*
* Lustre Light Super operations
*
- * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
+ * Copyright (c) 2002-2004 Cluster File Systems, Inc.
*
* This file is part of Lustre, http://www.lustre.org.
*
#include <stdlib.h>
#include <string.h>
-#include <error.h>
#include <assert.h>
#include <time.h>
#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
#include <sys/queue.h>
+#ifndef __CYGWIN__
+# include <sys/statvfs.h>
+#else
+# include <sys/statfs.h>
+#endif
+#ifdef HAVE_XTIO_H
+#include <xtio.h>
+#endif
#include <sysio.h>
#include <fs.h>
#include <mount.h>
#include <inode.h>
+#ifdef HAVE_FILE_H
#include <file.h>
+#endif
+
+#undef LIST_HEAD
#include "llite_lib.h"
+#ifndef MAY_EXEC
+#define MAY_EXEC 1
+#define MAY_WRITE 2
+#define MAY_READ 4
+#endif
+
+#define S_IXUGO (S_IXUSR|S_IXGRP|S_IXOTH)
+
+static int ll_permission(struct inode *inode, int mask)
+{
+ struct intnl_stat *st = llu_i2stat(inode);
+ mode_t mode = st->st_mode;
+
+ if (current->fsuid == st->st_uid)
+ mode >>= 6;
+ else if (in_group_p(st->st_gid))
+ mode >>= 3;
+
+ if ((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask)
+ return 0;
+
+ if ((mask & (MAY_READ|MAY_WRITE)) ||
+ (st->st_mode & S_IXUGO))
+ if (capable(CAP_DAC_OVERRIDE))
+ return 0;
+
+ if (mask == MAY_READ ||
+ (S_ISDIR(st->st_mode) && !(mask & MAY_WRITE))) {
+ if (capable(CAP_DAC_READ_SEARCH))
+ return 0;
+ }
+
+ return -EACCES;
+}
+
static void llu_fsop_gone(struct filesys *fs)
{
- /* FIXME */
+ struct llu_sb_info *sbi = (struct llu_sb_info *) fs->fs_private;
+ struct obd_device *obd = class_exp2obd(sbi->ll_mdc_exp);
+ int next = 0;
+ ENTRY;
+
+ list_del(&sbi->ll_conn_chain);
+ obd_disconnect(sbi->ll_osc_exp);
+ obd_disconnect(sbi->ll_mdc_exp);
+
+ while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
+ class_manual_cleanup(obd);
+
+ OBD_FREE(sbi, sizeof(*sbi));
+
+ liblustre_wait_idle();
+ EXIT;
}
static struct inode_ops llu_inode_ops;
struct lov_stripe_md *lsm)
{
struct llu_inode_info *lli = llu_i2info(inode);
+ struct intnl_stat *st = llu_i2stat(inode);
LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
if (lsm != NULL) {
- if (lli->lli_smd == NULL)
+ if (lli->lli_smd == NULL) {
lli->lli_smd = lsm;
- else
- LASSERT (!memcmp (lli->lli_smd, lsm,
- sizeof (*lsm)));
+ lli->lli_maxbytes = lsm->lsm_maxbytes;
+ if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
+ lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
+ } else {
+ if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
+ CERROR("lsm mismatch for inode %lld\n",
+ (long long)st->st_ino);
+ LBUG();
+ }
+ }
}
if (body->valid & OBD_MD_FLID)
- lli->lli_st_ino = body->ino;
- if (body->valid & OBD_MD_FLATIME)
- LTIME_S(lli->lli_st_atime) = body->atime;
- if (body->valid & OBD_MD_FLMTIME)
- LTIME_S(lli->lli_st_mtime) = body->mtime;
- if (body->valid & OBD_MD_FLCTIME)
- LTIME_S(lli->lli_st_ctime) = body->ctime;
+ st->st_ino = body->ino;
+ if (body->valid & OBD_MD_FLATIME &&
+ body->atime > LTIME_S(st->st_atime))
+ LTIME_S(st->st_atime) = body->atime;
+
+ /* mtime is always updated with ctime, but can be set in past.
+ As write and utime(2) may happen within 1 second, and utime's
+ mtime has a priority over write's one, so take mtime from mds
+ for the same ctimes. */
+ if (body->valid & OBD_MD_FLCTIME &&
+ body->ctime >= LTIME_S(st->st_ctime)) {
+ LTIME_S(st->st_ctime) = body->ctime;
+ if (body->valid & OBD_MD_FLMTIME)
+ LTIME_S(st->st_mtime) = body->mtime;
+ }
if (body->valid & OBD_MD_FLMODE)
- lli->lli_st_mode = (lli->lli_st_mode & S_IFMT)|(body->mode & ~S_IFMT);
+ st->st_mode = (st->st_mode & S_IFMT)|(body->mode & ~S_IFMT);
if (body->valid & OBD_MD_FLTYPE)
- lli->lli_st_mode = (lli->lli_st_mode & ~S_IFMT)|(body->mode & S_IFMT);
+ st->st_mode = (st->st_mode & ~S_IFMT)|(body->mode & S_IFMT);
+ if (S_ISREG(st->st_mode))
+ st->st_blksize = min(2UL * PTLRPC_MAX_BRW_SIZE, LL_MAX_BLKSIZE);
+ else
+ st->st_blksize = 4096;
if (body->valid & OBD_MD_FLUID)
- lli->lli_st_uid = body->uid;
+ st->st_uid = body->uid;
if (body->valid & OBD_MD_FLGID)
- lli->lli_st_gid = body->gid;
- if (body->valid & OBD_MD_FLFLAGS)
- lli->lli_st_flags = body->flags;
+ st->st_gid = body->gid;
if (body->valid & OBD_MD_FLNLINK)
- lli->lli_st_nlink = body->nlink;
- if (body->valid & OBD_MD_FLGENER)
- lli->lli_st_generation = body->generation;
+ st->st_nlink = body->nlink;
if (body->valid & OBD_MD_FLRDEV)
- lli->lli_st_rdev = body->rdev;
+ st->st_rdev = body->rdev;
if (body->valid & OBD_MD_FLSIZE)
- lli->lli_st_size = body->size;
+ st->st_size = body->size;
if (body->valid & OBD_MD_FLBLOCKS)
- lli->lli_st_blocks = body->blocks;
+ st->st_blocks = body->blocks;
+ if (body->valid & OBD_MD_FLFLAGS)
+ lli->lli_st_flags = body->flags;
+ if (body->valid & OBD_MD_FLGENER)
+ lli->lli_st_generation = body->generation;
/* fillin fid */
if (body->valid & OBD_MD_FLID)
void obdo_to_inode(struct inode *dst, struct obdo *src, obd_flag valid)
{
struct llu_inode_info *lli = llu_i2info(dst);
+ struct intnl_stat *st = llu_i2stat(dst);
valid &= src->o_valid;
+ if (valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME))
+ CDEBUG(D_INODE,"valid "LPX64", cur time %lu/%lu, new %lu/%lu\n",
+ src->o_valid,
+ LTIME_S(st->st_mtime), LTIME_S(st->st_ctime),
+ (long)src->o_mtime, (long)src->o_ctime);
+
if (valid & OBD_MD_FLATIME)
- LTIME_S(lli->lli_st_atime) = src->o_atime;
+ LTIME_S(st->st_atime) = src->o_atime;
if (valid & OBD_MD_FLMTIME)
- LTIME_S(lli->lli_st_mtime) = src->o_mtime;
- if (valid & OBD_MD_FLCTIME && src->o_ctime > LTIME_S(lli->lli_st_ctime))
- LTIME_S(lli->lli_st_ctime) = src->o_ctime;
+ LTIME_S(st->st_mtime) = src->o_mtime;
+ if (valid & OBD_MD_FLCTIME && src->o_ctime > LTIME_S(st->st_ctime))
+ LTIME_S(st->st_ctime) = src->o_ctime;
if (valid & OBD_MD_FLSIZE)
- lli->lli_st_size = src->o_size;
+ st->st_size = src->o_size;
if (valid & OBD_MD_FLBLOCKS) /* allocation of space */
- lli->lli_st_blocks = src->o_blocks;
+ st->st_blocks = src->o_blocks;
if (valid & OBD_MD_FLBLKSZ)
- lli->lli_st_blksize = src->o_blksize;
+ st->st_blksize = src->o_blksize;
if (valid & OBD_MD_FLTYPE)
- lli->lli_st_mode = (lli->lli_st_mode & ~S_IFMT) | (src->o_mode & S_IFMT);
+ st->st_mode = (st->st_mode & ~S_IFMT) | (src->o_mode & S_IFMT);
if (valid & OBD_MD_FLMODE)
- lli->lli_st_mode = (lli->lli_st_mode & S_IFMT) | (src->o_mode & ~S_IFMT);
+ st->st_mode = (st->st_mode & S_IFMT) | (src->o_mode & ~S_IFMT);
if (valid & OBD_MD_FLUID)
- lli->lli_st_uid = src->o_uid;
+ st->st_uid = src->o_uid;
if (valid & OBD_MD_FLGID)
- lli->lli_st_gid = src->o_gid;
+ st->st_gid = src->o_gid;
if (valid & OBD_MD_FLFLAGS)
lli->lli_st_flags = src->o_flags;
- if (valid & OBD_MD_FLNLINK)
- lli->lli_st_nlink = src->o_nlink;
if (valid & OBD_MD_FLGENER)
lli->lli_st_generation = src->o_generation;
- if (valid & OBD_MD_FLRDEV)
- lli->lli_st_rdev = src->o_rdev;
}
+#define S_IRWXUGO (S_IRWXU|S_IRWXG|S_IRWXO)
+#define S_IALLUGO (S_ISUID|S_ISGID|S_ISVTX|S_IRWXUGO)
+
void obdo_from_inode(struct obdo *dst, struct inode *src, obd_flag valid)
{
struct llu_inode_info *lli = llu_i2info(src);
+ struct intnl_stat *st = llu_i2stat(src);
+ obd_flag newvalid = 0;
- if (valid & OBD_MD_FLATIME)
- dst->o_atime = LTIME_S(lli->lli_st_atime);
- if (valid & OBD_MD_FLMTIME)
- dst->o_mtime = LTIME_S(lli->lli_st_mtime);
- if (valid & OBD_MD_FLCTIME)
- dst->o_ctime = LTIME_S(lli->lli_st_ctime);
- if (valid & OBD_MD_FLSIZE)
- dst->o_size = lli->lli_st_size;
- if (valid & OBD_MD_FLBLOCKS) /* allocation of space */
- dst->o_blocks = lli->lli_st_blocks;
- if (valid & OBD_MD_FLBLKSZ)
- dst->o_blksize = lli->lli_st_blksize;
- if (valid & OBD_MD_FLTYPE)
- dst->o_mode = (dst->o_mode & ~S_IFMT) | (lli->lli_st_mode & S_IFMT);
- if (valid & OBD_MD_FLMODE)
- dst->o_mode = (dst->o_mode & S_IFMT) | (lli->lli_st_mode & ~S_IFMT);
- if (valid & OBD_MD_FLUID)
- dst->o_uid = lli->lli_st_uid;
- if (valid & OBD_MD_FLGID)
- dst->o_gid = lli->lli_st_gid;
- if (valid & OBD_MD_FLFLAGS)
+ if (valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME))
+ CDEBUG(D_INODE, "valid %x, new time %lu/%lu\n",
+ valid, LTIME_S(st->st_mtime),
