From 8fc3c3862728373e0d0f5abccc6afc56c69e0c63 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Wed, 1 Jun 2016 21:32:24 -0400 Subject: NFS: Fix O_DIRECT verifier problems We should not be interested in looking at the value of the stable field, since that could take any value. Signed-off-by: Trond Myklebust --- fs/nfs/direct.c | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-) (limited to 'fs/nfs/direct.c') diff --git a/fs/nfs/direct.c b/fs/nfs/direct.c index 979b3c4dee6a..d6d43b5eafb3 100644 --- a/fs/nfs/direct.c +++ b/fs/nfs/direct.c @@ -196,6 +196,12 @@ static void nfs_direct_set_hdr_verf(struct nfs_direct_req *dreq, WARN_ON_ONCE(verfp->committed < 0); } +static int nfs_direct_cmp_verf(const struct nfs_writeverf *v1, + const struct nfs_writeverf *v2) +{ + return nfs_write_verifier_cmp(&v1->verifier, &v2->verifier); +} + /* * nfs_direct_cmp_hdr_verf - compare verifier for pgio header * @dreq - direct request possibly spanning multiple servers @@ -215,7 +221,7 @@ static int nfs_direct_set_or_cmp_hdr_verf(struct nfs_direct_req *dreq, nfs_direct_set_hdr_verf(dreq, hdr); return 0; } - return memcmp(verfp, &hdr->verf, sizeof(struct nfs_writeverf)); + return nfs_direct_cmp_verf(verfp, &hdr->verf); } /* @@ -238,7 +244,7 @@ static int nfs_direct_cmp_commit_data_verf(struct nfs_direct_req *dreq, if (verfp->committed < 0) return 1; - return memcmp(verfp, &data->verf, sizeof(struct nfs_writeverf)); + return nfs_direct_cmp_verf(verfp, &data->verf); } /** -- cgit v1.2.3 From a5314a74928fa6dbc4503a8c64f43bb5c1c12ac1 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Wed, 1 Jun 2016 21:42:32 -0400 Subject: NFS: Ensure we reset the write verifier 'committed' value on resend. Signed-off-by: Trond Myklebust --- fs/nfs/direct.c | 2 ++ fs/nfs/internal.h | 17 +++++++++++++++++ 2 files changed, 19 insertions(+) (limited to 'fs/nfs/direct.c') diff --git a/fs/nfs/direct.c b/fs/nfs/direct.c index d6d43b5eafb3..fb659bb50678 100644 --- a/fs/nfs/direct.c +++ b/fs/nfs/direct.c @@ -661,6 +661,8 @@ static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq) nfs_direct_write_scan_commit_list(dreq->inode, &reqs, &cinfo); dreq->count = 0; + dreq->verf.committed = NFS_INVALID_STABLE_HOW; + nfs_clear_pnfs_ds_commit_verifiers(&dreq->ds_cinfo); for (i = 0; i < dreq->mirror_count; i++) dreq->mirrors[i].count = 0; get_dreq(dreq); diff --git a/fs/nfs/internal.h b/fs/nfs/internal.h index 150a8eb0f323..0eb5c924886d 100644 --- a/fs/nfs/internal.h +++ b/fs/nfs/internal.h @@ -499,6 +499,23 @@ int nfs_key_timeout_notify(struct file *filp, struct inode *inode); bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx); void nfs_pageio_stop_mirroring(struct nfs_pageio_descriptor *pgio); +#ifdef CONFIG_NFS_V4_1 +static inline +void nfs_clear_pnfs_ds_commit_verifiers(struct pnfs_ds_commit_info *cinfo) +{ + int i; + + for (i = 0; i < cinfo->nbuckets; i++) + cinfo->buckets[i].direct_verf.committed = NFS_INVALID_STABLE_HOW; +} +#else +static inline +void nfs_clear_pnfs_ds_commit_verifiers(struct pnfs_ds_commit_info *cinfo) +{ +} +#endif + + #ifdef CONFIG_MIGRATION extern int nfs_migrate_page(struct address_space *, struct page *, struct page *, enum migrate_mode); -- cgit v1.2.3 From 2f3c7d87a347b12f725f6128b3097727b91b230e Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Wed, 