X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=nfs3%2Fnfs3.c;h=f7e7b4ef96ead8a8cb6fbd772f810841fe1ab6df;hb=4db23448ce252728f10df967b55b2787d5fb8db5;hp=72ec9cb7a26173e6a0732a428a901d0c0af99597;hpb=50c08ba526a6638e8e3c4eec0503365a2c110a85;p=bluesky.git diff --git a/nfs3/nfs3.c b/nfs3/nfs3.c index 72ec9cb..f7e7b4e 100644 --- a/nfs3/nfs3.c +++ b/nfs3/nfs3.c @@ -9,6 +9,9 @@ extern BlueSkyFS *fs; +#define NFS_BLOCKSIZE 32768 +#define NFS_MAXSIZE (1 << 20) + /* Check that a string is a valid file name. We require that it be valid * UTF-8, that it not be empty, and that it not contain embedded forward * slashes. Also checks that the length of the string is not more than the @@ -243,10 +246,14 @@ void nfsproc3_lookup_3_svc(diropargs3 *argp, RPCRequest *req) async_rpc_send_reply(req, &result); return; } + + result.lookup3res_u.resok.dir_attributes.present = TRUE; + encode_fattr3(&result.lookup3res_u.resok.dir_attributes.post_op_attr_u.attributes, dir); + g_mutex_unlock(dir->lock); + BlueSkyInode *inode = bluesky_get_inode(fs, inum); if (inode == NULL) { result.status = NFS3ERR_NOENT; - g_mutex_unlock(dir->lock); async_rpc_send_reply(req, &result); return; } @@ -254,8 +261,6 @@ void nfsproc3_lookup_3_svc(diropargs3 *argp, RPCRequest *req) schedule_inode_unref(req, inode); result.status = NFS3_OK; - result.lookup3res_u.resok.dir_attributes.present = TRUE; - encode_fattr3(&result.lookup3res_u.resok.dir_attributes.post_op_attr_u.attributes, dir); result.lookup3res_u.resok.obj_attributes.present = TRUE; encode_fattr3(&result.lookup3res_u.resok.obj_attributes.post_op_attr_u.attributes, inode); @@ -265,7 +270,6 @@ void nfsproc3_lookup_3_svc(diropargs3 *argp, RPCRequest *req) result.lookup3res_u.resok.object.data.data_val = (char *)&fh_bytes; g_mutex_unlock(inode->lock); - g_mutex_unlock(dir->lock); async_rpc_send_reply(req, &result); } @@ -322,7 +326,9 @@ void nfsproc3_read_3_svc(read3args *argp, RPCRequest *req) { read3res result; memset(&result, 0, sizeof(result)); - char buf[32768]; + char buf[NFS_MAXSIZE]; + + bluesky_flushd_invoke_conditional(fs); BlueSkyInode *inode = lookup_fh(req, &argp->file); if (inode == NULL) { @@ -339,6 +345,7 @@ void nfsproc3_read_3_svc(read3args *argp, RPCRequest *req) count = 0; result.read3res_u.resok.eof = TRUE; } else { + count = MIN(count, NFS_MAXSIZE); count = MIN(count, inode->size - argp->offset); if (argp->offset + count == inode->size) result.read3res_u.resok.eof = TRUE; @@ -367,6 +374,8 @@ void nfsproc3_write_3_svc(write3args *argp, RPCRequest *req) struct wcc_data wcc; memset(&wcc, 0, sizeof(wcc)); + bluesky_flushd_invoke_conditional(fs); + BlueSkyInode *inode = lookup_fh(req, &argp->file); if (inode == NULL) { result.status = NFS3ERR_STALE; @@ -375,6 +384,21 @@ void nfsproc3_write_3_svc(write3args *argp, RPCRequest *req) return; } +#if 0 + /* FIXME: Hack to throttle writes when there is too much dirty data still + * to be written out. */ + while (g_atomic_int_get(&fs->cache_dirty) > 4096 + || g_atomic_int_get(&fs->cache_total) > 8192) { + g_print("Too many dirty pages (%d) or total pages (%d); throttling writes...\n", + g_atomic_int_get(&fs->cache_dirty), + g_atomic_int_get(&fs->cache_total)); + struct timespec delay; + delay.tv_sec = 2; + delay.tv_nsec = 0; + nanosleep(&delay, NULL); + } +#endif + g_mutex_lock(inode->lock); encode_pre_wcc(&wcc, inode); @@ -749,17 +773,22 @@ void nfsproc3_rename_3_svc(rename3args *argp, RPCRequest *req) async_rpc_send_reply(req, &result); return; } - g_mutex_lock(dir1->lock); - encode_pre_wcc(wcc1, dir1); BlueSkyInode *dir2 = lookup_fh(req, &argp->to.dir); if (dir2 == NULL) { result.status = NFS3ERR_STALE; - g_mutex_unlock(dir1->lock); async_rpc_send_reply(req, &result); return; } - g_mutex_lock(dir2->lock); + + if (dir1->inum < dir2->inum) { + g_mutex_lock(dir1->lock); + g_mutex_lock(dir2->lock); + } else if (dir1->inum > dir2->inum) { + g_mutex_lock(dir2->lock); + g_mutex_lock(dir1->lock); + } + encode_pre_wcc(wcc1, dir1); encode_pre_wcc(wcc2, dir1); gboolean status = bluesky_rename(dir1, argp->from.name, @@ -775,8 +804,9 @@ void nfsproc3_rename_3_svc(rename3args *argp, RPCRequest *req) else result.status = NFS3ERR_PERM; - g_mutex_unlock(dir2->lock); g_mutex_unlock(dir1->lock); + if (dir1->inum != dir2->inum) + g_mutex_unlock(dir2->lock); async_rpc_send_reply(req, &result); } @@ -937,17 +967,37 @@ void nfsproc3_readdirplus_3_svc(readdirplus3args *argp, RPCRequest *req) uint64_t fh_bytes[MAX_READDIR_DIRENTS]; int count = 0; - /* TODO: Handle dircount, maxcount arguments from client. */ - + GSequenceIter *i; BlueSkyDirent start = {NULL, NULL, argp->cookie, 0}; - GSequenceIter *i = g_sequence_search(dir->dirents, &start, - bluesky_dirent_compare, NULL); + /* Perform a prefetch pass on inodes: for all the inodes we think we will + * return information about, try to load each one but don't wait. This + * should let multiple inodes be fetched in parallel, instead of + * sequentially in the loop that follows. */ + i = g_sequence_search(dir->dirents, &start, bluesky_dirent_compare, NULL); + while (count < MAX_READDIR_DIRENTS + && !g_sequence_iter_is_end(i) + && dircount <= argp->dircount + && dircount + attrcount <= argp->maxcount) + { + BlueSkyDirent *d = g_sequence_get(i); + BlueSkyInode *inode = bluesky_get_inode(fs, d->inum); + if (inode != NULL) + bluesky_inode_unref(inode); + dircount += 24 + ((strlen(d->name) + 3) & ~3); + attrcount += 88 + 8 + 8; + i = g_sequence_iter_next(i); + } + + i = g_sequence_search(dir->dirents, &start, bluesky_dirent_compare, NULL); + count = 0; + dircount = 88 + 16; + attrcount = 0; while (count < MAX_READDIR_DIRENTS && !g_sequence_iter_is_end(i)) { BlueSkyDirent *d = g_sequence_get(i); BlueSkyInode *inode = bluesky_get_inode(fs, d->inum); - g_mutex_lock(inode->lock); if (inode != NULL) { + g_mutex_lock(inode->lock); dircount += 24 + ((strlen(d->name) + 3) & ~3); attrcount += 88 + 8 + 8; if (dircount > argp->dircount @@ -1027,13 +1077,13 @@ void nfsproc3_fsinfo_3_svc(nfs_fh3 *argp, RPCRequest *req) result.status = NFS3_OK; result.fsinfo3res_u.resok.obj_attributes.present = TRUE; encode_fattr3(&result.fsinfo3res_u.resok.obj_attributes.post_op_attr_u.attributes, inode); - result.fsinfo3res_u.resok.rtmax = 32768; - result.fsinfo3res_u.resok.rtpref = 32768; - result.fsinfo3res_u.resok.rtmult = 4096; - result.fsinfo3res_u.resok.wtmax = 32768; - result.fsinfo3res_u.resok.wtpref = 32768; - result.fsinfo3res_u.resok.wtmult = 4096; - result.fsinfo3res_u.resok.dtpref = 4096; + result.fsinfo3res_u.resok.rtmax = NFS_MAXSIZE; + result.fsinfo3res_u.resok.rtpref = NFS_MAXSIZE; + result.fsinfo3res_u.resok.rtmult = NFS_BLOCKSIZE; + result.fsinfo3res_u.resok.wtmax = NFS_MAXSIZE; + result.fsinfo3res_u.resok.wtpref = NFS_MAXSIZE; + result.fsinfo3res_u.resok.wtmult = NFS_BLOCKSIZE; + result.fsinfo3res_u.resok.dtpref = NFS_BLOCKSIZE; result.fsinfo3res_u.resok.maxfilesize = 0x7fffffffffffffffULL; result.fsinfo3res_u.resok.time_delta.seconds = 0; result.fsinfo3res_u.resok.time_delta.nseconds = 1000; @@ -1072,7 +1122,24 @@ void nfsproc3_commit_3_svc(commit3args *argp, RPCRequest *req) commit3res result; memset(&result, 0, sizeof(result)); - result.status = NFS3ERR_NOTSUPP; + result.status = NFS3_OK; + + BlueSkyInode *inode = lookup_fh(req, &argp->file); + if (inode == NULL) { + result.status = NFS3ERR_STALE; + async_rpc_send_reply(req, &result); + return; + } + + g_mutex_lock(inode->lock); + encode_pre_wcc(&result.commit3res_u.resok.file_wcc, inode); + + //bluesky_inode_do_sync(inode); + + result.commit3res_u.resok.file_wcc.after.present = TRUE; + encode_fattr3(&result.commit3res_u.resok.file_wcc.after.post_op_attr_u.attributes, inode); + + g_mutex_unlock(inode->lock); async_rpc_send_reply(req, &result); }