/* for struct cl_lock_descr and struct cl_io */
#include <cl_object.h>
#include <lclient.h>
+#include <lustre_mdc.h>
+#include <linux/lustre_intent.h>
#ifndef FMODE_EXEC
#define FMODE_EXEC 0
* be sent to MDS. */
LLIF_SOM_DIRTY = (1 << 3),
/* File is contented */
- LLIF_CONTENDED = (1 << 4),
+ LLIF_CONTENDED = (1 << 4),
/* Truncate uses server lock for this file */
- LLIF_SRVLOCK = (1 << 5)
+ LLIF_SRVLOCK = (1 << 5),
};
struct ll_inode_info {
- int lli_inode_magic;
- cfs_semaphore_t lli_size_sem; /* protect open and change size */
- void *lli_size_sem_owner;
- cfs_semaphore_t lli_write_sem;
- cfs_rw_semaphore_t lli_trunc_sem;
- char *lli_symlink_name;
- __u64 lli_maxbytes;
- __u64 lli_ioepoch;
- unsigned long lli_flags;
- cfs_time_t lli_contention_time;
-
- /* this lock protects posix_acl, pending_write_llaps, mmap_cnt */
- cfs_spinlock_t lli_lock;
- cfs_list_t lli_close_list;
- /* handle is to be sent to MDS later on done_writing and setattr.
- * Open handle data are needed for the recovery to reconstruct
- * the inode state on the MDS. XXX: recovery is not ready yet. */
- struct obd_client_handle *lli_pending_och;
-
- /* for writepage() only to communicate to fsync */
- int lli_async_rc;
- int lli_write_rc;
-
- struct posix_acl *lli_posix_acl;
+ __u32 lli_inode_magic;
+ __u32 lli_flags;
+ __u64 lli_ioepoch;
- /* remote permission hash */
- cfs_hlist_head_t *lli_remote_perms;
- unsigned long lli_rmtperm_utime;
- cfs_semaphore_t lli_rmtperm_sem;
+ cfs_spinlock_t lli_lock;
+ struct posix_acl *lli_posix_acl;
- cfs_list_t lli_dead_list;
-
- cfs_semaphore_t lli_och_sem; /* Protects access to och pointers
- and their usage counters, also
- atomicity of check-update of
- lli_smd */
- /* We need all three because every inode may be opened in different
- modes */
- struct obd_client_handle *lli_mds_read_och;
- __u64 lli_open_fd_read_count;
- struct obd_client_handle *lli_mds_write_och;
- __u64 lli_open_fd_write_count;
- struct obd_client_handle *lli_mds_exec_och;
- __u64 lli_open_fd_exec_count;
-
- struct inode lli_vfs_inode;
+ cfs_hlist_head_t *lli_remote_perms;
+ cfs_semaphore_t lli_rmtperm_sem;
/* identifying fields for both metadata and data stacks. */
- struct lu_fid lli_fid;
+ struct lu_fid lli_fid;
/* Parent fid for accessing default stripe data on parent directory
* for allocating OST objects after a mknod() and later open-by-FID. */
- struct lu_fid lli_pfid;
- struct lov_stripe_md *lli_smd;
+ struct lu_fid lli_pfid;
- /* fid capability */
+ cfs_list_t lli_close_list;
+ cfs_list_t lli_oss_capas;
/* open count currently used by capability only, indicate whether
* capability needs renewal */
- cfs_atomic_t lli_open_count;
- struct obd_capa *lli_mds_capa;
- cfs_list_t lli_oss_capas;
-
- /* metadata statahead */
- /* protect statahead stuff: lli_opendir_pid, lli_opendir_key, lli_sai,
- * and so on. */
- cfs_spinlock_t lli_sa_lock;
- /*
- * "opendir_pid" is the token when lookup/revalid -- I am the owner of
- * dir statahead.
