc(cc->inode, cc->cluster_size);
        return cc->rpages ? 0 : -ENOMEM;
 }
 
 void f2fs_destroy_compress_ctx(struct compress_ctx *cc)
 {
-       kfree(cc->rpages);
+       page_array_free(cc->inode, cc->rpages, cc->cluster_size);
        cc->rpages = NULL;
        cc->nr_rpages = 0;
        cc->nr_cpages = 0;
@@ -382,16 +406,17 @@ static int zstd_init_decompress_ctx(struct decompress_io_ctx *dic)
        ZSTD_DStream *stream;
        void *workspace;
        unsigned int workspace_size;
+       unsigned int max_window_size =
+                       MAX_COMPRESS_WINDOW_SIZE(dic->log_cluster_size);
 
-       workspace_size = ZSTD_DStreamWorkspaceBound(MAX_COMPRESS_WINDOW_SIZE);
+       workspace_size = ZSTD_DStreamWorkspaceBound(max_window_size);
 
        workspace = f2fs_kvmalloc(F2FS_I_SB(dic->inode),
                                        workspace_size, GFP_NOFS);
        if (!workspace)
                return -ENOMEM;
 
-       stream = ZSTD_initDStream(MAX_COMPRESS_WINDOW_SIZE,
-                                       workspace, workspace_size);
+       stream = ZSTD_initDStream(max_window_size, workspace, workspace_size);
        if (!stream) {
                printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_initDStream failed\n",
                                KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id,
@@ -554,13 +579,30 @@ static void f2fs_compress_free_page(struct page *page)
        mempool_free(page, compress_page_pool);
 }
 
+#define MAX_VMAP_RETRIES       3
+
+static void *f2fs_vmap(struct page **pages, unsigned int count)
+{
+       int i;
+       void *buf = NULL;
+
+       for (i = 0; i < MAX_VMAP_RETRIES; i++) {
+               buf = vm_map_ram(pages, count, -1);
+               if (buf)
+                       break;
+               vm_unmap_aliases();
+       }
+       return buf;
+}
+
 static int f2fs_compress_pages(struct compress_ctx *cc)
 {
-       struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode);
        struct f2fs_inode_info *fi = F2FS_I(cc->inode);
        const struct f2fs_compress_ops *cops =
                                f2fs_cops[fi->i_compress_algorithm];
-       unsigned int max_len, nr_cpages;
+       unsigned int max_len, new_nr_cpages;
+       struct page **new_cpages;
+       u32 chksum = 0;
        int i, ret;
 
        trace_f2fs_compress_pages_start(cc->inode, cc->cluster_idx,
@@ -575,8 +617,7 @@ static int f2fs_compress_pages(struct compress_ctx *cc)
        max_len = COMPRESS_HEADER_SIZE + cc->clen;
        cc->nr_cpages = DIV_ROUND_UP(max_len, PAGE_SIZE);
 
-       cc->cpages = f2fs_kzalloc(sbi, sizeof(struct page *) *
-                                       cc->nr_cpages, GFP_NOFS);
+       cc->cpages = page_array_alloc(cc->inode, cc->nr_cpages);
        if (!cc->cpages) {
                ret = -ENOMEM;
                goto destroy_compress_ctx;
@@ -590,13 +631,13 @@ static int f2fs_compress_pages(struct compress_ctx *cc)
                }
        }
 
-       cc->rbuf = vmap(cc->rpages, cc->cluster_size, VM_MAP, PAGE_KERNEL_RO);
+       cc->rbuf = f2fs_vmap(cc->rpages, cc->cluster_size);
        if (!cc->rbuf) {
                ret = -ENOMEM;
                goto out_free_cpages;
        }
 
-       cc->cbuf = vmap(cc->cpages, cc->nr_cpages, VM_MAP, PAGE_KERNEL);
+       cc->cbuf = f2fs_vmap(cc->cpages, cc->nr_cpages);
        if (!cc->cbuf) {
                ret = -ENOMEM;
                goto out_vunmap_rbuf;
@@ -615,19 +656,36 @@ static int f2fs_compress_pages(struct compress_ctx *cc)
 
        cc->cbuf->clen = cpu_to_le32(cc->clen);
 