22 Jun 2016 14:38:06 -0400 Subject: NFS: Remove racy size manipulations in O_DIRECT On success, the RPC callbacks will ensure that we make the appropriate calls to nfs_writeback_update_inode() Signed-off-by: Trond Myklebust --- fs/nfs/direct.c | 16 ---------------- 1 file changed, 16 deletions(-) (limited to 'fs/nfs/direct.c') diff --git a/fs/nfs/direct.c b/fs/nfs/direct.c index fb659bb50678..826d4dace0e5 100644 --- a/fs/nfs/direct.c +++ b/fs/nfs/direct.c @@ -376,15 +376,6 @@ static void nfs_direct_complete(struct nfs_direct_req *dreq, bool write) { struct inode *inode = dreq->inode; - if (dreq->iocb && write) { - loff_t pos = dreq->iocb->ki_pos + dreq->count; - - spin_lock(&inode->i_lock); - if (i_size_read(inode) < pos) - i_size_write(inode, pos); - spin_unlock(&inode->i_lock); - } - if (write) nfs_zap_mapping(inode, inode->i_mapping); @@ -1058,14 +1049,7 @@ ssize_t nfs_file_direct_write(struct kiocb *iocb, struct iov_iter *iter) if (!result) { result = nfs_direct_wait(dreq); if (result > 0) { - struct inode *inode = mapping->host; - iocb->ki_pos = pos + result; - spin_lock(&inode->i_lock); - if (i_size_read(inode) < iocb->ki_pos) - i_size_write(inode, iocb->ki_pos); - spin_unlock(&inode->i_lock); - /* XXX: should check the generic_write_sync retval */ generic_write_sync(iocb, result); } -- cgit v1.2.3 From 89698b24d24f9c8b470a73351b0b7199c17e0153 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Thu, 23 Jun 2016 10:35:48 -0400 Subject: NFS Cleanup: move call to generic_write_checks() into fs/nfs/direct.c Signed-off-by: Trond Myklebust --- fs/nfs/direct.c | 12 ++++++++---- fs/nfs/file.c | 6 +----- 2 files changed, 9 insertions(+), 9 deletions(-) (limited to 'fs/nfs/direct.c') diff --git a/fs/nfs/direct.c b/fs/nfs/direct.c index 826d4dace0e5..0169eca8eb42 100644 --- a/fs/nfs/direct.c +++ b/fs/nfs/direct.c @@ -988,6 +988,7 @@ static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq, ssize_t nfs_file_direct_write(struct kiocb *iocb, struct iov_iter *iter) { ssize_t result = -EINVAL; + size_t count; struct file *file = iocb->ki_filp; struct address_space *mapping = file->f_mapping; struct inode *inode = mapping->host; @@ -998,8 +999,11 @@ ssize_t nfs_file_direct_write(struct kiocb *iocb, struct iov_iter *iter) dfprintk(FILE, "NFS: direct write(%pD2, %zd@%Ld)\n", file, iov_iter_count(iter), (long long) iocb->ki_pos); - nfs_add_stats(mapping->host, NFSIOS_DIRECTWRITTENBYTES, - iov_iter_count(iter)); + result = generic_write_checks(iocb, iter); + if (result <= 0) + return result; + count = result; + nfs_add_stats(mapping->host, NFSIOS_DIRECTWRITTENBYTES, count); pos = iocb->ki_pos; end = (pos + iov_iter_count(iter) - 1) >> PAGE_SHIFT; @@ -1017,7 +1021,7 @@ ssize_t nfs_file_direct_write(struct kiocb *iocb, struct iov_iter *iter) goto out_unlock; } - task_io_account_write(iov_iter_count(iter)); + task_io_account_write(count); result = -ENOMEM; dreq = nfs_direct_req_alloc(); @@ -1025,7 +1029,7 @@ ssize_t nfs_file_direct_write(struct kiocb *iocb, struct iov_iter *iter) goto out_unlock; dreq->inode = inode; - dreq->bytes_left = dreq->max_count = iov_iter_count(iter); + dreq->bytes_left = dreq->max_count = count; dreq->io_start = pos; dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp)); l_ctx = nfs_get_lock_context(dreq->ctx); diff --git a/fs/nfs/file.c b/fs/nfs/file.c index df4dd8e7e62e..c26847c84d00 100644 --- a/fs/nfs/file.c +++ b/fs/nfs/file.c @@ -629,12 +629,8 @@ ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from) if (result) return result; - if (iocb->ki_flags & IOCB_DIRECT) { - result = generic_write_checks(iocb, from); - if (result <= 0) - return result; + if (iocb->ki_flags & IOCB_DIRECT) return nfs_file_direct_write(iocb, from); - } dprintk("NFS: write(%pD2, %zu@%Ld)\n", file, count, (long long) iocb->ki_pos); -- cgit v1.2.3 From a5864c999de6703f7ce908f72337568520c6cad3 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Fri, 3 Jun 2016 17:07:19 -0400 Subject: NFS: Do not serialise O_DIRECT reads and writes Allow dio requests to be scheduled in parallel, but ensuring that they do not conflict with buffered I/O. Signed-off-by: Trond Myklebust --- fs/nfs/Makefile | 2 +- fs/nfs/direct.c | 41 +++----------- fs/nfs/file.c | 12 ++-- fs/nfs/internal.h | 8 +++ fs/nfs/io.c | 147 +++++++++++++++++++++++++++++++++++++++++++++++++ include/linux/nfs_fs.h | 1 + 6 files changed, 174 insertions(+), 37 deletions(-) create mode 100644 fs/nfs/io.c (limited to 'fs/nfs/direct.c') diff --git a/fs/nfs/Makefile b/fs/nfs/Makefile index 8664417955a2..6abdda209642 100644 --- a/fs/nfs/Makefile +++ b/fs/nfs/Makefile @@ -6,7 +6,7 @@ obj-$(CONFIG_NFS_FS) += nfs.o CFLAGS_nfstrace.o += -I$(src) nfs-y := client.o dir.o file.o getroot.o inode.o super.o \ - direct.o pagelist.o read.o symlink.o unlink.o \ + io.o direct.o pagelist.o read.o symlink.o unlink.o \ write.o namespace.o mount_clnt.o nfstrace.o nfs-$(CONFIG_ROOT_NFS) += nfsroot.o nfs-$(CONFIG_SYSCTL) += sysctl.o diff --git a/fs/nfs/direct.c b/fs/nfs/direct.c index 0169eca8eb42..6d0e88096440 100644 --- a/fs/nfs/direct.c +++ b/fs/nfs/direct.c @@ -578,17 +578,12 @@ ssize_t nfs_file_direct_read(struct kiocb *iocb, struct iov_iter *iter) if (!count) goto out; - inode_lock(inode); - result = nfs_sync_mapping(mapping); - if (result) - goto out_unlock; - task_io_account_read(count); result = -ENOMEM; dreq = nfs_direct_req_alloc(); if (dreq == NULL) - goto out_unlock; + goto out; dreq->inode = inode; dreq->bytes_left = dreq->max_count = count; @@ -603,10 +598,12 @@ ssize_t nfs_file_direct_read(struct kiocb *iocb, struct iov_iter *iter) if (!is_sync_kiocb(iocb)) dreq->iocb = iocb; + nfs_start_io_direct(inode); + NFS_I(inode)->read_io += count; result = nfs_direct_read_schedule_iovec(dreq, iter, iocb->ki_pos); - inode_unlock(inode); + nfs_end_io_direct(inode); if (!result) { result = nfs_direct_wait(dreq); @@ -614,13 +611,8 @@ ssize_t nfs_file_direct_read(struct kiocb *iocb, struct iov_iter *iter) iocb->ki_pos += result; } - nfs_direct_req_release(dreq); - return result; - out_release: nfs_direct_req_release(dreq); -out_unlock: - inode_unlock(inode); out: return result; } @@ -1008,25 +1000,12 @@ ssize_t nfs_file_direct_write(struct kiocb *iocb, struct iov_iter *iter) pos = iocb->ki_pos; end = (pos + iov_iter_count(iter) - 1) >> PAGE_SHIFT; - inode_lock(inode); - - result = nfs_sync_mapping(mapping); - if (result) - goto out_unlock; - - if (mapping->nrpages) { - result = invalidate_inode_pages2_range(mapping, - pos >> PAGE_SHIFT, end); - if (result) - goto out_unlock; - } - task_io_account_write(count); result = -ENOMEM; dreq = nfs_direct_req_alloc(); if (!dreq) - goto out_unlock; + goto out; dreq->inode = inode; dreq->bytes_left = dreq->max_count = count; @@ -1041,6 +1020,8 @@ ssize_t nfs_file_direct_write(struct kiocb *iocb, struct iov_iter *iter) if (!is_sync_kiocb(iocb)) dreq->iocb = iocb; + nfs_start_io_direct(inode); + result = nfs_direct_write_schedule_iovec(dreq, iter, pos); if (mapping->nrpages) { @@ -1048,7 +1029,7 @@ ssize_t nfs_file_direct_write(struct kiocb *iocb, struct iov_iter *iter) pos >> PAGE_SHIFT, end); } - inode_unlock(inode); + nfs_end_io_direct(inode); if (!result) { result = nfs_direct_wait(dreq); @@ -1058,13 +1039,9 @@ ssize_t nfs_file_direct_write(struct kiocb *iocb, struct iov_iter *iter) generic_write_sync(iocb, result); } } - nfs_direct_req_release(dreq); - return result; - out_release: nfs_direct_req_release(dreq); -out_unlock: - inode_unlock(inode); +out: return result; } diff --git a/fs/nfs/file.c b/fs/nfs/file.c index 46cf0afe3c0f..9f8da9e1b23f 100644 --- a/fs/nfs/file.c +++ b/fs/nfs/file.c @@ -170,12 +170,14 @@ nfs_file_read(struct kiocb *iocb, struct iov_iter *to) iocb->ki_filp, iov_iter_count(to), (unsigned long) iocb->ki_pos); - result = nfs_revalidate_mapping_protected(inode, iocb->ki_filp->f_mapping); + nfs_start_io_read(inode); + result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping); if (!result) { result = generic_file_read_iter(iocb, to); if (result > 0) nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result); } + nfs_end_io_read(inode); return result; } EXPORT_SYMBOL_GPL(nfs_file_read); @@ -191,12 +193,14 @@ nfs_file_splice_read(struct file *filp, loff_t *ppos, dprintk("NFS: splice_read(%pD2, %lu@%Lu)\n", filp, (unsigned long) count, (unsigned long long) *ppos); - res = nfs_revalidate_mapping_protected(inode, filp->f_mapping); + nfs_start_io_read(inode); + res = nfs_revalidate_mapping(inode, filp->f_mapping); if (!res) { res = generic_file_splice_read(filp, ppos, pipe, count, flags); if (res > 0) nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res); } + nfs_end_io_read(inode); return res; } EXPORT_SYMBOL_GPL(nfs_file_splice_read); @@ -645,14 +649,14 @@ ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from) goto out; } - inode_lock(inode); + nfs_start_io_write(inode); result = generic_write_checks(iocb, from); if (result > 0) { current->backing_dev_info = inode_to_bdi(inode); result = generic_perform_write(file, from, iocb->ki_pos); current->backing_dev_info = NULL; } - inode_unlock(inode); + nfs_end_io_write(inode); if (result <= 0) goto out; diff --git a/fs/nfs/internal.h b/fs/nfs/internal.h index 0eb5c924886d..159b64ede82a 100644 --- a/fs/nfs/internal.h +++ b/fs/nfs/internal.h @@ -411,6 +411,14 @@ extern void __exit unregister_nfs_fs(void); extern bool nfs_sb_active(struct super_block *sb); extern void nfs_sb_deactive(struct super_block *sb); +/* io.c */ +extern void nfs_start_io_read(struct inode *inode); +extern void nfs_end_io_read(struct inode *inode); +extern void nfs_start_io_write(struct inode *inode); +extern void nfs_end_io_write(struct inode *inode); +extern void nfs_start_io_direct(struct inode *inode); +extern void nfs_end_io_direct(struct inode *inode); + /* namespace.c */ #define NFS_PATH_CANONICAL 1 extern char *nfs_path(char **p, struct