+ LTIME_S(st->st_ctime));
+
+ if (valid & OBD_MD_FLATIME) {
+ dst->o_atime = LTIME_S(st->st_atime);
+ newvalid |= OBD_MD_FLATIME;
+ }
+ if (valid & OBD_MD_FLMTIME) {
+ dst->o_mtime = LTIME_S(st->st_mtime);
+ newvalid |= OBD_MD_FLMTIME;
+ }
+ if (valid & OBD_MD_FLCTIME) {
+ dst->o_ctime = LTIME_S(st->st_ctime);
+ newvalid |= OBD_MD_FLCTIME;
+ }
+ if (valid & OBD_MD_FLSIZE) {
+ dst->o_size = st->st_size;
+ newvalid |= OBD_MD_FLSIZE;
+ }
+ if (valid & OBD_MD_FLBLOCKS) { /* allocation of space (x512 bytes) */
+ dst->o_blocks = st->st_blocks;
+ newvalid |= OBD_MD_FLBLOCKS;
+ }
+ if (valid & OBD_MD_FLBLKSZ) { /* optimal block size */
+ dst->o_blksize = st->st_blksize;
+ newvalid |= OBD_MD_FLBLKSZ;
+ }
+ if (valid & OBD_MD_FLTYPE) {
+ dst->o_mode = (dst->o_mode & S_IALLUGO)|(st->st_mode & S_IFMT);
+ newvalid |= OBD_MD_FLTYPE;
+ }
+ if (valid & OBD_MD_FLMODE) {
+ dst->o_mode = (dst->o_mode & S_IFMT)|(st->st_mode & S_IALLUGO);
+ newvalid |= OBD_MD_FLMODE;
+ }
+ if (valid & OBD_MD_FLUID) {
+ dst->o_uid = st->st_uid;
+ newvalid |= OBD_MD_FLUID;
+ }
+ if (valid & OBD_MD_FLGID) {
+ dst->o_gid = st->st_gid;
+ newvalid |= OBD_MD_FLGID;
+ }
+ if (valid & OBD_MD_FLFLAGS) {
dst->o_flags = lli->lli_st_flags;
- if (valid & OBD_MD_FLNLINK)
- dst->o_nlink = lli->lli_st_nlink;
- if (valid & OBD_MD_FLGENER)
+ newvalid |= OBD_MD_FLFLAGS;
+ }
+ if (valid & OBD_MD_FLGENER) {
dst->o_generation = lli->lli_st_generation;
- if (valid & OBD_MD_FLRDEV)
- dst->o_rdev = (__u32)(lli->lli_st_rdev);
+ newvalid |= OBD_MD_FLGENER;
+ }
+ if (valid & OBD_MD_FLFID) {
+ dst->o_fid = st->st_ino;
+ newvalid |= OBD_MD_FLFID;
+ }
- dst->o_valid |= (valid & ~OBD_MD_FLID);
+ dst->o_valid |= newvalid;
}
-int llu_inode_getattr(struct inode *inode, struct lov_stripe_md *lsm,
- char *ostdata)
+/*
+ * really does the getattr on the inode and updates its fields
+ */
+int llu_inode_getattr(struct inode *inode, struct lov_stripe_md *lsm)
{
- struct llu_sb_info *sbi = llu_i2sbi(inode);
- struct obdo oa;
+ struct llu_inode_info *lli = llu_i2info(inode);
+ struct obd_export *exp = llu_i2obdexp(inode);
+ struct ptlrpc_request_set *set;
+ struct obd_info oinfo = { { { 0 } } };
+ struct obdo oa = { 0 };
+ obd_flag refresh_valid;
int rc;
ENTRY;
LASSERT(lsm);
- LASSERT(sbi);
+ LASSERT(lli);
- memset(&oa, 0, sizeof oa);
+ oinfo.oi_md = lsm;
+ oinfo.oi_oa = &oa;
oa.o_id = lsm->lsm_object_id;
oa.o_mode = S_IFREG;
oa.o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLSIZE |
- OBD_MD_FLBLOCKS | OBD_MD_FLMTIME | OBD_MD_FLCTIME;
+ OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ | OBD_MD_FLMTIME |
+ OBD_MD_FLCTIME;
- if (ostdata != NULL) {
- memcpy(&oa.o_inline, ostdata, FD_OSTDATA_SIZE);
- oa.o_valid |= OBD_MD_FLHANDLE;
+ set = ptlrpc_prep_set();
+ if (set == NULL) {
+ CERROR ("ENOMEM allocing request set\n");
+ rc = -ENOMEM;
+ } else {
+ rc = obd_getattr_async(exp, &oinfo, set);
+ if (rc == 0)
+ rc = ptlrpc_set_wait(set);
+ ptlrpc_set_destroy(set);
}
-
- rc = obd_getattr(&sbi->ll_osc_conn, &oa, lsm);
if (rc)
RETURN(rc);
- obdo_to_inode(inode, &oa, OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
- OBD_MD_FLMTIME | OBD_MD_FLCTIME);
+ refresh_valid = OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ | OBD_MD_FLMTIME |
+ OBD_MD_FLCTIME | OBD_MD_FLSIZE;
+
+ obdo_refresh_inode(inode, &oa, refresh_valid);
RETURN(0);
}
-struct inode* llu_new_inode(struct filesys *fs, ino_t ino, mode_t mode)
+static struct inode* llu_new_inode(struct filesys *fs,
+ struct ll_fid *fid)
{
- struct inode *inode;
+ struct inode *inode;
struct llu_inode_info *lli;
+ struct intnl_stat st = {
+ .st_dev = 0,
+#ifndef AUTOMOUNT_FILE_NAME
+ .st_mode = fid->f_type & S_IFMT,
+#else
+ .st_mode = fid->f_type /* all of the bits! */
+#endif
+ .st_uid = geteuid(),
+ .st_gid = getegid(),
+ };
OBD_ALLOC(lli, sizeof(*lli));
if (!lli)
lli->lli_smd = NULL;
lli->lli_symlink_name = NULL;
lli->lli_flags = 0;
- INIT_LIST_HEAD(&lli->lli_read_extents);
+ lli->lli_maxbytes = (__u64)(~0UL);
lli->lli_file_data = NULL;
- /* could file_identifier be 0 ? FIXME */
- inode = _sysio_i_new(fs, ino, NULL,
-#ifndef AUTOMOUNT_FILE_NAME
- mode & S_IFMT,
-#else
- mode, /* all of the bits! */
-#endif
- 0,
- &llu_inode_ops, lli);
+ lli->lli_sysio_fid.fid_data = &lli->lli_fid;
+ lli->lli_sysio_fid.fid_len = sizeof(lli->lli_fid);
+ lli->lli_fid = *fid;
+
+ /* file identifier is needed by functions like _sysio_i_find() */
+ inode = _sysio_i_new(fs, &lli->lli_sysio_fid,
+ &st, 0, &llu_inode_ops, lli);
- if (!inode)
- OBD_FREE(lli, sizeof(*lli));
+ if (!inode)
+ OBD_FREE(lli, sizeof(*lli));
return inode;
}
-static int llu_iop_lookup(struct pnode *pnode,
- struct inode **inop,
- struct intent *intnt __IS_UNUSED,
- const char *path __IS_UNUSED)
+static int llu_have_md_lock(struct inode *inode, __u64 lockpart)
{
- struct pnode_base *pb_dir = pnode->p_parent->p_base;
- struct ptlrpc_request *request = NULL;
- struct llu_sb_info *sbi = llu_i2sbi(pb_dir->pb_ino);
- struct ll_fid *fid = &llu_i2info(pb_dir->pb_ino)->lli_fid;
- struct qstr *name = &pnode->p_base->pb_name;
- struct mds_body *body;
- unsigned long valid;
- char *pname;
- int rc, easize;
- struct ll_read_inode2_cookie lic = {.lic_body = NULL, .lic_lsm = NULL};
-
- /* the mount root inode have no name, so don't call
- * remote in this case. but probably we need revalidate
- * it here? FIXME */
- if (pnode->p_mount->mnt_root == pnode) {
- struct inode *i = pnode->p_base->pb_ino;
- I_REF(i);
- *inop = i;
- return 0;
- }
+ struct llu_sb_info *sbi = llu_i2sbi(inode);
+ struct llu_inode_info *lli = llu_i2info(inode);
+ struct lustre_handle lockh;
+ struct ldlm_res_id res_id = { .name = {0} };
+ struct obd_device *obddev;
+ ldlm_policy_data_t policy = { .l_inodebits = { lockpart } };
+ int flags;
+ ENTRY;
- if (!name->len)
- return -EINVAL;
+ LASSERT(inode);
- /* mdc_getattr_name require NULL-terminated name */
- OBD_ALLOC(pname, name->len + 1);
- if (!pname)
- return -ENOMEM;
- memcpy(pname, name->name, name->len);
- pname[name->len] = 0;
-
- valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLSIZE;
-
- /* FIXME before getattr_name, we don't know whether
- * the inode we are finding is regular or not, so here
- * we blindly require server feed in EA data */
- easize = obd_size_diskmd(&sbi->ll_osc_conn, NULL);
- valid |= OBD_MD_FLEASIZE;
-
- rc = mdc_getattr_name(&sbi->ll_mdc_conn, fid,
- pname, name->len + 1,
- valid, easize, &request);
- if (rc < 0) {
- CERROR("mdc_getattr_name: %d\n", rc);
- rc = -ENOENT;
- goto out;
+ obddev = sbi->ll_mdc_exp->exp_obd;
+ res_id.name[0] = llu_i2stat(inode)->st_ino;
+ res_id.name[1] = lli->lli_st_generation;
+
+ CDEBUG(D_INFO, "trying to match res "LPU64"\n", res_id.name[0]);
+
+ flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
+ if (ldlm_lock_match(obddev->obd_namespace, flags, &res_id, LDLM_IBITS,
+ &policy, LCK_PW | LCK_PR, &lockh)) {
+ RETURN(1);
}
- body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
+ RETURN(0);
+}
- *inop = llu_new_inode(pnode->p_mount->mnt_fs, body->ino, body->mode);
- if (!inop)
- goto out;
+static int llu_inode_revalidate(struct inode *inode)
+{
+ struct lov_stripe_md *lsm = NULL;
+ ENTRY;
- lic.lic_body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*lic.lic_body));
- LASSERT (lic.lic_body != NULL);
- LASSERT_REPSWABBED (request, 0);
+ if (!inode) {
+ CERROR("REPORT THIS LINE TO PETER\n");
+ RETURN(0);
+ }
- if (S_ISREG(lic.lic_body->mode) &&
- lic.lic_body->valid & OBD_MD_FLEASIZE) {
- struct lov_mds_md *lmm;
- int lmm_size;
- int rc;
-
- lmm_size = lic.lic_body->eadatasize;
- if (lmm_size == 0) {
- CERROR ("OBD_MD_FLEASIZE set but eadatasize 0\n");
- RETURN (-EPROTO);
+ if (!llu_have_md_lock(inode, MDS_INODELOCK_UPDATE)) {
+ struct lustre_md md;
+ struct ptlrpc_request *req = NULL;
+ struct llu_sb_info *sbi = llu_i2sbi(inode);
+ struct ll_fid fid;
+ unsigned long valid = OBD_MD_FLGETATTR;
+ int rc, ealen = 0;
+
+ /* Why don't we update all valid MDS fields here, if we're
+ * doing an RPC anyways? -phil */
+ if (S_ISREG(llu_i2stat(inode)->st_mode)) {