+       if (fi->i_compress_flag & 1 << COMPRESS_CHKSUM)
+               chksum = f2fs_crc32(F2FS_I_SB(cc->inode),
+                                       cc->cbuf->cdata, cc->clen);
+       cc->cbuf->chksum = cpu_to_le32(chksum);
+
        for (i = 0; i < COMPRESS_DATA_RESERVED_SIZE; i++)
                cc->cbuf->reserved[i] = cpu_to_le32(0);
 
-       nr_cpages = DIV_ROUND_UP(cc->clen + COMPRESS_HEADER_SIZE, PAGE_SIZE);
+       new_nr_cpages = DIV_ROUND_UP(cc->clen + COMPRESS_HEADER_SIZE, PAGE_SIZE);
+
+       /* Now we're going to cut unnecessary tail pages */
+       new_cpages = page_array_alloc(cc->inode, new_nr_cpages);
+       if (!new_cpages) {
+               ret = -ENOMEM;
+               goto out_vunmap_cbuf;
+       }
 
        /* zero out any unused part of the last page */
        memset(&cc->cbuf->cdata[cc->clen], 0,
-              (nr_cpages * PAGE_SIZE) - (cc->clen + COMPRESS_HEADER_SIZE));
+                       (new_nr_cpages * PAGE_SIZE) -
+                       (cc->clen + COMPRESS_HEADER_SIZE));
 
-       vunmap(cc->cbuf);
-       vunmap(cc->rbuf);
+       vm_unmap_ram(cc->cbuf, cc->nr_cpages);
+       vm_unmap_ram(cc->rbuf, cc->cluster_size);
 
-       for (i = nr_cpages; i < cc->nr_cpages; i++) {
+       for (i = 0; i < cc->nr_cpages; i++) {
+               if (i < new_nr_cpages) {
+                       new_cpages[i] = cc->cpages[i];
+                       continue;
+               }
                f2fs_compress_free_page(cc->cpages[i]);
                cc->cpages[i] = NULL;
        }
@@ -635,22 +693,24 @@ static int f2fs_compress_pages(struct compress_ctx *cc)
        if (cops->destroy_compress_ctx)
                cops->destroy_compress_ctx(cc);
 
-       cc->nr_cpages = nr_cpages;
+       page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
+       cc->cpages = new_cpages;
+       cc->nr_cpages = new_nr_cpages;
 
        trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx,
                                                        cc->clen, ret);
        return 0;
 
 out_vunmap_cbuf:
-       vunmap(cc->cbuf);
+       vm_unmap_ram(cc->cbuf, cc->nr_cpages);
 out_vunmap_rbuf:
-       vunmap(cc->rbuf);
+       vm_unmap_ram(cc->rbuf, cc->cluster_size);
 out_free_cpages:
        for (i = 0; i < cc->nr_cpages; i++) {
                if (cc->cpages[i])
                        f2fs_compress_free_page(cc->cpages[i]);
        }
-       kfree(cc->cpages);
+       page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
        cc->cpages = NULL;
 destroy_compress_ctx:
        if (cops->destroy_compress_ctx)
@@ -677,7 +737,7 @@ void f2fs_decompress_pages(struct bio *bio, struct page *page, bool verity)
        if (bio->bi_status || PageError(page))
                dic->failed = true;
 
-       if (refcount_dec_not_one(&dic->ref))
+       if (atomic_dec_return(&dic->pending_pages))
                return;
 
        trace_f2fs_decompress_pages_start(dic->inode, dic->cluster_idx,
@@ -689,8 +749,7 @@ void f2fs_decompress_pages(struct bio *bio, struct page *page, bool verity)
                goto out_free_dic;
        }
 
-       dic->tpages = f2fs_kzalloc(sbi, sizeof(struct page *) *
-                                       dic->cluster_size, GFP_NOFS);
+       dic->tpages = page_array_alloc(dic->inode, dic->cluster_size);
        if (!dic->tpages) {
                ret = -ENOMEM;
                goto out_free_dic;
@@ -715,13 +774,13 @@ void f2fs_decompress_pages(struct bio *bio, struct page *page, bool verity)
                        goto out_free_dic;
        }
 
