Whamcloud - gitweb
LU-5260 llite: removing ll_cl_context when error occurs
[fs/lustre-release.git] / lustre / llite / file.c
index 9debc22..33c8e74 100644 (file)
@@ -45,6 +45,7 @@
 #include <lustre_lite.h>
 #include <linux/pagemap.h>
 #include <linux/file.h>
+#include <linux/sched.h>
 #include "llite_internal.h"
 #include <lustre/ll_fiemap.h>
 #include <lustre_ioctl.h>
@@ -207,7 +208,7 @@ static int ll_close_inode_openhandle(struct obd_export *md_exp,
        if (rc == 0 && op_data->op_bias & MDS_HSM_RELEASE) {
                struct mdt_body *body;
                body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
-               if (!(body->valid & OBD_MD_FLRELEASED))
+               if (!(body->mbo_valid & OBD_MD_FLRELEASED))
                        rc = -EBUSY;
        }
 
@@ -408,77 +409,69 @@ int ll_file_release(struct inode *inode, struct file *file)
         RETURN(rc);
 }
 
-static int ll_intent_file_open(struct file *file, void *lmm,
-                               int lmmsize, struct lookup_intent *itp)
+static int ll_intent_file_open(struct file *file, void *lmm, int lmmsize,
+                               struct lookup_intent *itp)
 {
-        struct ll_sb_info *sbi = ll_i2sbi(file->f_dentry->d_inode);
-        struct dentry *parent = file->f_dentry->d_parent;
-        const char *name = file->f_dentry->d_name.name;
-        const int len = file->f_dentry->d_name.len;
-        struct md_op_data *op_data;
-        struct ptlrpc_request *req;
-        __u32 opc = LUSTRE_OPC_ANY;
-        int rc;
-        ENTRY;
+       struct dentry *de = file->f_dentry;
+       struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
+       struct dentry *parent = de->d_parent;
+       const char *name = NULL;
+       int len = 0;
+       struct md_op_data *op_data;
+       struct ptlrpc_request *req = NULL;
+       int rc;
+       ENTRY;
 
-        if (!parent)
-                RETURN(-ENOENT);
-
-        /* Usually we come here only for NFSD, and we want open lock.
-           But we can also get here with pre 2.6.15 patchless kernels, and in
-           that case that lock is also ok */
-        /* We can also get here if there was cached open handle in revalidate_it
-         * but it disappeared while we were getting from there to ll_file_open.
-         * But this means this file was closed and immediatelly opened which
-         * makes a good candidate for using OPEN lock */
-        /* If lmmsize & lmm are not 0, we are just setting stripe info
-         * parameters. No need for the open lock */
-        if (lmm == NULL && lmmsize == 0) {
-                itp->it_flags |= MDS_OPEN_LOCK;
-                if (itp->it_flags & FMODE_WRITE)
-                        opc = LUSTRE_OPC_CREATE;
-        }
+       LASSERT(parent != NULL);
+       LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
 
-        op_data  = ll_prep_md_op_data(NULL, parent->d_inode,
-                                      file->f_dentry->d_inode, name, len,
-                                      O_RDWR, opc, NULL);
-        if (IS_ERR(op_data))
-                RETURN(PTR_ERR(op_data));
+       /* if server supports open-by-fid, or file name is invalid, don't pack
+        * name in open request */
+       if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID) &&
+           lu_name_is_valid_2(de->d_name.name, de->d_name.len)) {
+               name = de->d_name.name;
+               len = de->d_name.len;
+       }
 