+ ealen = obd_size_diskmd(sbi->ll_osc_exp, NULL);
+ valid |= OBD_MD_FLEASIZE;
}
- lmm = lustre_msg_buf(request->rq_repmsg, 0 + 1, lmm_size);
- LASSERT(lmm != NULL);
- LASSERT_REPSWABBED (request, 0 + 1);
-
- rc = obd_unpackmd (&sbi->ll_osc_conn,
- &lic.lic_lsm, lmm, lmm_size);
- if (rc < 0) {
- CERROR ("Error %d unpacking eadata\n", rc);
- RETURN (rc);
+ ll_inode2fid(&fid, inode);
+ rc = mdc_getattr(sbi->ll_mdc_exp, &fid, valid, ealen, &req);
+ if (rc) {
+ CERROR("failure %d inode %llu\n", rc,
+ (long long)llu_i2stat(inode)->st_ino);
+ RETURN(-abs(rc));
+ }
+ rc = mdc_req2lustre_md(req, REPLY_REC_OFF, sbi->ll_osc_exp,&md);
+
+ /* XXX Too paranoid? */
+ if (((md.body->valid ^ valid) & OBD_MD_FLEASIZE) &&
+ !((md.body->valid & OBD_MD_FLNLINK) &&
+ (md.body->nlink == 0))) {
+ CERROR("Asked for %s eadata but got %s (%d)\n",
+ (valid & OBD_MD_FLEASIZE) ? "some" : "no",
+ (md.body->valid & OBD_MD_FLEASIZE) ? "some":"none",
+ md.body->eadatasize);
+ }
+ if (rc) {
+ ptlrpc_req_finished(req);
+ RETURN(rc);
}
- LASSERT (rc >= sizeof (*lic.lic_lsm));
- } else {
- lic.lic_lsm = NULL;
- }
- llu_update_inode(*inop, body, lic.lic_lsm);
+ llu_update_inode(inode, md.body, md.lsm);
+ if (md.lsm != NULL && llu_i2info(inode)->lli_smd != md.lsm)
+ obd_free_memmd(sbi->ll_osc_exp, &md.lsm);
- if (llu_i2info(*inop)->lli_smd) {
- rc = llu_inode_getattr(*inop, llu_i2info(*inop)->lli_smd, NULL);
- if (rc)
- _sysio_i_gone(*inop);
+ if (md.body->valid & OBD_MD_FLSIZE)
+ set_bit(LLI_F_HAVE_MDS_SIZE_LOCK,
+ &llu_i2info(inode)->lli_flags);
+ ptlrpc_req_finished(req);
}
-out:
- ptlrpc_req_finished(request);
- OBD_FREE(pname, name->len + 1);
+ lsm = llu_i2info(inode)->lli_smd;
+ if (!lsm) /* object not yet allocated, don't validate size */
+ RETURN(0);
- return rc;
+ /* ll_glimpse_size will prefer locally cached writes if they extend
+ * the file */
+ RETURN(llu_glimpse_size(inode));
+}
+
+static void copy_stat_buf(struct inode *ino, struct intnl_stat *b)
+{
+ *b = *llu_i2stat(ino);
}
static int llu_iop_getattr(struct pnode *pno,
struct inode *ino,
struct intnl_stat *b)
{
- struct llu_inode_info *lli = llu_i2info(ino);
+ int rc;
+ ENTRY;
- b->st_dev = lli->lli_st_dev;
- b->st_ino = lli->lli_st_ino;
- b->st_mode = lli->lli_st_mode;
- b->st_nlink = lli->lli_st_nlink;
- b->st_uid = lli->lli_st_uid;
- b->st_gid = lli->lli_st_gid;
- b->st_rdev = lli->lli_st_rdev;
- b->st_size = lli->lli_st_size;
- b->st_blksize = lli->lli_st_blksize;
- b->st_blocks = lli->lli_st_blocks;
- b->st_atime = lli->lli_st_atime;
- b->st_mtime = lli->lli_st_mtime;
- b->st_ctime = lli->lli_st_ctime;
+ liblustre_wait_event(0);
- return 0;
+ if (!ino) {
+ LASSERT(pno);
+ LASSERT(pno->p_base->pb_ino);
+ ino = pno->p_base->pb_ino;
+ } else {
+ LASSERT(!pno || pno->p_base->pb_ino == ino);
+ }
+
+ /* libsysio might call us directly without intent lock,
+ * we must re-fetch the attrs here
+ */
+ rc = llu_inode_revalidate(ino);
+ if (!rc) {
+ copy_stat_buf(ino, b);
+ LASSERT(!llu_i2info(ino)->lli_it);
+ }
+
+ liblustre_wait_event(0);
+ RETURN(rc);
}
-int llu_mdc_cancel_unused(struct lustre_handle *conn,
- struct llu_inode_info *lli,
- int flags)
+static int null_if_equal(struct ldlm_lock *lock, void *data)
{
- struct ldlm_res_id res_id =
- { .name = {lli->lli_st_ino, lli->lli_st_generation} };
- struct obd_device *obddev = class_conn2obd(conn);
- ENTRY;
- RETURN(ldlm_cli_cancel_unused(obddev->obd_namespace, &res_id, flags));
+ if (data == lock->l_ast_data) {
+ lock->l_ast_data = NULL;
+
+ if (lock->l_req_mode != lock->l_granted_mode)
+ LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
+ }
+
+ return LDLM_ITER_CONTINUE;
}
-static void llu_clear_inode(struct inode *inode)
+void llu_clear_inode(struct inode *inode)
{
- struct llu_sb_info *sbi = llu_i2sbi(inode);
+ struct ll_fid fid;
struct llu_inode_info *lli = llu_i2info(inode);
- int rc;
+ struct llu_sb_info *sbi = llu_i2sbi(inode);
ENTRY;
- CDEBUG(D_INODE, "clear inode: %lu\n", lli->lli_st_ino);
- rc = llu_mdc_cancel_unused(&sbi->ll_mdc_conn, lli,
- LDLM_FL_NO_CALLBACK);
- if (rc < 0) {
- CERROR("ll_mdc_cancel_unused: %d\n", rc);
- /* XXX FIXME do something dramatic */
- }
+ CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu/%lu(%p)\n",
+ (long long)llu_i2stat(inode)->st_ino, lli->lli_st_generation,
+ inode);
- if (lli->lli_smd) {
- rc = obd_cancel_unused(&sbi->ll_osc_conn, lli->lli_smd, 0);
- if (rc < 0) {
- CERROR("obd_cancel_unused: %d\n", rc);
- /* XXX FIXME do something dramatic */
- }
- }
+ ll_inode2fid(&fid, inode);
+ clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(lli->lli_flags));
+ mdc_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
if (lli->lli_smd)
- obd_free_memmd(&sbi->ll_osc_conn, &lli->lli_smd);
+ obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
+ null_if_equal, inode);
+
+ if (lli->lli_smd) {
+ obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
+ lli->lli_smd = NULL;
+ }
if (lli->lli_symlink_name) {
OBD_FREE(lli->lli_symlink_name,
void llu_iop_gone(struct inode *inode)
{
struct llu_inode_info *lli = llu_i2info(inode);
+ ENTRY;
+ liblustre_wait_event(0);
llu_clear_inode(inode);
OBD_FREE(lli, sizeof(*lli));
+ EXIT;
}
-static int llu_setattr_raw(struct inode *inode, struct iattr *attr)
+static int inode_setattr(struct inode * inode, struct iattr * attr)
{
- struct ptlrpc_request *request = NULL;
+ unsigned int ia_valid = attr->ia_valid;
+ struct intnl_stat *st = llu_i2stat(inode);
+ int error = 0;
+
+ /*
+ * inode_setattr() is only ever invoked with ATTR_SIZE (by
+ * llu_setattr_raw()) when file has no bodies. Check this.
+ */
+ LASSERT(ergo(ia_valid & ATTR_SIZE, llu_i2info(inode)->lli_smd == NULL));
+
+ if (ia_valid & ATTR_SIZE)
+ st->st_size = attr->ia_size;
+ if (ia_valid & ATTR_UID)
+ st->st_uid = attr->ia_uid;
+ if (ia_valid & ATTR_GID)
+ st->st_gid = attr->ia_gid;
+ if (ia_valid & ATTR_ATIME)
+ st->st_atime = attr->ia_atime;
+ if (ia_valid & ATTR_MTIME)
+ st->st_mtime = attr->ia_mtime;
+ if (ia_valid & ATTR_CTIME)
+ st->st_ctime = attr->ia_ctime;
+ if (ia_valid & ATTR_MODE) {
+ st->st_mode = attr->ia_mode;
+ if (!in_group_p(st->st_gid) && !capable(CAP_FSETID))
+ st->st_mode &= ~S_ISGID;
+ }
+ /* mark_inode_dirty(inode); */
+ return error;
+}
+
+/* If this inode has objects allocated to it (lsm != NULL), then the OST
+ * object(s) determine the file size and mtime. Otherwise, the MDS will
+ * keep these values until such a time that objects are allocated for it.
+ * We do the MDS operations first, as it is checking permissions for us.
+ * We don't to the MDS RPC if there is nothing that we want to store there,
+ * otherwise there is no harm in updating mtime/atime on the MDS if we are
+ * going to do an RPC anyways.
+ *
+ * If we are doing a truncate, we will send the mtime and ctime updates
+ * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
+ * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
+ * at the same time.
+ */
+int llu_setattr_raw(struct inode *inode, struct iattr *attr)
+{
+ struct lov_stripe_md *lsm = llu_i2info(inode)->lli_smd;
struct llu_sb_info *sbi = llu_i2sbi(inode);
- struct llu_inode_info *lli = llu_i2info(inode);
+ struct intnl_stat *st = llu_i2stat(inode);
+ struct ptlrpc_request *request = NULL;
struct mdc_op_data op_data;
- int err = 0;
+ int ia_valid = attr->ia_valid;
+ int rc = 0;
ENTRY;
- CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", lli->lli_st_ino);
- /* if need truncate, do it at first */
- if (attr->ia_valid & ATTR_SIZE) {
- printf("************* don't support truncate now !!!!!!!!\n");
- LBUG();
+ CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu\n", (long long)st->st_ino);
+
+ if (ia_valid & ATTR_SIZE) {
+ if (attr->ia_size > ll_file_maxbytes(inode)) {
+ CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
+ (long long)attr->ia_size,
+ ll_file_maxbytes(inode));
+ RETURN(-EFBIG);
+ }
+
+ attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
}
- /* Don't send size changes to MDS to avoid "fast EA" problems, and
- * also avoid a pointless RPC (we get file size from OST anyways).