-       dic->rbuf = vmap(dic->tpages, dic->cluster_size, VM_MAP, PAGE_KERNEL);
+       dic->rbuf = f2fs_vmap(dic->tpages, dic->cluster_size);
        if (!dic->rbuf) {
                ret = -ENOMEM;
                goto destroy_decompress_ctx;
        }
 
-       dic->cbuf = vmap(dic->cpages, dic->nr_cpages, VM_MAP, PAGE_KERNEL_RO);
+       dic->cbuf = f2fs_vmap(dic->cpages, dic->nr_cpages);
        if (!dic->cbuf) {
                ret = -ENOMEM;
                goto out_vunmap_rbuf;
@@ -737,16 +796,30 @@ void f2fs_decompress_pages(struct bio *bio, struct page *page, bool verity)
 
        ret = cops->decompress_pages(dic);
 
+       if (!ret && (fi->i_compress_flag & 1 << COMPRESS_CHKSUM)) {
+               u32 provided = le32_to_cpu(dic->cbuf->chksum);
+               u32 calculated = f2fs_crc32(sbi, dic->cbuf->cdata, dic->clen);
+
+               if (provided != calculated) {
+                       if (!is_inode_flag_set(dic->inode, FI_COMPRESS_CORRUPT)) {
+                               set_inode_flag(dic->inode, FI_COMPRESS_CORRUPT);
+                               printk_ratelimited(
+                                       "%sF2FS-fs (%s): checksum invalid, nid = %lu, %x vs %x",
+                                       KERN_INFO, sbi->sb->s_id, dic->inode->i_ino,
+                                       provided, calculated);
+                       }
+                       set_sbi_flag(sbi, SBI_NEED_FSCK);
+               }
+       }
+
 out_vunmap_cbuf:
-       vunmap(dic->cbuf);
+       vm_unmap_ram(dic->cbuf, dic->nr_cpages);
 out_vunmap_rbuf:
-       vunmap(dic->rbuf);
+       vm_unmap_ram(dic->rbuf, dic->cluster_size);
 destroy_decompress_ctx:
        if (cops->destroy_decompress_ctx)
                cops->destroy_decompress_ctx(dic);
 out_free_dic:
-       if (verity)
-               refcount_set(&dic->ref, dic->nr_cpages);
        if (!verity)
                f2fs_decompress_end_io(dic->rpages, dic->cluster_size,
                                                                ret, false);
@@ -868,7 +941,7 @@ int f2fs_is_compressed_cluster(struct inode *inode, pgoff_t index)
 
 static bool cluster_may_compress(struct compress_ctx *cc)
 {
-       if (!f2fs_compressed_file(cc->inode))
+       if (!f2fs_need_compress_data(cc->inode))
                return false;
        if (f2fs_is_atomic_file(cc->inode))
                return false;
@@ -1029,6 +1102,7 @@ bool f2fs_compress_write_end(struct inode *inode, void *fsdata,
 
 {
        struct compress_ctx cc = {
+               .inode = inode,
                .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
                .cluster_size = F2FS_I(inode)->i_cluster_size,
                .rpages = fsdata,
@@ -1132,7 +1206,7 @@ static int f2fs_write_compressed_pages(struct compress_ctx *cc,
                 */
                down_read(&sbi->node_write);
        } else if (!f2fs_trylock_op(sbi)) {
-               return -EAGAIN;
+               goto out_free;
        }
 
        set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
@@ -1155,15 +1229,14 @@ static int f2fs_write_compressed_pages(struct compress_ctx *cc,
 
        fio.version = ni.version;
 
-       cic = f2fs_kzalloc(sbi, sizeof(struct compress_io_ctx), GFP_NOFS);
+       cic = kmem_cache_zalloc(cic_entry_slab, GFP_NOFS);
        if (!cic)
                goto out_put_dnode;
 
        cic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
        cic->inode = inode;
-       refcount_set(&cic->ref, cc->nr_cpages);
-       cic->rpages = f2fs_kzalloc(sbi, sizeof(struct page *) <<
-                       cc->log_cluster_size, GFP_NOFS);
+       atomic_set(&cic->pending_pages, cc->nr_cpages);
+       cic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
        if (!cic->rpages)
                goto out_put_cic;
 