-       itp->it_flags |= MDS_OPEN_BY_FID;
-        rc = md_intent_lock(sbi->ll_md_exp, op_data, lmm, lmmsize, itp,
-                            0 /*unused */, &req, ll_md_blocking_ast, 0);
-        ll_finish_md_op_data(op_data);
-        if (rc == -ESTALE) {
-                /* reason for keep own exit path - don`t flood log
-                * with messages with -ESTALE errors.
-                */
-                if (!it_disposition(itp, DISP_OPEN_OPEN) ||
-                     it_open_error(DISP_OPEN_OPEN, itp))
-                        GOTO(out, rc);
-                ll_release_openhandle(file->f_dentry, itp);
-                GOTO(out, rc);
-        }
+       op_data = ll_prep_md_op_data(NULL, parent->d_inode, de->d_inode,
+                                    name, len, 0, LUSTRE_OPC_ANY, NULL);
+       if (IS_ERR(op_data))
+               RETURN(PTR_ERR(op_data));
+       op_data->op_data = lmm;
+       op_data->op_data_size = lmmsize;
 
-        if (it_disposition(itp, DISP_LOOKUP_NEG))
-                GOTO(out, rc = -ENOENT);
+       rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
+                           &ll_md_blocking_ast, 0);
+       ll_finish_md_op_data(op_data);
+       if (rc == -ESTALE) {
+               /* reason for keep own exit path - don`t flood log
+                * with messages with -ESTALE errors.
+                */
+               if (!it_disposition(itp, DISP_OPEN_OPEN) ||
+                    it_open_error(DISP_OPEN_OPEN, itp))
+                       GOTO(out, rc);
+               ll_release_openhandle(de, itp);
+               GOTO(out, rc);
+       }
 
-        if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
-                rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
-                CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
-                GOTO(out, rc);
-        }
+       if (it_disposition(itp, DISP_LOOKUP_NEG))
+               GOTO(out, rc = -ENOENT);
 
-        rc = ll_prep_inode(&file->f_dentry->d_inode, req, NULL, itp);
-        if (!rc && itp->d.lustre.it_lock_mode)
-                ll_set_lock_data(sbi->ll_md_exp, file->f_dentry->d_inode,
-                                 itp, NULL);
+       if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
+               rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
+               CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
+               GOTO(out, rc);
+       }
+
+       rc = ll_prep_inode(&de->d_inode, req, NULL, itp);
+       if (!rc && itp->d.lustre.it_lock_mode)
+               ll_set_lock_data(sbi->ll_md_exp, de->d_inode, itp, NULL);
 
 out:
        ptlrpc_req_finished(req);
-        ll_intent_drop_lock(itp);
+       ll_intent_drop_lock(itp);
 
-        RETURN(rc);
+       RETURN(rc);
 }
 
 /**
@@ -502,8 +495,8 @@ static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
        struct mdt_body *body;
 
        body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
-       och->och_fh = body->handle;
-       och->och_fid = body->fid1;
+       och->och_fh = body->mbo_handle;
+       och->och_fid = body->mbo_fid1;
        och->och_lease_handle.cookie = it->d.lustre.it_lock_handle;
        och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
        och->och_flags = it->it_flags;
@@ -532,13 +525,17 @@ static int ll_local_open(struct file *file, struct lookup_intent *it,
                        RETURN(rc);
 
                body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
-               ll_ioepoch_open(lli, body->ioepoch);
+               ll_ioepoch_open(lli, body->mbo_ioepoch);
        }
 
        LUSTRE_FPRIVATE(file) = fd;
        ll_readahead_init(inode, &fd->fd_ras);
        fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
 
+       /* ll_cl_context initialize */
+       rwlock_init(&fd->fd_lock);
+       INIT_LIST_HEAD(&fd->fd_lccs);
+
        RETURN(0);
 }
 
@@ -665,9 +662,19 @@ restart:
                            would attempt to grab och_mutex as well, that would
                            result in a deadlock */
                        mutex_unlock(&lli->lli_och_mutex);
-                        it->it_create_mode |= M_CHECK_STALE;
+                       /*
+                        * Normally called under two situations:
+                        * 1. NFS export.
+                        * 2. revalidate with IT_OPEN (revalidate doesn't
+                        *    execute this intent any more).
+                        *
+                        * Always fetch MDS_OPEN_LOCK if this is not setstripe.
+                        *
+                        * Always specify MDS_OPEN_BY_FID because we don't want
+                        * to get file with different fid.
+                        */
+                       it->it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID;
                         rc = ll_intent_file_open(file, NULL, 0, it);
-                        it->it_create_mode &= ~M_CHECK_STALE;
                         if (rc)
                                 GOTO(out_openerr, rc);
 