- */
- attr->ia_valid &= ~ATTR_SIZE;
- if (!attr->ia_valid)
- RETURN(0);
+ /* We mark all of the fields "set" so MDS/OST does not re-set them */
+ if (attr->ia_valid & ATTR_CTIME) {
+ attr->ia_ctime = CURRENT_TIME;
+ attr->ia_valid |= ATTR_CTIME_SET;
+ }
+ if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
+ attr->ia_atime = CURRENT_TIME;
+ attr->ia_valid |= ATTR_ATIME_SET;
+ }
+ if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
+ attr->ia_mtime = CURRENT_TIME;
+ attr->ia_valid |= ATTR_MTIME_SET;
+ }
+ if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
+ /* To avoid stale mtime on mds, obtain it from ost and send
+ to mds. */
+ rc = llu_glimpse_size(inode);
+ if (rc)
+ RETURN(rc);
+
+ attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
+ attr->ia_mtime = inode->i_stbuf.st_mtime;
+ }
- llu_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
+ if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
+ CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
+ LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
+ LTIME_S(CURRENT_TIME));
+ if (lsm)
+ attr->ia_valid &= ~ATTR_SIZE;
+
+ /* If only OST attributes being set on objects, don't do MDS RPC.
+ * In that case, we need to check permissions and update the local
+ * inode ourselves so we can call obdo_from_inode() always. */
+ if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
+ struct lustre_md md;
+ llu_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
+
+ rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
+ attr, NULL, 0, NULL, 0, &request);
+
+ if (rc) {
+ ptlrpc_req_finished(request);
+ if (rc != -EPERM && rc != -EACCES)
+ CERROR("mdc_setattr fails: rc = %d\n", rc);
+ RETURN(rc);
+ }
- err = mdc_setattr(&sbi->ll_mdc_conn, &op_data,
- attr, NULL, 0, &request);
- if (err)
- CERROR("mdc_setattr fails: err = %d\n", err);
+ rc = mdc_req2lustre_md(request, REPLY_REC_OFF, sbi->ll_osc_exp,
+ &md);
+ if (rc) {
+ ptlrpc_req_finished(request);
+ RETURN(rc);
+ }
- ptlrpc_req_finished(request);
+ /* We call inode_setattr to adjust timestamps.
+ * If there is at least some data in file, we cleared ATTR_SIZE
+ * above to avoid invoking vmtruncate, otherwise it is important
+ * to call vmtruncate in inode_setattr to update inode->i_size
+ * (bug 6196) */
+ inode_setattr(inode, attr);
+ llu_update_inode(inode, md.body, md.lsm);
+ ptlrpc_req_finished(request);
+
+ if (!lsm || !S_ISREG(st->st_mode)) {
+ CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
+ RETURN(0);
+ }
+ } else {
+ /* The OST doesn't check permissions, but the alternative is
+ * a gratuitous RPC to the MDS. We already rely on the client
+ * to do read/write/truncate permission checks, so is mtime OK?
+ */
+ if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
+ /* from sys_utime() */
+ if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
+ if (current->fsuid != st->st_uid &&
+ (rc = ll_permission(inode, MAY_WRITE)) != 0)
+ RETURN(rc);
+ } else {
+ /* from inode_change_ok() */
+ if (current->fsuid != st->st_uid &&
+ !capable(CAP_FOWNER))
+ RETURN(-EPERM);
+ }
+ }
+
+ /* Won't invoke llu_vmtruncate(), as we already cleared
+ * ATTR_SIZE */
+ inode_setattr(inode, attr);
+ }
+
+ if (ia_valid & ATTR_SIZE) {
+ ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
+ OBD_OBJECT_EOF} };
+ struct lustre_handle lockh = { 0, };
+ struct lustre_handle match_lockh = { 0, };
+
+ int err;
+ int flags = LDLM_FL_TEST_LOCK; /* for assertion check below */
+ int lock_mode;
+ obd_flag obd_flags;
+
+ /* check that there are no matching locks */
+ LASSERT(obd_match(sbi->ll_osc_exp, lsm, LDLM_EXTENT, &policy,
+ LCK_PW, &flags, inode, &match_lockh) <= 0);
+
+ /* XXX when we fix the AST intents to pass the discard-range
+ * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
+ * XXX here. */
+ flags = (attr->ia_size == 0) ? LDLM_AST_DISCARD_DATA : 0;
+
+ if (sbi->ll_lco.lco_flags & OBD_CONNECT_TRUNCLOCK) {
+ lock_mode = LCK_NL;
+ obd_flags = OBD_FL_TRUNCLOCK;
+ CDEBUG(D_INODE, "delegating locking to the OST");
+ } else {
+ lock_mode = LCK_PW;
+ obd_flags = 0;
+ }
- if (S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_MTIME_SET) {
- struct lov_stripe_md *lsm = lli->lli_smd;
+ /* with lock_mode == LK_NL no lock is taken. */
+ rc = llu_extent_lock(NULL, inode, lsm, lock_mode, &policy,
+ &lockh, flags);
+ if (rc != ELDLM_OK) {
+ if (rc > 0)
+ RETURN(-ENOLCK);
+ RETURN(rc);
+ }
+
+ rc = llu_vmtruncate(inode, attr->ia_size, obd_flags);
+
+ /* unlock now as we don't mind others file lockers racing with
+ * the mds updates below? */
+ err = llu_extent_unlock(NULL, inode, lsm, lock_mode, &lockh);
+ if (err) {
+ CERROR("llu_extent_unlock failed: %d\n", err);
+ if (!rc)
+ rc = err;
+ }
+ } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
+ struct obd_info oinfo = { { { 0 } } };
struct obdo oa;
- int err2;
- CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
- lli->lli_st_ino, attr->ia_mtime);
+ CDEBUG(D_INODE, "set mtime on OST inode %llu to %lu\n",
+ (long long)st->st_ino, LTIME_S(attr->ia_mtime));
oa.o_id = lsm->lsm_object_id;
- oa.o_mode = S_IFREG;
- oa.o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLMTIME;
- oa.o_mtime = attr->ia_mtime;
- err2 = obd_setattr(&sbi->ll_osc_conn, &oa, lsm, NULL);
- if (err2) {
- CERROR("obd_setattr fails: rc=%d\n", err);
- if (!err)
- err = err2;
- }
+ oa.o_valid = OBD_MD_FLID;
+
+ obdo_from_inode(&oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
+ OBD_MD_FLMTIME | OBD_MD_FLCTIME);
+
+ oinfo.oi_oa = &oa;
+ oinfo.oi_md = lsm;
+
+ rc = obd_setattr_rqset(sbi->ll_osc_exp, &oinfo, NULL);
+ if (rc)
+ CERROR("obd_setattr_async fails: rc=%d\n", rc);
}
- RETURN(err);
+ RETURN(rc);
}
-/* FIXME here we simply act as a thin layer to glue it with
+/* here we simply act as a thin layer to glue it with
* llu_setattr_raw(), which is copy from kernel
*/
static int llu_iop_setattr(struct pnode *pno,
struct intnl_stat *stbuf)
{
struct iattr iattr;
+ int rc;
+ ENTRY;
+ liblustre_wait_event(0);
+
+ LASSERT(!(mask & ~(SETATTR_MTIME | SETATTR_ATIME |
+ SETATTR_UID | SETATTR_GID |
+ SETATTR_LEN | SETATTR_MODE)));
memset(&iattr, 0, sizeof(iattr));
if (mask & SETATTR_MODE) {
}
if (mask & SETATTR_MTIME) {
iattr.ia_mtime = stbuf->st_mtime;
- iattr.ia_valid |= ATTR_MTIME;
+ iattr.ia_valid |= ATTR_MTIME | ATTR_MTIME_SET;
}
if (mask & SETATTR_ATIME) {
iattr.ia_atime = stbuf->st_atime;
- iattr.ia_valid |= ATTR_ATIME;
+ iattr.ia_valid |= ATTR_ATIME | ATTR_ATIME_SET;
}
if (mask & SETATTR_UID) {
iattr.ia_uid = stbuf->st_uid;
iattr.ia_valid |= ATTR_GID;
}
if (mask & SETATTR_LEN) {
- iattr.ia_size = stbuf->st_size; /* FIXME signed expansion problem */
+ iattr.ia_size = stbuf->st_size; /* XXX signed expansion problem */
iattr.ia_valid |= ATTR_SIZE;
}
- iattr.ia_valid |= ATTR_RAW;
- /* FIXME FIXME FIXME FIXME FIXME FIXME FIXME
- * without ATTR_FROM_OPEN, mds_reint_setattr will call
- * mds_fid2locked_dentry() and deadlocked at completion_ast call.
- * Here we workaround it and avoid any locking.