@@ -1257,11 +1330,13 @@ unlock_continue:
        spin_unlock(&fi->i_size_lock);
 
        f2fs_put_rpages(cc);
+       page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
+       cc->cpages = NULL;
        f2fs_destroy_compress_ctx(cc);
        return 0;
 
 out_destroy_crypt:
-       kfree(cic->rpages);
+       page_array_free(cc->inode, cic->rpages, cc->cluster_size);
 
        for (--i; i >= 0; i--)
                fscrypt_finalize_bounce_page(&cc->cpages[i]);
@@ -1271,7 +1346,7 @@ out_destroy_crypt:
                f2fs_put_page(cc->cpages[i], 1);
        }
 out_put_cic:
-       kfree(cic);
+       kmem_cache_free(cic_entry_slab, cic);
 out_put_dnode:
        f2fs_put_dnode(&dn);
 out_unlock_op:
@@ -1279,6 +1354,9 @@ out_unlock_op:
                up_read(&sbi->node_write);
        else
                f2fs_unlock_op(sbi);
+out_free:
+       page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
+       cc->cpages = NULL;
        return -EAGAIN;
 }
 
@@ -1296,7 +1374,7 @@ void f2fs_compress_write_end_io(struct bio *bio, struct page *page)
 
        dec_page_count(sbi, F2FS_WB_DATA);
 
-       if (refcount_dec_not_one(&cic->ref))
+       if (atomic_dec_return(&cic->pending_pages))
                return;
 
        for (i = 0; i < cic->nr_rpages; i++) {
@@ -1305,8 +1383,8 @@ void f2fs_compress_write_end_io(struct bio *bio, struct page *page)
                end_page_writeback(cic->rpages[i]);
        }
 
-       kfree(cic->rpages);
-       kfree(cic);
+       page_array_free(cic->inode, cic->rpages, cic->nr_rpages);
+       kmem_cache_free(cic_entry_slab, cic);
 }
 
 static int f2fs_write_raw_pages(struct compress_ctx *cc,
@@ -1388,9 +1466,6 @@ int f2fs_write_multi_pages(struct compress_ctx *cc,
                                        struct writeback_control *wbc,
                                        enum iostat_type io_type)
 {
-       struct f2fs_inode_info *fi = F2FS_I(cc->inode);
-       const struct f2fs_compress_ops *cops =
-                       f2fs_cops[fi->i_compress_algorithm];
        int err;
 
        *submitted = 0;
@@ -1405,9 +1480,6 @@ int f2fs_write_multi_pages(struct compress_ctx *cc,
 
                err = f2fs_write_compressed_pages(cc, submitted,
                                                        wbc, io_type);
-               cops->destroy_compress_ctx(cc);
-               kfree(cc->cpages);
-               cc->cpages = NULL;
                if (!err)
                        return 0;
                f2fs_bug_on(F2FS_I_SB(cc->inode), err != -EAGAIN);
@@ -1424,25 +1496,23 @@ destroy_out:
 
 struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc)
 {
-       struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode);
        struct decompress_io_ctx *dic;
        pgoff_t start_idx = start_idx_of_cluster(cc);
        int i;
 
-       dic = f2fs_kzalloc(sbi, sizeof(struct decompress_io_ctx), GFP_NOFS);
+       dic = kmem_cache_zalloc(dic_entry_slab, GFP_NOFS);
        if (!dic)
                return ERR_PTR(-ENOMEM);
 