@@ -775,7 +782,7 @@ ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
        struct lookup_intent it = { .it_op = IT_OPEN };
        struct ll_sb_info *sbi = ll_i2sbi(inode);
        struct md_op_data *op_data;
-       struct ptlrpc_request *req;
+       struct ptlrpc_request *req = NULL;
        struct lustre_handle old_handle = { 0 };
        struct obd_client_handle *och = NULL;
        int rc;
@@ -841,15 +848,15 @@ ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
 
        it.it_flags = fmode | open_flags;
        it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
-       rc = md_intent_lock(sbi->ll_md_exp, op_data, NULL, 0, &it, 0, &req,
-                               ll_md_blocking_lease_ast,
+       rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
+                           &ll_md_blocking_lease_ast,
        /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
         * it can be cancelled which may mislead applications that the lease is
         * broken;
         * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
         * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
         * doesn't deal with openhandle, so normal openhandle will be leaked. */
-                               LDLM_FL_NO_LRU | LDLM_FL_EXCL);
+                           LDLM_FL_NO_LRU | LDLM_FL_EXCL);
        ll_finish_md_op_data(op_data);
        ptlrpc_req_finished(req);
        if (rc < 0)
@@ -1156,8 +1163,8 @@ restart:
                 struct ccc_io *cio = ccc_env_io(env);
                 int write_mutex_locked = 0;
 
-                cio->cui_fd  = LUSTRE_FPRIVATE(file);
-                vio->cui_io_subtype = args->via_io_subtype;
+               cio->cui_fd  = LUSTRE_FPRIVATE(file);
+               vio->cui_io_subtype = args->via_io_subtype;
 
                 switch (vio->cui_io_subtype) {
                 case IO_NORMAL:
@@ -1182,7 +1189,11 @@ restart:
                         CERROR("Unknow IO type - %u\n", vio->cui_io_subtype);
                         LBUG();
                 }
+
+               ll_cl_add(file, env, io);
                 result = cl_io_loop(env, io);
+               ll_cl_remove(file, env);
+
                if (args->via_io_subtype == IO_NORMAL)
                        up_read(&lli->lli_trunc_sem);
                if (write_mutex_locked)
@@ -1513,6 +1524,7 @@ int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file,
        }
 
        ll_inode_size_lock(inode);
+       oit.it_flags |= MDS_OPEN_BY_FID;
        rc = ll_intent_file_open(file, lum, lum_size, &oit);
        if (rc)
                GOTO(out_unlock, rc);
@@ -1567,9 +1579,9 @@ int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
         LASSERT(body != NULL); /* checked by mdc_getattr_name */
 
-        lmmsize = body->eadatasize;
+       lmmsize = body->mbo_eadatasize;
 
-        if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
+       if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
                         lmmsize == 0) {
                 GOTO(out, rc = -ENODATA);
         }
@@ -1598,13 +1610,14 @@ int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
                  * avoid swab not existent lsm objects */
                 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) {
                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
-                        if (S_ISREG(body->mode))
-                                lustre_swab_lov_user_md_objects(
-                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
-                                 stripe_count);
-                } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
-                        lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
-                        if (S_ISREG(body->mode))
+                       if (S_ISREG(body->mbo_mode))
+                               lustre_swab_lov_user_md_objects(
+                                   ((struct lov_user_md_v1 *)lmm)->lmm_objects,
+                                   stripe_count);
+               } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
+                       lustre_swab_lov_user_md_v3(
+                               (struct lov_user_md_v3 *)lmm);
+                       if (S_ISREG(body->mbo_mode))
                                 lustre_swab_lov_user_md_objects(
                                  ((struct lov_user_md_v3 *)lmm)->lmm_objects,
                                  stripe_count);
@@ -2990,8 +3003,8 @@ ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
               flock.l_flock.pid, flags, einfo.ei_mode,
               flock.l_flock.start, flock.l_flock.end);
 