- */
- pid_t lli_opendir_pid;
- /*
- * since parent-child threads can share the same @file struct,
- * "opendir_key" is the token when dir close for case of parent exit
- * before child -- it is me should cleanup the dir readahead. */
- void *lli_opendir_key;
- struct ll_statahead_info *lli_sai;
- __u64 lli_sa_pos;
- struct cl_object *lli_clob;
+ cfs_atomic_t lli_open_count;
+ struct obd_capa *lli_mds_capa;
+ cfs_time_t lli_rmtperm_time;
+
+ /* handle is to be sent to MDS later on done_writing and setattr.
+ * Open handle data are needed for the recovery to reconstruct
+ * the inode state on the MDS. XXX: recovery is not ready yet. */
+ struct obd_client_handle *lli_pending_och;
+
+ /* We need all three because every inode may be opened in different
+ * modes */
+ struct obd_client_handle *lli_mds_read_och;
+ struct obd_client_handle *lli_mds_write_och;
+ struct obd_client_handle *lli_mds_exec_och;
+ __u64 lli_open_fd_read_count;
+ __u64 lli_open_fd_write_count;
+ __u64 lli_open_fd_exec_count;
+ /* Protects access to och pointers and their usage counters, also
+ * atomicity of check-update of lli_smd */
+ cfs_semaphore_t lli_och_sem;
+
+ struct inode lli_vfs_inode;
+
/* the most recent timestamps obtained from mds */
- struct ost_lvb lli_lvb;
- /**
- * serialize normal readdir and statahead-readdir
+ struct ost_lvb lli_lvb;
+ cfs_spinlock_t lli_agl_lock;
+
+ /* Try to make the d::member and f::member are aligned. Before using
+ * these members, make clear whether it is directory or not. */
+ union {
+ /* for directory */
+ struct {
+ /* serialize normal readdir and statahead-readdir. */
+ cfs_semaphore_t d_readdir_sem;
+
+ /* metadata statahead */
+ /* since parent-child threads can share the same @file
+ * struct, "opendir_key" is the token when dir close for
+ * case of parent exit before child -- it is me should
+ * cleanup the dir readahead. */
+ void *d_opendir_key;
+ struct ll_statahead_info *d_sai;
+ __u64 d_sa_pos;
+ struct posix_acl *d_def_acl;
+ /* protect statahead stuff. */
+ cfs_spinlock_t d_sa_lock;
+ /* "opendir_pid" is the token when lookup/revalid
+ * -- I am the owner of dir statahead. */
+ pid_t d_opendir_pid;
+ } d;
+
+#define lli_readdir_sem u.d.d_readdir_sem
+#define lli_opendir_key u.d.d_opendir_key
+#define lli_sai u.d.d_sai
+#define lli_sa_pos u.d.d_sa_pos
+#define lli_def_acl u.d.d_def_acl
+#define lli_sa_lock u.d.d_sa_lock
+#define lli_opendir_pid u.d.d_opendir_pid
+
+ /* for non-directory */
+ struct {
+ cfs_semaphore_t f_size_sem;
+ void *f_size_sem_owner;
+ char *f_symlink_name;
+ __u64 f_maxbytes;
+ /*
+ * cfs_rw_semaphore_t {
+ * signed long count; // align u.d.d_def_acl
+ * cfs_spinlock_t wait_lock; // align u.d.d_sa_lock
+ * struct list_head wait_list;
+ * }
+ */
+ cfs_rw_semaphore_t f_trunc_sem;
+ cfs_semaphore_t f_write_sem;
+
+ /* for writepage() only to communicate to fsync */
+ int f_async_rc;
+ int f_write_rc;
+
+ cfs_rw_semaphore_t f_glimpse_sem;
+ cfs_time_t f_glimpse_time;
+ cfs_list_t f_agl_list;
+ __u64 f_agl_index;
+ } f;
+
+#define lli_size_sem u.f.f_size_sem
+#define lli_size_sem_owner u.f.f_size_sem_owner
+#define lli_symlink_name u.f.f_symlink_name
+#define lli_maxbytes u.f.f_maxbytes
+#define lli_trunc_sem u.f.f_trunc_sem
+#define lli_write_sem u.f.f_write_sem
+#define lli_async_rc u.f.f_async_rc
+#define lli_write_rc u.f.f_write_rc
+#define lli_glimpse_sem u.f.f_glimpse_sem
+#define lli_glimpse_time u.f.f_glimpse_time
+#define lli_agl_list u.f.f_agl_list
+#define lli_agl_index u.f.f_agl_index
+
+ } u;
+
+ /* XXX: For following frequent used members, although they maybe special
+ * used for non-directory object, it is some time-wasting to check
+ * whether the object is directory or not before using them. On the
+ * other hand, currently, sizeof(f) > sizeof(d), it cannot reduce
+ * the "ll_inode_info" size even if moving those members into u.f.