-       dic->rpages = f2fs_kzalloc(sbi, sizeof(struct page *) <<
-                       cc->log_cluster_size, GFP_NOFS);
+       dic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
        if (!dic->rpages) {
-               kfree(dic);
+               kmem_cache_free(dic_entry_slab, dic);
                return ERR_PTR(-ENOMEM);
        }
 
        dic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
        dic->inode = cc->inode;
-       refcount_set(&dic->ref, cc->nr_cpages);
+       atomic_set(&dic->pending_pages, cc->nr_cpages);
        dic->cluster_idx = cc->cluster_idx;
        dic->cluster_size = cc->cluster_size;
        dic->log_cluster_size = cc->log_cluster_size;
@@ -1453,8 +1523,7 @@ struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc)
                dic->rpages[i] = cc->rpages[i];
        dic->nr_rpages = cc->cluster_size;
 
-       dic->cpages = f2fs_kzalloc(sbi, sizeof(struct page *) *
-                                       dic->nr_cpages, GFP_NOFS);
+       dic->cpages = page_array_alloc(dic->inode, dic->nr_cpages);
        if (!dic->cpages)
                goto out_free;
 
@@ -1489,7 +1558,7 @@ void f2fs_free_dic(struct decompress_io_ctx *dic)
                                continue;
                        f2fs_compress_free_page(dic->tpages[i]);
                }
-               kfree(dic->tpages);
+               page_array_free(dic->inode, dic->tpages, dic->cluster_size);
        }
 
        if (dic->cpages) {
@@ -1498,11 +1567,11 @@ void f2fs_free_dic(struct decompress_io_ctx *dic)
                                continue;
                        f2fs_compress_free_page(dic->cpages[i]);
                }
-               kfree(dic->cpages);
+               page_array_free(dic->inode, dic->cpages, dic->nr_cpages);
        }
 
-       kfree(dic->rpages);
-       kfree(dic);
+       page_array_free(dic->inode, dic->rpages, dic->nr_rpages);
+       kmem_cache_free(dic_entry_slab, dic);
 }
 
 void f2fs_decompress_end_io(struct page **rpages,
@@ -1530,3 +1599,76 @@ unlock:
                unlock_page(rpage);
        }
 }
+
+int f2fs_init_page_array_cache(struct f2fs_sb_info *sbi)
+{
+       dev_t dev = sbi->sb->s_bdev->bd_dev;
+       char slab_name[32];
+
+       sprintf(slab_name, "f2fs_page_array_entry-%u:%u", MAJOR(dev), MINOR(dev));
+
+       sbi->page_array_slab_size = sizeof(struct page *) <<
+                                       F2FS_OPTION(sbi).compress_log_size;
+
+       sbi->page_array_slab = f2fs_kmem_cache_create(slab_name,
+                                       sbi->page_array_slab_size);
+       if (!sbi->page_array_slab)
+               return -ENOMEM;
+       return 0;
+}
+
+void f2fs_destroy_page_array_cache(struct f2fs_sb_info *sbi)
+{
+       kmem_cache_destroy(sbi->page_array_slab);
+}
+
+static int __init f2fs_init_cic_cache(void)
+{
+       cic_entry_slab = f2fs_kmem_cache_create("f2fs_cic_entry",
+                                       sizeof(struct compress_io_ctx));
+       if (!cic_entry_slab)
+               return -ENOMEM;
+       return 0;
+}
+
+static void f2fs_destroy_cic_cache(void)
+{
+       kmem_cache_destroy(cic_entry_slab);
+}
+
+static int __init f2fs_init_dic_cache(void)
+{
+       dic_entry_slab = f2fs_kmem_cache_create("f2fs_dic_entry",
+                                       sizeof(struct decompress_io_ctx));
+       if (!dic_entry_slab)
+               return -ENOMEM;
+       return 0;
+}
+
+static void f2fs_destroy_dic_cache(void)
+{
+       kmem_cache_destroy(dic_entry_slab);
+}
+
+int __init f2fs_init_compress_cache(void)
+{
+       int err;
+
+       err = f2fs_init_cic_cache();
+       if (err)
+               goto out;
+       err = f2fs_init_dic_cache();
+       if (err)
+               goto free_cic;
+       return 0;
+free_cic:
+       f2fs_destroy_cic_cache();
+out:
+       return -ENOMEM;
+}
+
+void f2fs_destroy_compress_cache(void)
+{
+       f2fs_destroy_dic_cache();
+       f2fs_destroy_cic_cache();
+}