- * FIXME FIXME FIXME FIXME FIXME FIXME FIXME
- */
- iattr.ia_valid |= ATTR_FROM_OPEN;
+ iattr.ia_valid |= ATTR_RAW | ATTR_CTIME;
+ iattr.ia_ctime = CURRENT_TIME;
- return llu_setattr_raw(ino, &iattr);
+ rc = llu_setattr_raw(ino, &iattr);
+ liblustre_wait_event(0);
+ RETURN(rc);
}
+#define EXT2_LINK_MAX 32000
-static int llu_mkdir2(struct inode *dir, const char *name, int len, int mode)
+static int llu_iop_symlink_raw(struct pnode *pno, const char *tgt)
{
+ struct inode *dir = pno->p_base->pb_parent->pb_ino;
+ struct qstr *qstr = &pno->p_base->pb_name;
+ const char *name = qstr->name;
+ int len = qstr->len;
struct ptlrpc_request *request = NULL;
- time_t curtime = CURRENT_TIME;
struct llu_sb_info *sbi = llu_i2sbi(dir);
- struct llu_inode_info *lli = llu_i2info(dir);
struct mdc_op_data op_data;
int err = -EMLINK;
ENTRY;
- CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu\n",
- name, lli->lli_st_ino);
- /* FIXME check this later */
-#if 0
- if (dir->i_nlink >= EXT2_LINK_MAX)
+ liblustre_wait_event(0);
+ if (llu_i2stat(dir)->st_nlink >= EXT2_LINK_MAX)
RETURN(err);
- mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
-#endif
- mode |= S_IFDIR;
+
llu_prepare_mdc_op_data(&op_data, dir, NULL, name, len, 0);
- err = mdc_create(&sbi->ll_mdc_conn, &op_data, NULL, 0, mode,
- current->fsuid, current->fsgid,
- curtime, 0, &request);
+ err = mdc_create(sbi->ll_mdc_exp, &op_data,
+ tgt, strlen(tgt) + 1, S_IFLNK | S_IRWXUGO,
+ current->fsuid, current->fsgid, current->cap_effective,
+ 0, &request);
ptlrpc_req_finished(request);
+ liblustre_wait_event(0);
RETURN(err);
}
-static int llu_iop_mkdir(struct pnode *pno, mode_t mode)
+static int llu_readlink_internal(struct inode *inode,
+ struct ptlrpc_request **request,
+ char **symname)
{
- struct inode *dir = pno->p_base->pb_parent->pb_ino;
- struct qstr *qstr = &pno->p_base->pb_name;
- int rc;
+ struct llu_inode_info *lli = llu_i2info(inode);
+ struct llu_sb_info *sbi = llu_i2sbi(inode);
+ struct ll_fid fid;
+ struct mds_body *body;
+ struct intnl_stat *st = llu_i2stat(inode);
+ int rc, symlen = st->st_size + 1;
+ ENTRY;
- LASSERT(dir);
+ *request = NULL;
- rc = llu_mkdir2(dir, qstr->name, qstr->len, mode);
+ if (lli->lli_symlink_name) {
+ *symname = lli->lli_symlink_name;
+ CDEBUG(D_INODE, "using cached symlink %s\n", *symname);
+ RETURN(0);
+ }
- return rc;
+ ll_inode2fid(&fid, inode);
+ rc = mdc_getattr(sbi->ll_mdc_exp, &fid,
+ OBD_MD_LINKNAME, symlen, request);
+ if (rc) {
+ CERROR("inode %llu: rc = %d\n", (long long)st->st_ino, rc);
+ RETURN(rc);
+ }
+
+ body = lustre_msg_buf((*request)->rq_repmsg, REPLY_REC_OFF,
+ sizeof(*body));
+ LASSERT(body != NULL);
+ LASSERT_REPSWABBED(*request, REPLY_REC_OFF);
+
+ if ((body->valid & OBD_MD_LINKNAME) == 0) {
+ CERROR ("OBD_MD_LINKNAME not set on reply\n");
+ GOTO (failed, rc = -EPROTO);
+ }
+
+ LASSERT(symlen != 0);
+ if (body->eadatasize != symlen) {
+ CERROR("inode %llu: symlink length %d not expected %d\n",
+ (long long)st->st_ino, body->eadatasize - 1, symlen - 1);
+ GOTO(failed, rc = -EPROTO);
+ }
+
+ *symname = lustre_msg_buf((*request)->rq_repmsg, REPLY_REC_OFF + 1,
+ symlen);
+ if (*symname == NULL ||
+ strnlen(*symname, symlen) != symlen - 1) {
+ /* not full/NULL terminated */
+ CERROR("inode %llu: symlink not NULL terminated string"
+ "of length %d\n", (long long)st->st_ino, symlen - 1);
+ GOTO(failed, rc = -EPROTO);
+ }
+
+ OBD_ALLOC(lli->lli_symlink_name, symlen);
+ /* do not return an error if we cannot cache the symlink locally */
+ if (lli->lli_symlink_name)
+ memcpy(lli->lli_symlink_name, *symname, symlen);
+
+ RETURN(0);
+
+ failed:
+ ptlrpc_req_finished (*request);
+ RETURN (-EPROTO);
}
-#ifndef S_IRWXUGO
-#define S_IRWXUGO (S_IRWXU|S_IRWXG|S_IRWXO)
-#endif
+static int llu_iop_readlink(struct pnode *pno, char *data, size_t bufsize)
+{
+ struct inode *inode = pno->p_base->pb_ino;
+ struct ptlrpc_request *request;
+ char *symname;
+ int rc;
+ ENTRY;
+
+ liblustre_wait_event(0);
+ rc = llu_readlink_internal(inode, &request, &symname);
+ if (rc)
+ GOTO(out, rc);
+
+ LASSERT(symname);
+ strncpy(data, symname, bufsize);
+ rc = strlen(symname);
-static int llu_symlink2(struct inode *dir, const char *name, int len,
- const char *tgt)
+ ptlrpc_req_finished(request);
+ out:
+ liblustre_wait_event(0);
+ RETURN(rc);
+}
+
+static int llu_iop_mknod_raw(struct pnode *pno,
+ mode_t mode,
+ dev_t dev)
{
struct ptlrpc_request *request = NULL;
- time_t curtime = CURRENT_TIME;
+ struct inode *dir = pno->p_parent->p_base->pb_ino;
struct llu_sb_info *sbi = llu_i2sbi(dir);
- struct llu_inode_info *lli = llu_i2info(dir);
struct mdc_op_data op_data;
int err = -EMLINK;
ENTRY;
- CDEBUG(D_VFSTRACE, "VFS Op:name=%s,dir=%lu,target=%s\n",
- name, lli->lli_st_ino, tgt);
+ liblustre_wait_event(0);
+ CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu\n",
+ (int)pno->p_base->pb_name.len, pno->p_base->pb_name.name,
+ (long long)llu_i2stat(dir)->st_ino);
-#if 0
- if (dir->i_nlink >= EXT2_LINK_MAX)
+ if (llu_i2stat(dir)->st_nlink >= EXT2_LINK_MAX)
RETURN(err);
-#endif
- llu_prepare_mdc_op_data(&op_data, dir, NULL, name, len, 0);
- err = mdc_create(&sbi->ll_mdc_conn, &op_data,
- tgt, strlen(tgt) + 1, S_IFLNK | S_IRWXUGO,
- current->fsuid, current->fsgid, curtime, 0, &request);
- ptlrpc_req_finished(request);
+
+ switch (mode & S_IFMT) {
+ case 0:
+ case S_IFREG:
+ mode |= S_IFREG; /* for mode = 0 case, fallthrough */
+ case S_IFCHR:
+ case S_IFBLK:
+ case S_IFIFO:
+ case S_IFSOCK:
+ llu_prepare_mdc_op_data(&op_data, dir, NULL,
+ pno->p_base->pb_name.name,
+ pno->p_base->pb_name.len,
+ 0);
+ err = mdc_create(sbi->ll_mdc_exp, &op_data, NULL, 0, mode,
+ current->fsuid, current->fsgid,
+ current->cap_effective, dev, &request);
+ ptlrpc_req_finished(request);
+ break;
+ case S_IFDIR:
+ err = -EPERM;
+ break;
+ default:
+ err = -EINVAL;
+ }
+ liblustre_wait_event(0);
RETURN(err);
}
-static int llu_iop_symlink(struct pnode *pno, const char *data)
+static int llu_iop_link_raw(struct pnode *old, struct pnode *new)
{
- struct inode *dir = pno->p_base->pb_parent->pb_ino;
- struct qstr *qstr = &pno->p_base->pb_name;
+ struct inode *src = old->p_base->pb_ino;
+ struct inode *dir = new->p_parent->p_base->pb_ino;
+ const char *name = new->p_base->pb_name.name;
+ int namelen = new->p_base->pb_name.len;
+ struct ptlrpc_request *request = NULL;
+ struct mdc_op_data op_data;
int rc;
-
+ ENTRY;
+
+ LASSERT(src);
LASSERT(dir);
- rc = llu_symlink2(dir, qstr->name, qstr->len, data);
+ liblustre_wait_event(0);
+ llu_prepare_mdc_op_data(&op_data, src, dir, name, namelen, 0);
+ rc = mdc_link(llu_i2sbi(src)->ll_mdc_exp, &op_data, &request);
+ ptlrpc_req_finished(request);
+ liblustre_wait_event(0);
- return rc;
+ RETURN(rc);
}
-struct filesys_ops llu_filesys_ops =
+/*
+ * libsysio will clear the inode immediately after return
+ */
+static int llu_iop_unlink_raw(struct pnode *pno)
{
- fsop_gone: llu_fsop_gone,
-};
+ struct inode *dir = pno->p_base->pb_parent->pb_ino;
+ struct qstr *qstr = &pno->p_base->pb_name;
+ const char *name = qstr->name;
+ int len = qstr->len;
+ struct inode *target = pno->p_base->pb_ino;
+ struct ptlrpc_request *request = NULL;
+ struct mdc_op_data op_data;
+ int rc;
+ ENTRY;
+ LASSERT(target);
-static struct inode_ops llu_inode_ops = {
- inop_lookup: llu_iop_lookup,
- inop_getattr: llu_iop_getattr,
- inop_setattr: llu_iop_setattr,
- inop_getdirentries: NULL,
- inop_mkdir: llu_iop_mkdir,
- inop_rmdir: NULL,
- inop_symlink: llu_iop_symlink,
- inop_readlink: NULL,
- inop_open: llu_iop_open,
- inop_close: llu_iop_close,
- inop_unlink: NULL,
- inop_ipreadv: llu_iop_ipreadv,
- inop_ipwritev: llu_iop_ipwritev,
- inop_iodone: llu_iop_iodone,
- inop_fcntl: NULL,
- inop_sync: NULL,
- inop_datasync: NULL,
- inop_ioctl: NULL,
- inop_mknod: NULL,
- inop_statvfs: NULL,
- inop_gone: llu_iop_gone,
+ liblustre_wait_event(0);
+ llu_prepare_mdc_op_data(&op_data, dir, NULL, name, len, 0);
+ rc = mdc_unlink(llu_i2sbi(dir)->ll_mdc_exp, &op_data, &request);
+ if (!rc)
+ rc = llu_objects_destroy(request, dir);
+ ptlrpc_req_finished(request);
+ liblustre_wait_event(0);
+
+ RETURN(rc);
+}
+
+static int llu_iop_rename_raw(struct pnode *old, struct pnode *new)
+{
+ struct inode *src = old->p_parent->p_base->pb_ino;
+ struct inode *tgt = new->p_parent->p_base->pb_ino;
+ const char *oldname = old->p_base->pb_name.name;
+ int oldnamelen = old->p_base->pb_name.len;
+ const char *newname = new->p_base->pb_name.name;
+ int newnamelen = new->p_base->pb_name.len;
+ struct ptlrpc_request *request = NULL;
+ struct mdc_op_data op_data;
+ int rc;
+ ENTRY;
+
+ LASSERT(src);
+ LASSERT(tgt);
+
+ liblustre_wait_event(0);
+ llu_prepare_mdc_op_data(&op_data, src, tgt, NULL, 0, 0);
+ rc = mdc_rename(llu_i2sbi(src)->ll_mdc_exp, &op_data,
+ oldname, oldnamelen, newname, newnamelen,
+ &request);
+ if (!rc) {
+ rc = llu_objects_destroy(request, src);
+ }
+
+ ptlrpc_req_finished(request);
+ liblustre_wait_event(0);
+
+ RETURN(rc);
+}
+
+#ifdef _HAVE_STATVFS
+static int llu_statfs_internal(struct llu_sb_info *sbi,
+ struct obd_statfs *osfs, __u64 max_age)
+{
+ struct obd_statfs obd_osfs;
+ int rc;
+ ENTRY;
+
+ rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age);
+ if (rc) {
+ CERROR("mdc_statfs fails: rc = %d\n", rc);
+ RETURN(rc);
+ }
+
+ CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
+ osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
+
+ rc = obd_statfs_rqset(class_exp2obd(sbi->ll_osc_exp),
+ &obd_osfs, max_age);
+ if (rc) {
+ CERROR("obd_statfs fails: rc = %d\n", rc);
+ RETURN(rc);
+ }
+
+ CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
+ obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
+ obd_osfs.os_files);
+
+ osfs->os_blocks = obd_osfs.os_blocks;
+ osfs->os_bfree = obd_osfs.os_bfree;
+ osfs->os_bavail = obd_osfs.os_bavail;
+
+ /* If we don't have as many objects free on the OST as inodes
+ * on the MDS, we reduce the total number of inodes to
+ * compensate, so that the "inodes in use" number is correct.
+ */
+ if (obd_osfs.os_ffree < osfs->os_ffree) {
+ osfs->os_files = (osfs->os_files - osfs->os_ffree) +
+ obd_osfs.os_ffree;
+ osfs->os_ffree = obd_osfs.os_ffree;
+ }
+
+ RETURN(rc);
+}
+
+static int llu_statfs(struct llu_sb_info *sbi, struct statfs *sfs)
+{
+ struct obd_statfs osfs;
+ int rc;
+
+ CDEBUG(D_VFSTRACE, "VFS Op:\n");
+
+ /* For now we will always get up-to-date statfs values, but in the
+ * future we may allow some amount of caching on the client (e.g.