+ * So keep them out side.
+ *
+ * In the future, if more members are added only for directory,
+ * some of the following members can be moved into u.f.
*/
- cfs_semaphore_t lli_readdir_sem;
+ struct lov_stripe_md *lli_smd;
+ struct cl_object *lli_clob;
};
/*
#define LL_SBI_SOM_PREVIEW 0x1000 /* SOM preview mount option */
#define LL_SBI_32BIT_API 0x2000 /* generate 32 bit inodes. */
#define LL_SBI_64BIT_HASH 0x4000 /* support 64-bits dir hash/offset */
+#define LL_SBI_AGL_ENABLED 0x8000 /* enable agl */
/* default value for ll_sb_info->contention_time */
#define SBI_DEFAULT_CONTENTION_SECONDS 60
unsigned int ll_offset_process_count;
struct ll_rw_process_info ll_rw_offset_info[LL_OFFSET_HIST_MAX];
unsigned int ll_rw_offset_entry_count;
- enum stats_track_type ll_stats_track_type;
int ll_stats_track_id;
+ enum stats_track_type ll_stats_track_type;
int ll_rw_stats_on;
/* metadata stat-ahead */
* count */
atomic_t ll_sa_wrong; /* statahead thread stopped for
* low hit ratio */
+ atomic_t ll_agl_total; /* AGL thread started count */
dev_t ll_sdev_orig; /* save s_dev before assign for
* clustred nfs */
/* llite/dir.c */
-static inline void ll_put_page(struct page *page)
-{
- kunmap(page);
- page_cache_release(page);
-}
-
+void ll_release_page(struct page *page, int remove);
extern struct file_operations ll_dir_operations;
extern struct inode_operations ll_dir_inode_operations;
struct page *ll_get_dir_page(struct file *filp, struct inode *dir, __u64 hash,
- int exact, struct ll_dir_chain *chain);
+ struct ll_dir_chain *chain);
int ll_readdir(struct file *filp, void *cookie, filldir_t filldir);
int ll_get_mdt_idx(struct inode *inode);
int lookup_flags);
int ll_lookup_it_finish(struct ptlrpc_request *request,
struct lookup_intent *it, void *data);
+struct dentry *ll_find_alias(struct inode *inode, struct dentry *de);
/* llite/rw.c */
int ll_prepare_write(struct file *, struct page *, unsigned from, unsigned to);
int ll_som_update(struct inode *inode, struct md_op_data *op_data);
int ll_inode_getattr(struct inode *inode, struct obdo *obdo,
__u64 ioepoch, int sync);
-int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
+int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
struct md_open_data **mod);
void ll_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data,
struct lustre_handle *fh);
struct lookup_intent *it, struct kstat *stat);
int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat);
struct ll_file_data *ll_file_data_get(void);
-#ifndef HAVE_INODE_PERMISION_2ARGS
-int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd);
+#ifdef HAVE_GENERIC_PERMISSION_4ARGS
+int ll_inode_permission(struct inode *inode, int mask, unsigned int flags);
#else
+# ifndef HAVE_INODE_PERMISION_2ARGS
+int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd);
+# else
int ll_inode_permission(struct inode *inode, int mask);
+# endif
#endif
int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file,
int flags, struct lov_user_md *lum,
int set_default);
int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmm,
int *lmm_size, struct ptlrpc_request **request);
+#ifndef HAVE_FILE_FSYNC_2ARGS
int ll_fsync(struct file *file, struct dentry *dentry, int data);
+#else
+int ll_fsync(struct file *file, int data);
+#endif
int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
int num_bytes);
int ll_merge_lvb(struct inode *inode);