-        rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL,
-                        op_data, &lockh, &flock, 0, NULL /* req */, flags);
+       rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, NULL, op_data, &lockh,
+                       flags);
 
        /* Restore the file lock type if not TEST lock. */
        if (!(flags & LDLM_FL_TEST_LOCK))
@@ -3007,8 +3020,8 @@ ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
 
        if (rc2 && file_lock->fl_type != F_UNLCK) {
                einfo.ei_mode = LCK_NL;
-               md_enqueue(sbi->ll_md_exp, &einfo, NULL,
-                       op_data, &lockh, &flock, 0, NULL /* req */, flags);
+               md_enqueue(sbi->ll_md_exp, &einfo, &flock, NULL, op_data,
+                          &lockh, flags);
                rc = rc2;
        }
 
@@ -3041,7 +3054,7 @@ int ll_get_fid_by_name(struct inode *parent, const char *name,
        if (body == NULL)
                GOTO(out_req, rc = -EFAULT);
        if (fid != NULL)
-               *fid = body->fid1;
+               *fid = body->mbo_fid1;
 out_req:
        ptlrpc_req_finished(req);
        RETURN(rc);
@@ -3252,14 +3265,9 @@ static int __ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
                 if (IS_ERR(op_data))
                         RETURN(PTR_ERR(op_data));
 
-                oit.it_create_mode |= M_CHECK_STALE;
-                rc = md_intent_lock(exp, op_data, NULL, 0,
-                                    /* we are not interested in name
-                                       based lookup */
-                                    &oit, 0, &req,
-                                    ll_md_blocking_ast, 0);
+               rc = md_intent_lock(exp, op_data, &oit, &req,
+                                   &ll_md_blocking_ast, 0);
                 ll_finish_md_op_data(op_data);
-                oit.it_create_mode &= ~M_CHECK_STALE;
                 if (rc < 0) {
                         rc = ll_inode_revalidate_fini(inode, rc);
                         GOTO (out, rc);
@@ -3507,8 +3515,14 @@ int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
 # endif
 #endif
 {
-        int rc = 0;
-        ENTRY;
+       int rc = 0;
+       struct ll_sb_info *sbi;
+       struct root_squash_info *squash;
+       struct cred *cred = NULL;
+       const struct cred *old_cred = NULL;
+       cfs_cap_t cap;
+       bool squash_id = false;
+       ENTRY;
 
 #ifdef MAY_NOT_BLOCK
        if (mask & MAY_NOT_BLOCK)
@@ -3531,11 +3545,46 @@ int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
        CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
               PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
 
-       if (ll_i2sbi(inode)->ll_flags & LL_SBI_RMT_CLIENT)
-               return lustre_check_remote_perm(inode, mask);
+       /* squash fsuid/fsgid if needed */
+       sbi = ll_i2sbi(inode);
+       squash = &sbi->ll_squash;
+       if (unlikely(squash->rsi_uid != 0 &&
+                    uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
+                    !(sbi->ll_flags & LL_SBI_NOROOTSQUASH))) {
+                       squash_id = true;
+       }
+       if (squash_id) {
+               CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
+                      __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
+                      squash->rsi_uid, squash->rsi_gid);
+
+               /* update current process's credentials
+                * and FS capability */
+               cred = prepare_creds();
+               if (cred == NULL)
+                       RETURN(-ENOMEM);
 