+ * from QOS or lprocfs updates). */
+ rc = llu_statfs_internal(sbi, &osfs, cfs_time_current_64() - HZ);
+ if (rc)
+ return rc;
+
+ statfs_unpack(sfs, &osfs);
+
+ if (sizeof(sfs->f_blocks) == 4) {
+ while (osfs.os_blocks > ~0UL) {
+ sfs->f_bsize <<= 1;
+
+ osfs.os_blocks >>= 1;
+ osfs.os_bfree >>= 1;
+ osfs.os_bavail >>= 1;
+ }
+ }
+
+ sfs->f_blocks = osfs.os_blocks;
+ sfs->f_bfree = osfs.os_bfree;
+ sfs->f_bavail = osfs.os_bavail;
+
+ return 0;
+}
+
+static int llu_iop_statvfs(struct pnode *pno,
+ struct inode *ino,
+ struct intnl_statvfs *buf)
+{
+ struct statfs fs;
+ int rc;
+ ENTRY;
+
+ liblustre_wait_event(0);
+
+#ifndef __CYGWIN__
+ LASSERT(pno->p_base->pb_ino);
+ rc = llu_statfs(llu_i2sbi(pno->p_base->pb_ino), &fs);
+ if (rc)
+ RETURN(rc);
+
+ /* from native driver */
+ buf->f_bsize = fs.f_bsize; /* file system block size */
+ buf->f_frsize = fs.f_bsize; /* file system fundamental block size */
+ buf->f_blocks = fs.f_blocks;
+ buf->f_bfree = fs.f_bfree;
+ buf->f_bavail = fs.f_bavail;
+ buf->f_files = fs.f_files; /* Total number serial numbers */
+ buf->f_ffree = fs.f_ffree; /* Number free serial numbers */
+ buf->f_favail = fs.f_ffree; /* Number free ser num for non-privileged*/
+ buf->f_fsid = fs.f_fsid.__val[1];
+ buf->f_flag = 0; /* No equiv in statfs; maybe use type? */
+ buf->f_namemax = fs.f_namelen;
+#endif
+
+ liblustre_wait_event(0);
+ RETURN(0);
+}
+#endif /* _HAVE_STATVFS */
+
+static int llu_iop_mkdir_raw(struct pnode *pno, mode_t mode)
+{
+ struct inode *dir = pno->p_base->pb_parent->pb_ino;
+ struct qstr *qstr = &pno->p_base->pb_name;
+ const char *name = qstr->name;
+ int len = qstr->len;
+ struct ptlrpc_request *request = NULL;
+ struct intnl_stat *st = llu_i2stat(dir);
+ struct mdc_op_data op_data;
+ int err = -EMLINK;
+ ENTRY;
+
+ liblustre_wait_event(0);
+ CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu/%lu(%p)\n", len, name,
+ (long long)st->st_ino, llu_i2info(dir)->lli_st_generation, dir);
+
+ if (st->st_nlink >= EXT2_LINK_MAX)
+ RETURN(err);
+
+ llu_prepare_mdc_op_data(&op_data, dir, NULL, name, len, 0);
+ err = mdc_create(llu_i2sbi(dir)->ll_mdc_exp, &op_data, NULL, 0, mode,
+ current->fsuid, current->fsgid, current->cap_effective,
+ 0, &request);
+ ptlrpc_req_finished(request);
+ liblustre_wait_event(0);
+ RETURN(err);
+}
+
+static int llu_iop_rmdir_raw(struct pnode *pno)
+{
+ struct inode *dir = pno->p_base->pb_parent->pb_ino;
+ struct qstr *qstr = &pno->p_base->pb_name;
+ const char *name = qstr->name;
+ int len = qstr->len;
+ struct ptlrpc_request *request = NULL;
+ struct mdc_op_data op_data;
+ int rc;
+ ENTRY;
+
+ liblustre_wait_event(0);
+ CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu/%lu(%p)\n", len, name,
+ (long long)llu_i2stat(dir)->st_ino,
+ llu_i2info(dir)->lli_st_generation, dir);
+
+ llu_prepare_mdc_op_data(&op_data, dir, NULL, name, len, S_IFDIR);
+ rc = mdc_unlink(llu_i2sbi(dir)->ll_mdc_exp, &op_data, &request);
+ ptlrpc_req_finished(request);
+
+ liblustre_wait_event(0);
+ RETURN(rc);
+}
+
+#ifdef O_DIRECT
+#define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC|O_DIRECT)
+#else
+#define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC)
+#endif
+#define FCNTL_FLMASK_INVALID (O_NONBLOCK|O_ASYNC)
+
+/* refer to ll_file_flock() for details */
+static int llu_file_flock(struct inode *ino,
+ int cmd,
+ struct file_lock *file_lock)
+{
+ struct llu_inode_info *lli = llu_i2info(ino);
+ struct intnl_stat *st = llu_i2stat(ino);
+ struct ldlm_res_id res_id =
+ { .name = {st->st_ino,
+ lli->lli_st_generation, LDLM_FLOCK} };
+ struct lustre_handle lockh = {0};
+ ldlm_policy_data_t flock;
+ ldlm_mode_t mode = 0;
+ int flags = 0;
+ int rc;
+
+ CDEBUG(D_VFSTRACE, "VFS Op:inode="LPU64" file_lock=%p\n",
+ st->st_ino, file_lock);
+
+ flock.l_flock.pid = file_lock->fl_pid;
+ flock.l_flock.start = file_lock->fl_start;
+ flock.l_flock.end = file_lock->fl_end;
+
+ switch (file_lock->fl_type) {
+ case F_RDLCK:
+ mode = LCK_PR;
+ break;
+ case F_UNLCK:
+ mode = LCK_NL;
+ break;
+ case F_WRLCK:
+ mode = LCK_PW;
+ break;
+ default:
+ CERROR("unknown fcntl lock type: %d\n", file_lock->fl_type);
+ LBUG();
+ }
+
+ switch (cmd) {
+ case F_SETLKW:
+#ifdef F_SETLKW64
+#if F_SETLKW64 != F_SETLKW
+ case F_SETLKW64:
+#endif
+#endif
+ flags = 0;
+ break;
+ case F_SETLK:
+#ifdef F_SETLK64
+#if F_SETLK64 != F_SETLK
+ case F_SETLK64:
+#endif
+#endif
+ flags = LDLM_FL_BLOCK_NOWAIT;
+ break;
+ case F_GETLK:
+#ifdef F_GETLK64
+#if F_GETLK64 != F_GETLK
+ case F_GETLK64:
+#endif
+#endif
+ flags = LDLM_FL_TEST_LOCK;
+ file_lock->fl_type = mode;
+ break;
+ default:
+ CERROR("unknown fcntl cmd: %d\n", cmd);
+ LBUG();
+ }
+
+ CDEBUG(D_DLMTRACE, "inode="LPU64", pid=%u, flags=%#x, mode=%u, "
+ "start="LPU64", end="LPU64"\n", st->st_ino, flock.l_flock.pid,
+ flags, mode, flock.l_flock.start, flock.l_flock.end);
+
+ rc = ldlm_cli_enqueue(llu_i2mdcexp(ino), NULL, res_id,
+ LDLM_FLOCK, &flock, mode, &flags, NULL,
+ ldlm_flock_completion_ast, NULL,
+ file_lock, NULL, 0, NULL, &lockh, 0);
+ RETURN(rc);
+}
+
+static int assign_type(struct file_lock *fl, int type)
+{
+ switch (type) {
+ case F_RDLCK:
+ case F_WRLCK:
+ case F_UNLCK:
+ fl->fl_type = type;
+ return 0;
+ default:
+ return -EINVAL;
+ }
+}
+
+static int flock_to_posix_lock(struct inode *ino,
+ struct file_lock *fl,
+ struct flock *l)
+{
+ switch (l->l_whence) {
+ /* XXX: only SEEK_SET is supported in lustre */
+ case SEEK_SET:
+ fl->fl_start = 0;
+ break;
+ default:
+ return -EINVAL;
+ }
+
+ fl->fl_end = l->l_len - 1;
+ if (l->l_len < 0)
+ return -EINVAL;
+ if (l->l_len == 0)
+ fl->fl_end = OFFSET_MAX;
+
+ fl->fl_pid = getpid();
+ fl->fl_flags = FL_POSIX;
+ fl->fl_notify = NULL;
+ fl->fl_insert = NULL;
+ fl->fl_remove = NULL;
+ /* XXX: these fields can't be filled with suitable values,
+ but I think lustre doesn't use them.