int ll_fid2path(struct obd_export *exp, void *arg);
/* llite/dcache.c */
-/* llite/namei.c */
-/**
- * protect race ll_find_aliases vs ll_revalidate_it vs ll_unhash_aliases
- */
int ll_dops_init(struct dentry *de, int block, int init_sa);
extern cfs_spinlock_t ll_lookup_lock;
extern struct dentry_operations ll_d_ops;
void ll_intent_drop_lock(struct lookup_intent *);
void ll_intent_release(struct lookup_intent *);
int ll_drop_dentry(struct dentry *dentry);
-int ll_drop_dentry(struct dentry *dentry);
void ll_unhash_aliases(struct inode *);
void ll_frob_intent(struct lookup_intent **itp, struct lookup_intent *deft);
void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry);
void ll_kill_super(struct super_block *sb);
struct inode *ll_inode_from_lock(struct ldlm_lock *lock);
void ll_clear_inode(struct inode *inode);
-int ll_setattr_raw(struct inode *inode, struct iattr *attr);
+int ll_setattr_raw(struct dentry *dentry, struct iattr *attr);
int ll_setattr(struct dentry *de, struct iattr *attr);
#ifndef HAVE_STATFS_DENTRY_PARAM
int ll_statfs(struct super_block *sb, struct kstatfs *sfs);
const char *name, int namelen,
int mode, __u32 opc, void *data);
void ll_finish_md_op_data(struct md_op_data *op_data);
+int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg);
/* llite/llite_nfs.c */
extern struct export_operations lustre_export_operations;
struct vm_area_struct *vma, unsigned long addr, size_t count);
struct vm_area_struct *our_vma(unsigned long addr, size_t count);
+static inline void ll_invalidate_page(struct page *vmpage)
+{
+ struct address_space *mapping = vmpage->mapping;
+ loff_t offset = vmpage->index << PAGE_CACHE_SHIFT;
+
+ LASSERT(PageLocked(vmpage));
+ if (mapping == NULL)
+ return;
+
+ ll_teardown_mmaps(mapping, offset, offset + CFS_PAGE_SIZE);
+ truncate_complete_page(mapping, vmpage);
+}
+
#define ll_s2sbi(sb) (s2lsi(sb)->lsi_llsbi)
/* don't need an addref as the sb_info should be holding one */
/**
* Common IO arguments for various VFS I/O interfaces.
*/
-
int cl_sb_init(struct super_block *sb);
int cl_sb_fini(struct super_block *sb);
-int cl_inode_init(struct inode *inode, struct lustre_md *md);
-void cl_inode_fini(struct inode *inode);
-
enum cl_lock_mode vvp_mode_from_vma(struct vm_area_struct *vma);
void ll_io_init(struct cl_io *io, const struct file *file, int write);
/* statahead.c */
-#define LL_SA_RPC_MIN 2
-#define LL_SA_RPC_DEF 32
-#define LL_SA_RPC_MAX 8192
+#define LL_SA_RPC_MIN 2
+#define LL_SA_RPC_DEF 32
+#define LL_SA_RPC_MAX 8192
+
+#define LL_SA_CACHE_BIT 5
+#define LL_SA_CACHE_SIZE (1 << LL_SA_CACHE_BIT)
+#define LL_SA_CACHE_MASK (LL_SA_CACHE_SIZE - 1)
/* per inode struct, for dir only */
struct ll_statahead_info {
struct inode *sai_inode;
- unsigned int sai_generation; /* generation for statahead */
cfs_atomic_t sai_refcount; /* when access this struct, hold
* refcount */
- unsigned int sai_sent; /* stat requests sent count */
- unsigned int sai_replied; /* stat requests which received
- * reply */
+ unsigned int sai_generation; /* generation for statahead */
unsigned int sai_max; /* max ahead of lookup */
- unsigned int sai_index; /* index of statahead entry */
- unsigned int sai_index_next; /* index for the next statahead
- * entry to be stated */
- unsigned int sai_hit; /* hit count */
- unsigned int sai_miss; /* miss count:
+ __u64 sai_sent; /* stat requests sent count */
+ __u64 sai_replied; /* stat requests which received