dentry *dentry, diff --git a/fs/nfs/io.c b/fs/nfs/io.c new file mode 100644 index 000000000000..1fc5d1ce327e --- /dev/null +++ b/fs/nfs/io.c @@ -0,0 +1,147 @@ +/* + * Copyright (c) 2016 Trond Myklebust + * + * I/O and data path helper functionality. + */ + +#include +#include +#include +#include +#include +#include + +#include "internal.h" + +/* Call with exclusively locked inode->i_rwsem */ +static void nfs_block_o_direct(struct nfs_inode *nfsi, struct inode *inode) +{ + if (test_bit(NFS_INO_ODIRECT, &nfsi->flags)) { + clear_bit(NFS_INO_ODIRECT, &nfsi->flags); + inode_dio_wait(inode); + } +} + +/** + * nfs_start_io_read - declare the file is being used for buffered reads + * @inode - file inode + * + * Declare that a buffered read operation is about to start, and ensure + * that we block all direct I/O. + * On exit, the function ensures that the NFS_INO_ODIRECT flag is unset, + * and holds a shared lock on inode->i_rwsem to ensure that the flag + * cannot be changed. + * In practice, this means that buffered read operations are allowed to + * execute in parallel, thanks to the shared lock, whereas direct I/O + * operations need to wait to grab an exclusive lock in order to set + * NFS_INO_ODIRECT. + * Note that buffered writes and truncates both take a write lock on + * inode->i_rwsem, meaning that those are serialised w.r.t. the reads. + */ +void +nfs_start_io_read(struct inode *inode) +{ + struct nfs_inode *nfsi = NFS_I(inode); + /* Be an optimist! */ + down_read(&inode->i_rwsem); + if (test_bit(NFS_INO_ODIRECT, &nfsi->flags) == 0) + return; + up_read(&inode->i_rwsem); + /* Slow path.... */ + down_write(&inode->i_rwsem); + nfs_block_o_direct(nfsi, inode); + downgrade_write(&inode->i_rwsem); +} + +/** + * nfs_end_io_read - declare that the buffered read operation is done + * @inode - file inode + * + * Declare that a buffered read operation is done, and release the shared + * lock on inode->i_rwsem. + */ +void +nfs_end_io_read(struct inode *inode) +{ + up_read(&inode->i_rwsem); +} + +/** + * nfs_start_io_write - declare the file is being used for buffered writes + * @inode - file inode + * + * Declare that a buffered read operation is about to start, and ensure + * that we block all direct I/O. + */ +void +nfs_start_io_write(struct inode *inode) +{ + down_write(&inode->i_rwsem); + nfs_block_o_direct(NFS_I(inode), inode); +} + +/** + * nfs_end_io_write - declare that the buffered write operation is done + * @inode - file inode + * + * Declare that a buffered write operation is done, and release the + * lock on inode->i_rwsem. + */ +void +nfs_end_io_write(struct inode *inode) +{ + up_write(&inode->i_rwsem); +} + +/* Call with exclusively locked inode->i_rwsem */ +static void nfs_block_buffered(struct nfs_inode *nfsi, struct inode *inode) +{ + if (!test_bit(NFS_INO_ODIRECT, &nfsi->flags)) { + set_bit(NFS_INO_ODIRECT, &nfsi->flags); + nfs_wb_all(inode); + } +} + +/** + * nfs_end_io_direct - declare the file is being used for direct i/o + * @inode - file inode + * + * Declare that a direct I/O operation is about to start, and ensure + * that we block all buffered I/O. + * On exit, the function ensures that the NFS_INO_ODIRECT flag is set, + * and holds a shared lock on inode->i_rwsem to ensure that the flag + * cannot be