+ */
+ fl->fl_owner = NULL;
+ fl->fl_file = NULL;
+
+ return assign_type(fl, l->l_type);
+}
+
+static int llu_fcntl_getlk(struct inode *ino, struct flock *flock)
+{
+ struct file_lock fl;
+ int error;
+
+ error = EINVAL;
+ if ((flock->l_type != F_RDLCK) && (flock->l_type != F_WRLCK))
+ goto out;
+
+ error = flock_to_posix_lock(ino, &fl, flock);
+ if (error)
+ goto out;
+
+ error = llu_file_flock(ino, F_GETLK, &fl);
+ if (error)
+ goto out;
+
+ flock->l_type = F_UNLCK;
+ if (fl.fl_type != F_UNLCK) {
+ flock->l_pid = fl.fl_pid;
+ flock->l_start = fl.fl_start;
+ flock->l_len = fl.fl_end == OFFSET_MAX ? 0:
+ fl.fl_end - fl.fl_start + 1;
+ flock->l_whence = SEEK_SET;
+ flock->l_type = fl.fl_type;
+ }
+
+out:
+ return error;
+}
+
+static int llu_fcntl_setlk(struct inode *ino, int cmd, struct flock *flock)
+{
+ struct file_lock fl;
+ int flags = llu_i2info(ino)->lli_open_flags + 1;
+ int error;
+
+ error = flock_to_posix_lock(ino, &fl, flock);
+ if (error)
+ goto out;
+ if (cmd == F_SETLKW)
+ fl.fl_flags |= FL_SLEEP;
+
+ error = -EBADF;
+ switch (flock->l_type) {
+ case F_RDLCK:
+ if (!(flags & FMODE_READ))
+ goto out;
+ break;
+ case F_WRLCK:
+ if (!(flags & FMODE_WRITE))
+ goto out;
+ break;
+ case F_UNLCK:
+ break;
+ default:
+ error = -EINVAL;
+ goto out;
+ }
+
+ error = llu_file_flock(ino, cmd, &fl);
+ if (error)
+ goto out;
+
+out:
+ return error;
+}
+
+static int llu_iop_fcntl(struct inode *ino, int cmd, va_list ap, int *rtn)
+{
+ struct llu_inode_info *lli = llu_i2info(ino);
+ long flags;
+ struct flock *flock;
+ long err = 0;
+
+ liblustre_wait_event(0);
+ switch (cmd) {
+ case F_GETFL:
+ *rtn = lli->lli_open_flags;
+ break;
+ case F_SETFL:
+ flags = va_arg(ap, long);
+ flags &= FCNTL_FLMASK;
+ if (flags & FCNTL_FLMASK_INVALID) {
+ CERROR("liblustre don't support O_NONBLOCK, O_ASYNC, "
+ "and O_DIRECT on file descriptor\n");
+ *rtn = -EINVAL;
+ err = EINVAL;
+ break;
+ }
+ lli->lli_open_flags = (int)(flags & FCNTL_FLMASK) |
+ (lli->lli_open_flags & ~FCNTL_FLMASK);
+ *rtn = 0;
+ break;
+ case F_GETLK:
+#ifdef F_GETLK64
+#if F_GETLK64 != F_GETLK
+ case F_GETLK64:
+#endif
+#endif
+ flock = va_arg(ap, struct flock *);
+ err = llu_fcntl_getlk(ino, flock);
+ *rtn = err? -1: 0;
+ break;
+ case F_SETLK:
+#ifdef F_SETLKW64
+#if F_SETLKW64 != F_SETLKW
+ case F_SETLKW64:
+#endif
+#endif
+ case F_SETLKW:
+#ifdef F_SETLK64
+#if F_SETLK64 != F_SETLK
+ case F_SETLK64:
+#endif
+#endif
+ flock = va_arg(ap, struct flock *);
+ err = llu_fcntl_setlk(ino, cmd, flock);
+ *rtn = err? -1: 0;
+ break;
+ default:
+ CERROR("unsupported fcntl cmd %x\n", cmd);
+ *rtn = -ENOSYS;
+ err = ENOSYS;
+ break;
+ }
+
+ liblustre_wait_event(0);
+ return err;
+}
+
+static int llu_get_grouplock(struct inode *inode, unsigned long arg)
+{
+ struct llu_inode_info *lli = llu_i2info(inode);
+ struct ll_file_data *fd = lli->lli_file_data;
+ ldlm_policy_data_t policy = { .l_extent = { .start = 0,
+ .end = OBD_OBJECT_EOF}};
+ struct lustre_handle lockh = { 0 };
+ struct lov_stripe_md *lsm = lli->lli_smd;
+ ldlm_error_t err;
+ int flags = 0;
+ ENTRY;
+
+ if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
+ RETURN(-EINVAL);
+ }
+
+ policy.l_extent.gid = arg;
+ if (lli->lli_open_flags & O_NONBLOCK)
+ flags = LDLM_FL_BLOCK_NOWAIT;
+
+ err = llu_extent_lock(fd, inode, lsm, LCK_GROUP, &policy, &lockh,
+ flags);
+ if (err)
+ RETURN(err);
+
+ fd->fd_flags |= LL_FILE_GROUP_LOCKED|LL_FILE_IGNORE_LOCK;
+ fd->fd_gid = arg;
+ memcpy(&fd->fd_cwlockh, &lockh, sizeof(lockh));
+
+ RETURN(0);
+}
+
+static int llu_put_grouplock(struct inode *inode, unsigned long arg)
+{
+ struct llu_inode_info *lli = llu_i2info(inode);
+ struct ll_file_data *fd = lli->lli_file_data;
+ struct lov_stripe_md *lsm = lli->lli_smd;
+ ldlm_error_t err;
+ ENTRY;
+
+ if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED))
+ RETURN(-EINVAL);
+
+ if (fd->fd_gid != arg)
+ RETURN(-EINVAL);
+
+ fd->fd_flags &= ~(LL_FILE_GROUP_LOCKED|LL_FILE_IGNORE_LOCK);
+
+ err = llu_extent_unlock(fd, inode, lsm, LCK_GROUP, &fd->fd_cwlockh);
+ if (err)
+ RETURN(err);
+
+ fd->fd_gid = 0;
+ memset(&fd->fd_cwlockh, 0, sizeof(fd->fd_cwlockh));
+
+ RETURN(0);
+}
+
+static int llu_lov_dir_setstripe(struct inode *ino, unsigned long arg)
+{
+ struct llu_sb_info *sbi = llu_i2sbi(ino);
+ struct ptlrpc_request *request = NULL;
+ struct mdc_op_data op_data;
+ struct iattr attr = { 0 };
+ struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
+ int rc = 0;
+
+ llu_prepare_mdc_op_data(&op_data, ino, NULL, NULL, 0, 0);
+
+ LASSERT(sizeof(lum) == sizeof(*lump));
+ LASSERT(sizeof(lum.lmm_objects[0]) ==
+ sizeof(lump->lmm_objects[0]));
+ rc = copy_from_user(&lum, lump, sizeof(lum));
+ if (rc)
+ return(-EFAULT);
+
+ if (lum.lmm_magic != LOV_USER_MAGIC)
+ RETURN(-EINVAL);
+
+ if (lum.lmm_magic != cpu_to_le32(LOV_USER_MAGIC))
+ lustre_swab_lov_user_md(&lum);
+
+ /* swabbing is done in lov_setstripe() on server side */
+ rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
+ &attr, &lum, sizeof(lum), NULL, 0, &request);
+ if (rc) {
+ ptlrpc_req_finished(request);
+ if (rc != -EPERM && rc != -EACCES)
+ CERROR("mdc_setattr fails: rc = %d\n", rc);
+ return rc;
+ }
+ ptlrpc_req_finished(request);
+
+ return rc;
+}
+
+static int llu_lov_setstripe_ea_info(struct inode *ino, int flags,
+ struct lov_user_md *lum, int lum_size)
+{
+ struct llu_sb_info *sbi = llu_i2sbi(ino);
+ struct obd_export *exp = llu_i2obdexp(ino);
+ struct llu_inode_info *lli = llu_i2info(ino);
+ struct llu_inode_info *lli2 = NULL;
+ struct lov_stripe_md *lsm;
+ struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
+ struct ptlrpc_request *req = NULL;
+ struct lustre_md md;
+ struct mdc_op_data data;
+ struct lustre_handle lockh;
+ int rc = 0;
+ ENTRY;
+
+ lsm = lli->lli_smd;
+ if (lsm) {
+ CDEBUG(D_IOCTL, "stripe already exists for ino "LPU64"\n",
+ lli->lli_fid.id);
+ return -EEXIST;
+ }
+
+ OBD_ALLOC(lli2, sizeof(struct llu_inode_info));
+ if (!lli2)
+ return -ENOMEM;
+
+ memcpy(lli2, lli, sizeof(struct llu_inode_info));
+ lli2->lli_open_count = 0;
+ lli2->lli_it = NULL;
+ lli2->lli_file_data = NULL;
+ lli2->lli_smd = NULL;
+ lli2->lli_symlink_name = NULL;
+ ino->i_private = lli2;
+
+ llu_prepare_mdc_op_data(&data, NULL, ino, NULL, 0, O_RDWR);
+
+ rc = mdc_enqueue(sbi->ll_mdc_exp, LDLM_IBITS, &oit, LCK_CR, &data,
+ &lockh, lum, lum_size, ldlm_completion_ast,
+ llu_mdc_blocking_ast, NULL, LDLM_FL_INTENT_ONLY);
+ if (rc)
+ GOTO(out, rc);
+
+ req = oit.d.lustre.it_data;
+ rc = it_open_error(DISP_IT_EXECD, &oit);
+ if (rc) {
+ req->rq_replay = 0;
+ GOTO(out, rc);
+ }
+
+ rc = it_open_error(DISP_OPEN_OPEN, &oit);
+ if (rc) {
+ req->rq_replay = 0;
+ GOTO(out, rc);
+ }
+
+ rc = mdc_req2lustre_md(req, DLM_REPLY_REC_OFF, exp, &md);
+ if (rc)
+ GOTO(out, rc);
+
+ llu_update_inode(ino, md.body, md.lsm);
+ lli->lli_smd = lli2->lli_smd;
+ lli2->lli_smd = NULL;
+
+ llu_local_open(lli2, &oit);
+
+ /* release intent */
+ if (lustre_handle_is_used(&lockh))
+ ldlm_lock_decref(&lockh, LCK_CR);
+
+ ptlrpc_req_finished(req);
+ req = NULL;
+
+ rc = llu_file_release(ino);
+ out:
+ ino->i_private = lli;
+ if (lli2)
+ OBD_FREE(lli2, sizeof(struct llu_inode_info));
+ if (req != NULL)
+ ptlrpc_req_finished(req);
+ RETURN(rc);
+}
+
+static int llu_lov_file_setstripe(struct inode *ino, unsigned long arg)
+{
+ struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
+ int rc;
+ int flags = FMODE_WRITE;
+ ENTRY;
+
+ LASSERT(sizeof(lum) == sizeof(*lump));
+ LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lump->lmm_objects[0]));
+ rc = copy_from_user(&lum, lump, sizeof(lum));
+ if (rc)
+ RETURN(-EFAULT);
+
+ rc = llu_lov_setstripe_ea_info(ino, flags, &lum, sizeof(lum));
+ RETURN(rc);
+}
+
+static int llu_lov_setstripe(struct inode *ino, unsigned long arg)
+{
+ struct intnl_stat *st = llu_i2stat(ino);
+ if (S_ISREG(st->st_mode))
+ return llu_lov_file_setstripe(ino, arg);
+ if (S_ISDIR(st->st_mode))
+ return llu_lov_dir_setstripe(ino, arg);
+
+ return -EINVAL;
+}
+
+static int llu_lov_getstripe(struct inode *ino, unsigned long arg)
+{
+ struct lov_stripe_md *lsm = llu_i2info(ino)->lli_smd;
+
+ if (!lsm)
+ RETURN(-ENODATA);
+
+ return obd_iocontrol(LL_IOC_LOV_GETSTRIPE, llu_i2obdexp(ino), 0, lsm,
+ (void *)arg);
+}
+
+static int llu_iop_ioctl(struct inode *ino, unsigned long int request,
+ va_list ap)
+{
+ unsigned long arg;
+ int rc;
+
+ liblustre_wait_event(0);
+
+ switch (request) {
+ case LL_IOC_GROUP_LOCK:
+ arg = va_arg(ap, unsigned long);
+ rc = llu_get_grouplock(ino, arg);
+ break;
+ case LL_IOC_GROUP_UNLOCK:
+ arg = va_arg(ap, unsigned long);
+ rc = llu_put_grouplock(ino, arg);
+ break;
+ case LL_IOC_LOV_SETSTRIPE:
+ arg = va_arg(ap, unsigned long);
+ rc = llu_lov_setstripe(ino, arg);
+ break;
+ case LL_IOC_LOV_GETSTRIPE:
+ arg = va_arg(ap, unsigned long);
+ rc = llu_lov_getstripe(ino, arg);
+ break;
+ default:
+ CERROR("did not support ioctl cmd %lx\n", request);
+ rc = -ENOSYS;
+ break;
+ }
+
+ liblustre_wait_event(0);
+ return rc;
+}
+
+/*
+ * we already do syncronous read/write
+ */