+ * reply */
+ __u64 sai_index; /* index of statahead entry */
+ __u64 sai_index_wait; /* index of entry which is the
+ * caller is waiting for */
+ __u64 sai_hit; /* hit count */
+ __u64 sai_miss; /* miss count:
* for "ls -al" case, it includes
* hidden dentry miss;
* for "ls -l" case, it does not
unsigned int sai_miss_hidden;/* "ls -al", but first dentry
* is not a hidden one */
unsigned int sai_skip_hidden;/* skipped hidden dentry count */
- unsigned int sai_ls_all:1; /* "ls -al", do stat-ahead for
+ unsigned int sai_ls_all:1, /* "ls -al", do stat-ahead for
* hidden entries */
+ sai_in_readpage:1,/* statahead is in readdir()*/
+ sai_agl_valid:1;/* AGL is valid for the dir */
cfs_waitq_t sai_waitq; /* stat-ahead wait queue */
struct ptlrpc_thread sai_thread; /* stat-ahead thread */
+ struct ptlrpc_thread sai_agl_thread; /* AGL thread */
cfs_list_t sai_entries_sent; /* entries sent out */
cfs_list_t sai_entries_received; /* entries returned */
cfs_list_t sai_entries_stated; /* entries stated */
- pid_t sai_pid; /* pid of statahead itself */
+ cfs_list_t sai_entries_agl; /* AGL entries to be sent */
+ cfs_list_t sai_cache[LL_SA_CACHE_SIZE];
+ cfs_spinlock_t sai_cache_lock[LL_SA_CACHE_SIZE];
+ cfs_atomic_t sai_cache_count; /* entry count in cache */
};
-int do_statahead_enter(struct inode *dir, struct dentry **dentry, int lookup);
-void ll_statahead_exit(struct inode *dir, struct dentry *dentry, int result);
+int do_statahead_enter(struct inode *dir, struct dentry **dentry,
+ int only_unplug);
void ll_stop_statahead(struct inode *dir, void *key);
-static inline
-void ll_statahead_mark(struct inode *dir, struct dentry *dentry)
+static inline int ll_glimpse_size(struct inode *inode)
{
- struct ll_inode_info *lli;
- struct ll_dentry_data *ldd = ll_d2d(dentry);
+ struct ll_inode_info *lli = ll_i2info(inode);
+ int rc;
+
+ cfs_down_read(&lli->lli_glimpse_sem);
+ rc = cl_glimpse_size(inode);
+ lli->lli_glimpse_time = cfs_time_current();
+ cfs_up_read(&lli->lli_glimpse_sem);
+ return rc;
+}
- /* dentry has been move to other directory, no need mark */
- if (unlikely(dir != dentry->d_parent->d_inode))
- return;
+static inline void
+ll_statahead_mark(struct inode *dir, struct dentry *dentry)
+{
+ struct ll_inode_info *lli = ll_i2info(dir);
+ struct ll_statahead_info *sai = lli->lli_sai;
+ struct ll_dentry_data *ldd = ll_d2d(dentry);
- lli = ll_i2info(dir);
/* not the same process, don't mark */
if (lli->lli_opendir_pid != cfs_curproc_pid())
return;
- cfs_spin_lock(&lli->lli_sa_lock);
- if (likely(lli->lli_sai != NULL && ldd != NULL))
- ldd->lld_sa_generation = lli->lli_sai->sai_generation;
- cfs_spin_unlock(&lli->lli_sa_lock);
+ if (sai != NULL && ldd != NULL)
+ ldd->lld_sa_generation = sai->sai_generation;
}
-static inline
-int ll_statahead_enter(struct inode *dir, struct dentry **dentryp, int lookup)
+static inline int
+ll_statahead_enter(struct inode *dir, struct dentry **dentryp, int only_unplug)
{
struct ll_inode_info *lli;
- struct ll_dentry_data *ldd = ll_d2d(*dentryp);
-
- if (unlikely(dir == NULL))
- return -EAGAIN;
+ struct ll_dentry_data *ldd;
if (ll_i2sbi(dir)->ll_sa_max == 0)
return -ENOTSUPP;
if (lli->lli_opendir_pid != cfs_curproc_pid())
return -EAGAIN;
+ ldd = ll_d2d(*dentryp);
/*
- * When "ls" a dentry, the system trigger more than once "revalidate" or
- * "lookup", for "getattr", for "getxattr", and maybe for others.