changed. + * In practice, this means that direct I/O operations are allowed to + * execute in parallel, thanks to the shared lock, whereas buffered I/O + * operations need to wait to grab an exclusive lock in order to clear + * NFS_INO_ODIRECT. + * Note that buffered writes and truncates both take a write lock on + * inode->i_rwsem, meaning that those are serialised w.r.t. O_DIRECT. + */ +void +nfs_start_io_direct(struct inode *inode) +{ + struct nfs_inode *nfsi = NFS_I(inode); + /* Be an optimist! */ + down_read(&inode->i_rwsem); + if (test_bit(NFS_INO_ODIRECT, &nfsi->flags) != 0) + return; + up_read(&inode->i_rwsem); + /* Slow path.... */ + down_write(&inode->i_rwsem); + nfs_block_buffered(nfsi, inode); + downgrade_write(&inode->i_rwsem); +} + +/** + * nfs_end_io_direct - declare that the direct i/o operation is done + * @inode - file inode + * + * Declare that a direct I/O operation is done, and release the shared + * lock on inode->i_rwsem. + */ +void +nfs_end_io_direct(struct inode *inode) +{ + up_read(&inode->i_rwsem); +} diff --git a/include/linux/nfs_fs.h b/include/linux/nfs_fs.h index 120dd04b553c..225d17d35277 100644 --- a/include/linux/nfs_fs.h +++ b/include/linux/nfs_fs.h @@ -210,6 +210,7 @@ struct nfs_inode { #define NFS_INO_LAYOUTCOMMIT (9) /* layoutcommit required */ #define NFS_INO_LAYOUTCOMMITTING (10) /* layoutcommit inflight */ #define NFS_INO_LAYOUTSTATS (11) /* layoutstats inflight */ +#define NFS_INO_ODIRECT (12) /* I/O setting is O_DIRECT */ static inline struct nfs_inode *NFS_I(const struct inode *inode) { -- cgit v1.2.3 From f7b5c340aca87d736a6b15aa40bf135f1baab011 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Thu, 23 Jun 2016 09:29:47 -0400 Subject: NFS: Cleanup nfs_direct_complete() There is only one caller that sets the "write" argument to true, so just move the call to nfs_zap_mapping() and get rid of the now redundant argument. Signed-off-by: Trond Myklebust --- fs/nfs/direct.c | 12 +++++------- 1 file changed, 5 insertions(+), 7 deletions(-) (limited to 'fs/nfs/direct.c') diff --git a/fs/nfs/direct.c b/fs/nfs/direct.c index 6d0e88096440..c16d33eb1ddf 100644 --- a/fs/nfs/direct.c +++ b/fs/nfs/direct.c @@ -372,13 +372,10 @@ out: * Synchronous I/O uses a stack-allocated iocb. Thus we can't trust * the iocb is still valid here if this is a synchronous request. */ -static void nfs_direct_complete(struct nfs_direct_req *dreq, bool write) +static void nfs_direct_complete(struct nfs_direct_req *dreq) { struct inode *inode = dreq->inode; - if (write) - nfs_zap_mapping(inode, inode->i_mapping); - inode_dio_end(inode); if (dreq->iocb) { @@ -431,7 +428,7 @@ static void nfs_direct_read_completion(struct nfs_pgio_header *hdr) } out_put: if (put_dreq(dreq)) - nfs_direct_complete(dreq, false); + nfs_direct_complete(dreq); hdr->release(hdr); } @@ -537,7 +534,7 @@ static ssize_t nfs_direct_read_schedule_iovec(struct nfs_direct_req *dreq, } if (put_dreq(dreq)) - nfs_direct_complete(dreq, false); + nfs_direct_complete(dreq); return 0; } @@ -764,7 +761,8 @@ static void nfs_direct_write_schedule_work(struct work_struct *work) nfs_direct_write_reschedule(dreq); break; default: - nfs_direct_complete(dreq, true); + nfs_zap_mapping(dreq->inode, dreq->inode->i_mapping); + nfs_direct_complete(dreq); } } -- cgit v1.2.3