+static int llu_iop_sync(struct inode *inode)
+{
+ liblustre_wait_event(0);
+ return 0;
+}
+
+static int llu_iop_datasync(struct inode *inode)
+{
+ liblustre_wait_event(0);
+ return 0;
+}
+
+struct filesys_ops llu_filesys_ops =
+{
+ fsop_gone: llu_fsop_gone,
};
+struct inode *llu_iget(struct filesys *fs, struct lustre_md *md)
+{
+ struct inode *inode;
+ struct ll_fid fid;
+ struct file_identifier fileid = {&fid, sizeof(fid)};
+
+ if ((md->body->valid &
+ (OBD_MD_FLGENER | OBD_MD_FLID | OBD_MD_FLTYPE)) !=
+ (OBD_MD_FLGENER | OBD_MD_FLID | OBD_MD_FLTYPE)) {
+ CERROR("bad md body valid mask "LPX64"\n", md->body->valid);
+ LBUG();
+ return ERR_PTR(-EPERM);
+ }
+
+ /* try to find existing inode */
+ fid.id = md->body->ino;
+ fid.generation = md->body->generation;
+ fid.f_type = md->body->mode & S_IFMT;
+
+ inode = _sysio_i_find(fs, &fileid);
+ if (inode) {
+ struct llu_inode_info *lli = llu_i2info(inode);
+
+ if (inode->i_zombie ||
+ lli->lli_st_generation != md->body->generation) {
+ I_RELE(inode);
+ }
+ else {
+ llu_update_inode(inode, md->body, md->lsm);
+ return inode;
+ }
+ }
+
+ inode = llu_new_inode(fs, &fid);
+ if (inode)
+ llu_update_inode(inode, md->body, md->lsm);
+
+ return inode;
+}
+
+extern struct list_head lustre_profile_list;
static int
llu_fsswop_mount(const char *source,
struct filesys *fs;
struct inode *root;
struct pnode_base *rootpb;
- static struct qstr noname = { NULL, 0, 0 };
+ struct obd_device *obd;
struct ll_fid rootfid;
-
struct llu_sb_info *sbi;
- struct ptlrpc_connection *mdc_conn;
+ struct obd_statfs osfs;
+ static struct qstr noname = { NULL, 0, 0 };
struct ptlrpc_request *request = NULL;
- struct mds_body *root_body;
- struct obd_uuid param_uuid;
+ struct lustre_handle mdc_conn = {0, };
+ struct lustre_handle osc_conn = {0, };
+ struct lustre_md md;
class_uuid_t uuid;
- struct obd_device *obd;
- char *osc=mount_option.osc_uuid;
- char *mdc=mount_option.mdc_uuid;
- int err = -EINVAL;
+ struct config_llog_instance cfg;
+ char ll_instance[sizeof(sbi) * 2 + 1];
+ struct lustre_profile *lprof;
+ char *zconf_mgsnid, *zconf_profile;
+ char *osc = NULL, *mdc = NULL;
+ int async = 1, err = -EINVAL;
+ struct obd_connect_data ocd = {0,};
ENTRY;
+ if (ll_parse_mount_target(source,
+ &zconf_mgsnid,
+ &zconf_profile)) {
+ CERROR("mal-formed target %s\n", source);
+ RETURN(err);
+ }
+ if (!zconf_mgsnid || !zconf_profile) {
+ printf("Liblustre: invalid target %s\n", source);
+ RETURN(err);
+ }
+ /* allocate & initialize sbi */
OBD_ALLOC(sbi, sizeof(*sbi));
if (!sbi)
RETURN(-ENOMEM);
INIT_LIST_HEAD(&sbi->ll_conn_chain);
- generate_random_uuid(uuid);
+ ll_generate_random_uuid(uuid);
class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
+ /* generate a string unique to this super, let's try
+ the address of the super itself.*/
+ sprintf(ll_instance, "%p", sbi);
+
+ /* retrive & parse config log */
+ cfg.cfg_instance = ll_instance;
+ cfg.cfg_uuid = sbi->ll_sb_uuid;
+ cfg.cfg_last_idx = 0;
+ err = liblustre_process_log(&cfg, zconf_mgsnid, zconf_profile, 1);
+ if (err < 0) {
+ CERROR("Unable to process log: %s\n", zconf_profile);
+ GOTO(out_free, err);
+ }
+
+ lprof = class_get_profile(zconf_profile);
+ if (lprof == NULL) {
+ CERROR("No profile found: %s\n", zconf_profile);
+ GOTO(out_free, err = -EINVAL);
+ }
+ OBD_ALLOC(osc, strlen(lprof->lp_osc) + strlen(ll_instance) + 2);
+ sprintf(osc, "%s-%s", lprof->lp_osc, ll_instance);
+
+ OBD_ALLOC(mdc, strlen(lprof->lp_mdc) + strlen(ll_instance) + 2);
+ sprintf(mdc, "%s-%s", lprof->lp_mdc, ll_instance);
+
+ if (!osc) {
+ CERROR("no osc\n");
+ GOTO(out_free, err = -EINVAL);
+ }
+ if (!mdc) {
+ CERROR("no mdc\n");
+ GOTO(out_free, err = -EINVAL);
+ }
+
fs = _sysio_fs_new(&llu_filesys_ops, flags, sbi);
if (!fs) {
err = -ENOMEM;
obd = class_name2obd(mdc);
if (!obd) {
CERROR("MDC %s: not setup or attached\n", mdc);
- err = -EINVAL;
- goto out_free;
+ GOTO(out_free, err = -EINVAL);
}
+ obd_set_info_async(obd->obd_self_export, strlen("async"), "async",
+ sizeof(async), &async, NULL);
+
+ ocd.ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_VERSION;
+ ocd.ocd_ibits_known = MDS_INODELOCK_FULL;
+ ocd.ocd_version = LUSTRE_VERSION_CODE;
/* setup mdc */
- /* FIXME need recover stuff */
- err = obd_connect(&sbi->ll_mdc_conn, obd, &sbi->ll_sb_uuid);
+ err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid, &ocd);
if (err) {
CERROR("cannot connect to %s: rc = %d\n", mdc, err);
- goto out_free;
+ GOTO(out_free, err);
}
+ sbi->ll_mdc_exp = class_conn2export(&mdc_conn);
- mdc_conn = sbi2mdc(sbi)->cl_import->imp_connection;
+ err = obd_statfs(obd, &osfs, 100000000);
+ if (err)
+ GOTO(out_mdc, err);
+
+ /*
+ * FIXME fill fs stat data into sbi here!!! FIXME
+ */
/* setup osc */
obd = class_name2obd(osc);
if (!obd) {
CERROR("OSC %s: not setup or attached\n", osc);
- err = -EINVAL;
- goto out_mdc;
+ GOTO(out_mdc, err = -EINVAL);
}
+ obd_set_info_async(obd->obd_self_export, strlen("async"), "async",
+ sizeof(async), &async, NULL);
+
+ obd->obd_upcall.onu_owner = &sbi->ll_lco;
+ obd->obd_upcall.onu_upcall = ll_ocd_update;
- err = obd_connect(&sbi->ll_osc_conn, obd, &sbi->ll_sb_uuid);
+ ocd.ocd_connect_flags = OBD_CONNECT_SRVLOCK | OBD_CONNECT_REQPORTAL |
+ OBD_CONNECT_VERSION | OBD_CONNECT_TRUNCLOCK;
+ ocd.ocd_version = LUSTRE_VERSION_CODE;
+ err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid, &ocd);
if (err) {
CERROR("cannot connect to %s: rc = %d\n", osc, err);
- goto out_mdc;
+ GOTO(out_mdc, err);
}
+ sbi->ll_osc_exp = class_conn2export(&osc_conn);
+ sbi->ll_lco.lco_flags = ocd.ocd_connect_flags;
- err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
+ mdc_init_ea_size(sbi->ll_mdc_exp, sbi->ll_osc_exp);
+
+ err = mdc_getstatus(sbi->ll_mdc_exp, &rootfid);
if (err) {
CERROR("cannot mds_connect: rc = %d\n", err);
- goto out_osc;
+ GOTO(out_osc, err);
}
CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
sbi->ll_rootino = rootfid.id;
-/* XXX do we need this??
- memset(&osfs, 0, sizeof(osfs));
- rc = obd_statfs(&sbi->ll_mdc_conn, &osfs);
-*/
/* fetch attr of root inode */
- err = mdc_getattr(&sbi->ll_mdc_conn, &rootfid,
- OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
+ err = mdc_getattr(sbi->ll_mdc_exp, &rootfid,
+ OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS, 0, &request);
if (err) {
CERROR("mdc_getattr failed for root: rc = %d\n", err);
- goto out_request;
+ GOTO(out_osc, err);
+ }
+
+ err = mdc_req2lustre_md(request, REPLY_REC_OFF, sbi->ll_osc_exp, &md);
+ if (err) {
+ CERROR("failed to understand root inode md: rc = %d\n",err);
+ GOTO(out_request, err);
}
- root_body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*root_body));
LASSERT(sbi->ll_rootino != 0);
- root = llu_new_inode(fs, root_body->ino, root_body->mode);
- if (!root) {
- err = -ENOMEM;
- goto out_request;
+ root = llu_iget(fs, &md);
+ if (!root || IS_ERR(root)) {
+ CERROR("fail to generate root inode\n");
+ GOTO(out_request, err = -EBADF);
}
- llu_update_inode(root, root_body, NULL);
-
- /*
- * Generate base path-node for root.
- */
- rootpb = _sysio_pb_new(&noname, NULL, root);
- if (!rootpb) {
- err = -ENOMEM;
- goto out_inode;
- }
+ /*
+ * Generate base path-node for root.
+ */
+ rootpb = _sysio_pb_new(&noname, NULL, root);
+ if (!rootpb) {
+ err = -ENOMEM;
+ goto out_inode;
+ }
- err = _sysio_do_mount(fs, rootpb, flags, NULL, mntp);
- if (err) {
+ err = _sysio_do_mount(fs, rootpb, flags, tocover, mntp);
+ if (err) {
_sysio_pb_gone(rootpb);
- goto out_inode;
+ goto out_inode;
}
ptlrpc_req_finished(request);
- request = NULL;
- printf("************************************************\n");
- printf("* Mount successfully!!!!!!! *\n");
- printf("************************************************\n");
+ CDEBUG(D_SUPER, "LibLustre: %s mounted successfully!\n", source);
+ liblustre_wait_idle();
return 0;
out_request:
ptlrpc_req_finished(request);
out_osc:
- obd_disconnect(&sbi->ll_osc_conn);
+ obd_disconnect(sbi->ll_osc_exp);
out_mdc:
- obd_disconnect(&sbi->ll_mdc_conn);
+ obd_disconnect(sbi->ll_mdc_exp);
out_free:
+ if (osc)
+ OBD_FREE(osc, strlen(osc) + 1);
+ if (mdc)
+ OBD_FREE(mdc, strlen(mdc) + 1);
OBD_FREE(sbi, sizeof(*sbi));
+
+ liblustre_wait_idle();
return err;
}
llu_fsswop_mount
};
+static struct inode_ops llu_inode_ops = {
+ inop_lookup: llu_iop_lookup,
+ inop_getattr: llu_iop_getattr,
+ inop_setattr: llu_iop_setattr,
+ inop_filldirentries: llu_iop_filldirentries,
+ inop_mkdir: llu_iop_mkdir_raw,
+ inop_rmdir: llu_iop_rmdir_raw,
+ inop_symlink: llu_iop_symlink_raw,
+ inop_readlink: llu_iop_readlink,
+ inop_open: llu_iop_open,
+ inop_close: llu_iop_close,
+ inop_link: llu_iop_link_raw,
+ inop_unlink: llu_iop_unlink_raw,
+ inop_rename: llu_iop_rename_raw,
+ inop_pos: llu_iop_pos,
+ inop_read: llu_iop_read,
+ inop_write: llu_iop_write,
+ inop_iodone: llu_iop_iodone,
+ inop_fcntl: llu_iop_fcntl,
+ inop_sync: llu_iop_sync,
+ inop_datasync: llu_iop_datasync,
+ inop_ioctl: llu_iop_ioctl,
+ inop_mknod: llu_iop_mknod_raw,
+#ifdef _HAVE_STATVFS
+ inop_statvfs: llu_iop_statvfs,
+#endif
+ inop_gone: llu_iop_gone,
+};