+ * When stats a dentry, the system trigger more than once "revalidate"
+ * or "lookup", for "getattr", for "getxattr", and maybe for others.
* Under patchless client mode, the operation intent is not accurate,
- * it maybe misguide the statahead thread. For example:
+ * which maybe misguide the statahead thread. For example:
* The "revalidate" call for "getattr" and "getxattr" of a dentry maybe
* have the same operation intent -- "IT_GETATTR".
* In fact, one dentry should has only one chance to interact with the
ldd->lld_sa_generation == lli->lli_sai->sai_generation)
return -EAGAIN;
- return do_statahead_enter(dir, dentryp, lookup);
+ return do_statahead_enter(dir, dentryp, only_unplug);
}
/* llite ioctl register support rountine */
int ldp_nr;
};
+static inline void cl_stats_tally(struct cl_device *dev, enum cl_req_type crt,
+ int rc)
+{
+ int opc = (crt == CRT_READ) ? LPROC_LL_OSC_READ :
+ LPROC_LL_OSC_WRITE;
+
+ ll_stats_ops_tally(ll_s2sbi(cl2ccc_dev(dev)->cdv_sb), opc, rc);
+}
+
extern ssize_t ll_direct_rw_pages(const struct lu_env *env, struct cl_io *io,
int rw, struct inode *inode,
struct ll_dio_pages *pv);
(ll_i2sbi(inode)->ll_flags & LL_SBI_NOLCK));
}
+static inline void ll_set_lock_data(struct obd_export *exp, struct inode *inode,
+ struct lookup_intent *it, __u64 *bits)
+{
+ if (!it->d.lustre.it_lock_set) {
+ CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
+ inode, inode->i_ino, inode->i_generation);
+ md_set_lock_data(exp, &it->d.lustre.it_lock_handle,
+ inode, &it->d.lustre.it_lock_bits);
+ it->d.lustre.it_lock_set = 1;
+ }
+
+ if (bits != NULL)
+ *bits = it->d.lustre.it_lock_bits;
+}
+
+static inline void ll_dentry_rehash(struct dentry *dentry, int locked)
+{
+ if (!locked) {
+ cfs_spin_lock(&ll_lookup_lock);
+ spin_lock(&dcache_lock);
+ }
+ if (d_unhashed(dentry))
+ d_rehash_cond(dentry, 0);
+ if (!locked) {
+ spin_unlock(&dcache_lock);
+ cfs_spin_unlock(&ll_lookup_lock);
+ }
+}
+
+static inline void ll_dentry_reset_flags(struct dentry *dentry, __u64 bits)
+{
+ if (bits & MDS_INODELOCK_LOOKUP &&
+ dentry->d_flags & DCACHE_LUSTRE_INVALID) {
+ lock_dentry(dentry);
+ dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
+ unlock_dentry(dentry);
+ }
+}
+
#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0)
/* Compatibility for old (1.8) compiled userspace quota code */
struct if_quotactl_18 {
git://git.whamcloud.com / fs / lustre-release.git / blobdiff