From 9897713fe1077c90b4a86c9af0a878d56c8888a2 Mon Sep 17 00:00:00 2001 From: Michal Hocko Date: Thu, 26 Sep 2024 19:11:50 +0200 Subject: bcachefs: do not use PF_MEMALLOC_NORECLAIM Patch series "remove PF_MEMALLOC_NORECLAIM" v3. This patch (of 2): bch2_new_inode relies on PF_MEMALLOC_NORECLAIM to try to allocate a new inode to achieve GFP_NOWAIT semantic while holding locks. If this allocation fails it will drop locks and use GFP_NOFS allocation context. We would like to drop PF_MEMALLOC_NORECLAIM because it is really dangerous to use if the caller doesn't control the full call chain with this flag set. E.g. if any of the function down the chain needed GFP_NOFAIL request the PF_MEMALLOC_NORECLAIM would override this and cause unexpected failure. While this is not the case in this particular case using the scoped gfp semantic is not really needed bacause we can easily pus the allocation context down the chain without too much clutter. [akpm@linux-foundation.org: fix kerneldoc warnings] Link: https://lkml.kernel.org/r/20240926172940.167084-1-mhocko@kernel.org Link: https://lkml.kernel.org/r/20240926172940.167084-2-mhocko@kernel.org Signed-off-by: Michal Hocko Reviewed-by: Christoph Hellwig Reviewed-by: Dave Chinner Reviewed-by: Jan Kara # For vfs changes Cc: Al Viro Cc: Christian Brauner Cc: James Morris Cc: Kent Overstreet Cc: Paul Moore Cc: Serge E. Hallyn Cc: Yafang Shao Cc: Matthew Wilcox (Oracle) Cc: Vlastimil Babka Signed-off-by: Andrew Morton --- fs/bcachefs/fs.c | 14 ++++++-------- fs/inode.c | 10 ++++++---- 2 files changed, 12 insertions(+), 12 deletions(-) (limited to 'fs') diff --git a/fs/bcachefs/fs.c b/fs/bcachefs/fs.c index 5bfc26d58270..02969dff165d 100644 --- a/fs/bcachefs/fs.c +++ b/fs/bcachefs/fs.c @@ -300,10 +300,10 @@ static struct inode *bch2_alloc_inode(struct super_block *sb) BUG(); } -static struct bch_inode_info *__bch2_new_inode(struct bch_fs *c) +static struct bch_inode_info *__bch2_new_inode(struct bch_fs *c, gfp_t gfp) { struct bch_inode_info *inode = alloc_inode_sb(c->vfs_sb, - bch2_inode_cache, GFP_NOFS); + bch2_inode_cache, gfp); if (!inode) return NULL; @@ -315,7 +315,7 @@ static struct bch_inode_info *__bch2_new_inode(struct bch_fs *c) mutex_init(&inode->ei_quota_lock); memset(&inode->ei_devs_need_flush, 0, sizeof(inode->ei_devs_need_flush)); - if (unlikely(inode_init_always(c->vfs_sb, &inode->v))) { + if (unlikely(inode_init_always_gfp(c->vfs_sb, &inode->v, gfp))) { kmem_cache_free(bch2_inode_cache, inode); return NULL; } @@ -328,12 +328,10 @@ static struct bch_inode_info *__bch2_new_inode(struct bch_fs *c) */ static struct bch_inode_info *bch2_new_inode(struct btree_trans *trans) { - struct bch_inode_info *inode = - memalloc_flags_do(PF_MEMALLOC_NORECLAIM|PF_MEMALLOC_NOWARN, - __bch2_new_inode(trans->c)); + struct bch_inode_info *inode = __bch2_new_inode(trans->c, GFP_NOWAIT); if (unlikely(!inode)) { - int ret = drop_locks_do(trans, (inode = __bch2_new_inode(trans->c)) ? 0 : -ENOMEM); + int ret = drop_locks_do(trans, (inode = __bch2_new_inode(trans->c, GFP_NOFS)) ? 0 : -ENOMEM); if (ret && inode) { __destroy_inode(&inode->v); kmem_cache_free(bch2_inode_cache, inode); @@ -407,7 +405,7 @@ __bch2_create(struct mnt_idmap *idmap, if (ret) return ERR_PTR(ret); #endif - inode = __bch2_new_inode(c); + inode = __bch2_new_inode(c, GFP_NOFS); if (unlikely(!inode)) { inode = ERR_PTR(-ENOMEM); goto err; diff --git a/fs/inode.c b/fs/inode.c index 471ae4a31549..8dabb224f941 100644 --- a/fs/inode.c +++ b/fs/inode.c @@ -146,14 +146,16 @@ static int no_open(struct inode *inode, struct file *file) } /** - * inode_init_always - perform inode structure initialisation + * inode_init_always_gfp - perform inode structure initialisation * @sb: superblock inode belongs to * @inode: inode to initialise + * @gfp: allocation flags * * These are initializations that need to be done on every inode * allocation as the fields are not initialised by slab allocation. + * If there are additional allocations required @gfp is used. */ -int inode_init_always(struct super_block *sb, struct inode *inode) +int inode_init_always_gfp(struct super_block *sb, struct inode *inode, gfp_t gfp) { static const struct inode_operations empty_iops; static const struct file_operations no_open_fops = {.open = no_open}; @@ -230,14 +232,14 @@ int inode_init_always(struct super_block *sb, struct inode *inode) #endif inode->i_flctx = NULL; - if (unlikely(security_inode_alloc(inode))) + if (unlikely(security_inode_alloc(inode, gfp))) return -ENOMEM; this_cpu_inc(nr_inodes); return 0; } -EXPORT_SYMBOL(inode_init_always); +EXPORT_SYMBOL(inode_init_always_gfp); void free_inode_nonrcu(struct inode *inode) { -- cgit v1.2.3-70-g09d2 From 3d5854d75e3187147613130561b58f0b06166172 Mon Sep 17 00:00:00 2001 From: Alexander Gordeev Date: Mon, 30 Sep 2024 14:21:19 +0200 Subject: fs/proc/kcore.c: allow translation of physical memory addresses When /proc/kcore is read an attempt to read the first two pages results in HW-specific page swap on s390 and another (so called prefix) pages are accessed instead. That leads to a wrong read. Allow architecture-specific translation of memory addresses using kc_xlate_dev_mem_ptr() and kc_unxlate_dev_mem_ptr() callbacks similarily to /dev/mem xlate_dev_mem_ptr() and unxlate_dev_mem_ptr() callbacks. That way an architecture can deal with specific physical memory ranges. Re-use the existing /dev/mem callback implementation on s390, which handles the described prefix pages swapping correctly. For other architectures the default callback is basically NOP. It is expected the condition (vaddr == __va(__pa(vaddr))) always holds true for KCORE_RAM memory type. Link: https://lkml.kernel.org/r/20240930122119.1651546-1-agordeev@linux.ibm.com Signed-off-by: Alexander Gordeev Suggested-by: Heiko Carstens Cc: Vasily Gorbik Cc: Signed-off-by: Andrew Morton --- arch/s390/include/asm/io.h | 2 ++ fs/proc/kcore.c | 36 ++++++++++++++++++++++++++++++++++-- 2 files changed, 36 insertions(+), 2 deletions(-) (limited to 'fs') diff --git a/arch/s390/include/asm/io.h b/arch/s390/include/asm/io.h index 0fbc992d7a5e..fc9933a743d6 100644 --- a/arch/s390/include/asm/io.h +++ b/arch/s390/include/asm/io.h @@ -16,8 +16,10 @@ #include #define xlate_dev_mem_ptr xlate_dev_mem_ptr +#define kc_xlate_dev_mem_ptr xlate_dev_mem_ptr void *xlate_dev_mem_ptr(phys_addr_t phys); #define unxlate_dev_mem_ptr unxlate_dev_mem_ptr +#define kc_unxlate_dev_mem_ptr unxlate_dev_mem_ptr void unxlate_dev_mem_ptr(phys_addr_t phys, void *addr); #define IO_SPACE_LIMIT 0 diff --git a/fs/proc/kcore.c b/fs/proc/kcore.c index 7d0acdad74e2..51446c59388f 100644 --- a/fs/proc/kcore.c +++ b/fs/proc/kcore.c @@ -50,6 +50,20 @@ static struct proc_dir_entry *proc_root_kcore; #define kc_offset_to_vaddr(o) ((o) + PAGE_OFFSET) #endif +#ifndef kc_xlate_dev_mem_ptr +#define kc_xlate_dev_mem_ptr kc_xlate_dev_mem_ptr +static inline void *kc_xlate_dev_mem_ptr(phys_addr_t phys) +{ + return __va(phys); +} +#endif +#ifndef kc_unxlate_dev_mem_ptr +#define kc_unxlate_dev_mem_ptr kc_unxlate_dev_mem_ptr +static inline void kc_unxlate_dev_mem_ptr(phys_addr_t phys, void *virt) +{ +} +#endif + static LIST_HEAD(kclist_head); static DECLARE_RWSEM(kclist_lock); static int kcore_need_update = 1; @@ -471,6 +485,8 @@ static ssize_t read_kcore_iter(struct kiocb *iocb, struct iov_iter *iter) while (buflen) { struct page *page; unsigned long pfn; + phys_addr_t phys; + void *__start; /* * If this is the first iteration or the address is not within @@ -537,7 +553,8 @@ static ssize_t read_kcore_iter(struct kiocb *iocb, struct iov_iter *iter) } break; case KCORE_RAM: - pfn = __pa(start) >> PAGE_SHIFT; + phys = __pa(start); + pfn = phys >> PAGE_SHIFT; page = pfn_to_online_page(pfn); /* @@ -557,13 +574,28 @@ static ssize_t read_kcore_iter(struct kiocb *iocb, struct iov_iter *iter) fallthrough; case KCORE_VMEMMAP: case KCORE_TEXT: + if (m->type == KCORE_RAM) { + __start = kc_xlate_dev_mem_ptr(phys); + if (!__start) { + ret = -ENOMEM; + if (iov_iter_zero(tsz, iter) != tsz) + ret = -EFAULT; + goto out; + } + } else { + __start = (void *)start; + } + /* * Sadly we must use a bounce buffer here to be able to * make use of copy_from_kernel_nofault(), as these * memory regions might not always be mapped on all * architectures. */ - if (copy_from_kernel_nofault(buf, (void *)start, tsz)) { + ret = copy_from_kernel_nofault(buf, __start, tsz); + if (m->type == KCORE_RAM) + kc_unxlate_dev_mem_ptr(phys, __start); + if (ret) { if (iov_iter_zero(tsz, iter) != tsz) { ret = -EFAULT; goto out; -- cgit v1.2.3-70-g09d2