-       ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1);
-       rc = ll_generic_permission(inode, mask, flags, ll_check_acl);
+               cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
+               cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
+               for (cap = 0; cap < sizeof(cfs_cap_t) * 8; cap++) {
+                       if ((1 << cap) & CFS_CAP_FS_MASK)
+                               cap_lower(cred->cap_effective, cap);
+               }
+               old_cred = override_creds(cred);
+       }
+
+       ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM, 1);
+
+       if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
+               rc = lustre_check_remote_perm(inode, mask);
+       else
+               rc = ll_generic_permission(inode, mask, flags, ll_check_acl);
+
+       /* restore current process's credentials and FS capability */
+       if (squash_id) {
+               revert_creds(old_cred);
+               put_cred(cred);
+       }
 
        RETURN(rc);
 }
@@ -3608,15 +3657,15 @@ struct inode_operations ll_file_inode_operations = {
 /* dynamic ioctl number support routins */
 static struct llioc_ctl_data {
        struct rw_semaphore     ioc_sem;
-        cfs_list_t              ioc_head;
+       struct list_head        ioc_head;
 } llioc = {
-        __RWSEM_INITIALIZER(llioc.ioc_sem),
-        CFS_LIST_HEAD_INIT(llioc.ioc_head)
+       __RWSEM_INITIALIZER(llioc.ioc_sem),
+       LIST_HEAD_INIT(llioc.ioc_head)
 };
 
 
 struct llioc_data {
-        cfs_list_t              iocd_list;
+       struct list_head        iocd_list;
         unsigned int            iocd_size;
         llioc_callback_t        iocd_cb;
         unsigned int            iocd_count;
@@ -3645,7 +3694,7 @@ void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd)
         memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count);
 
        down_write(&llioc.ioc_sem);
-        cfs_list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
+       list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
        up_write(&llioc.ioc_sem);
 
         RETURN(in_data);
@@ -3659,11 +3708,11 @@ void ll_iocontrol_unregister(void *magic)
                 return;
 
        down_write(&llioc.ioc_sem);
-        cfs_list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
+       list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
                 if (tmp == magic) {
                         unsigned int size = tmp->iocd_size;
 
-                        cfs_list_del(&tmp->iocd_list);
+                       list_del(&tmp->iocd_list);
                        up_write(&llioc.ioc_sem);
 
                         OBD_FREE(tmp, size);
@@ -3687,7 +3736,7 @@ ll_iocontrol_call(struct inode *inode, struct file *file,
         int rc = -EINVAL, i;
 
        down_read(&llioc.ioc_sem);
-        cfs_list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
+       list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
                 for (i = 0; i < data->iocd_count; i++) {
                         if (cmd != data->iocd_cmd[i])
                                 continue;
@@ -3793,7 +3842,7 @@ static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
        if (body == NULL)
                GOTO(out, rc = -EPROTO);
 
-       lmmsize = body->eadatasize;
+       lmmsize = body->mbo_eadatasize;
        if (lmmsize == 0) /* empty layout */
                GOTO(out, rc = 0);
 
@@ -3956,8 +4005,8 @@ int ll_layout_refresh(struct inode *inode, __u32 *gen)
        struct ldlm_enqueue_info einfo = {
                .ei_type = LDLM_IBITS,
                .ei_mode = LCK_CR,
-               .ei_cb_bl = ll_md_blocking_ast,
-               .ei_cb_cp = ldlm_completion_ast,
+               .ei_cb_bl = &ll_md_blocking_ast,
+               .ei_cb_cp = &ldlm_completion_ast,
        };
        int rc;
        ENTRY;
@@ -4003,8 +4052,7 @@ again:
                          ll_get_fsname(inode->i_sb, NULL, 0),
                          PFID(&lli->lli_fid), inode);
 
-       rc = md_enqueue(sbi->ll_md_exp, &einfo, &it, op_data, &lockh,
-                       NULL, 0, NULL, 0);
+       rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL, &it, op_data, &lockh, 0);
        if (it.d.lustre.it_data != NULL)
                ptlrpc_req_finished(it.d.lustre.it_data);
        it.d.lustre.it_data = NULL;