summaryrefslogtreecommitdiffstats
path: root/security/nss/lib/softoken/legacydb/dbmshim.c
diff options
context:
space:
mode:
authorMatt A. Tobin <mattatobin@localhost.localdomain>2018-02-02 04:16:08 -0500
committerMatt A. Tobin <mattatobin@localhost.localdomain>2018-02-02 04:16:08 -0500
commit5f8de423f190bbb79a62f804151bc24824fa32d8 (patch)
tree10027f336435511475e392454359edea8e25895d /security/nss/lib/softoken/legacydb/dbmshim.c
parent49ee0794b5d912db1f95dce6eb52d781dc210db5 (diff)
downloadUXP-5f8de423f190bbb79a62f804151bc24824fa32d8.tar
UXP-5f8de423f190bbb79a62f804151bc24824fa32d8.tar.gz
UXP-5f8de423f190bbb79a62f804151bc24824fa32d8.tar.lz
UXP-5f8de423f190bbb79a62f804151bc24824fa32d8.tar.xz
UXP-5f8de423f190bbb79a62f804151bc24824fa32d8.zip
Add m-esr52 at 52.6.0
Diffstat (limited to 'security/nss/lib/softoken/legacydb/dbmshim.c')
-rw-r--r--security/nss/lib/softoken/legacydb/dbmshim.c600
1 files changed, 600 insertions, 0 deletions
diff --git a/security/nss/lib/softoken/legacydb/dbmshim.c b/security/nss/lib/softoken/legacydb/dbmshim.c
new file mode 100644
index 000000000..ae498faea
--- /dev/null
+++ b/security/nss/lib/softoken/legacydb/dbmshim.c
@@ -0,0 +1,600 @@
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+/*
+ * Berkeley DB 1.85 Shim code to handle blobs.
+ */
+#include "mcom_db.h"
+#include "secitem.h"
+#include "nssb64.h"
+#include "blapi.h"
+#include "secerr.h"
+
+#include "lgdb.h"
+
+/*
+ * Blob block:
+ * Byte 0 CERTDB Version -+ -+
+ * Byte 1 certDBEntryTypeBlob | BLOB_HEAD_LEN |
+ * Byte 2 flags (always '0'); | |
+ * Byte 3 reserved (always '0'); -+ |
+ * Byte 4 LSB length | <--BLOB_LENGTH_START | BLOB_BUF_LEN
+ * Byte 5 . | |
+ * Byte 6 . | BLOB_LENGTH_LEN |
+ * Byte 7 MSB length | |
+ * Byte 8 blob_filename -+ -+ <-- BLOB_NAME_START |
+ * Byte 9 . | BLOB_NAME_LEN |
+ * . . | |
+ * Byte 37 . -+ -+
+ */
+#define DBS_BLOCK_SIZE (16 * 1024) /* 16 k */
+#define DBS_MAX_ENTRY_SIZE (DBS_BLOCK_SIZE - (2048)) /* 14 k */
+#define DBS_CACHE_SIZE DBS_BLOCK_SIZE * 8
+#define ROUNDDIV(x, y) (x + (y - 1)) / y
+#define BLOB_HEAD_LEN 4
+#define BLOB_LENGTH_START BLOB_HEAD_LEN
+#define BLOB_LENGTH_LEN 4
+#define BLOB_NAME_START BLOB_LENGTH_START + BLOB_LENGTH_LEN
+#define BLOB_NAME_LEN 1 + ROUNDDIV(SHA1_LENGTH, 3) * 4 + 1
+#define BLOB_BUF_LEN BLOB_HEAD_LEN + BLOB_LENGTH_LEN + BLOB_NAME_LEN
+
+/* a Shim data structure. This data structure has a db built into it. */
+typedef struct DBSStr DBS;
+
+struct DBSStr {
+ DB db;
+ char *blobdir;
+ int mode;
+ PRBool readOnly;
+ PRFileMap *dbs_mapfile;
+ unsigned char *dbs_addr;
+ PRUint32 dbs_len;
+ char staticBlobArea[BLOB_BUF_LEN];
+};
+
+/*
+ * return true if the Datablock contains a blobtype
+ */
+static PRBool
+dbs_IsBlob(DBT *blobData)
+{
+ unsigned char *addr = (unsigned char *)blobData->data;
+ if (blobData->size < BLOB_BUF_LEN) {
+ return PR_FALSE;
+ }
+ return addr && ((certDBEntryType)addr[1] == certDBEntryTypeBlob);
+}
+
+/*
+ * extract the filename in the blob of the real data set.
+ * This value is not malloced (does not need to be freed by the caller.
+ */
+static const char *
+dbs_getBlobFileName(DBT *blobData)
+{
+ char *addr = (char *)blobData->data;
+
+ return &addr[BLOB_NAME_START];
+}
+
+/*
+ * extract the size of the actual blob from the blob record
+ */
+static PRUint32
+dbs_getBlobSize(DBT *blobData)
+{
+ unsigned char *addr = (unsigned char *)blobData->data;
+
+ return (PRUint32)(addr[BLOB_LENGTH_START + 3] << 24) |
+ (addr[BLOB_LENGTH_START + 2] << 16) |
+ (addr[BLOB_LENGTH_START + 1] << 8) |
+ addr[BLOB_LENGTH_START];
+}
+
+/* We are using base64 data for the filename, but base64 data can include a
+ * '/' which is interpreted as a path separator on many platforms. Replace it
+ * with an inocuous '-'. We don't need to convert back because we never actual
+ * decode the filename.
+ */
+
+static void
+dbs_replaceSlash(char *cp, int len)
+{
+ while (len--) {
+ if (*cp == '/')
+ *cp = '-';
+ cp++;
+ }
+}
+
+/*
+ * create a blob record from a key, data and return it in blobData.
+ * NOTE: The data element is static data (keeping with the dbm model).
+ */
+static void
+dbs_mkBlob(DBS *dbsp, const DBT *key, const DBT *data, DBT *blobData)
+{
+ unsigned char sha1_data[SHA1_LENGTH];
+ char *b = dbsp->staticBlobArea;
+ PRUint32 length = data->size;
+ SECItem sha1Item;
+
+ b[0] = CERT_DB_FILE_VERSION; /* certdb version number */
+ b[1] = (char)certDBEntryTypeBlob; /* type */
+ b[2] = 0; /* flags */
+ b[3] = 0; /* reserved */
+ b[BLOB_LENGTH_START] = length & 0xff;
+ b[BLOB_LENGTH_START + 1] = (length >> 8) & 0xff;
+ b[BLOB_LENGTH_START + 2] = (length >> 16) & 0xff;
+ b[BLOB_LENGTH_START + 3] = (length >> 24) & 0xff;
+ sha1Item.data = sha1_data;
+ sha1Item.len = SHA1_LENGTH;
+ SHA1_HashBuf(sha1_data, key->data, key->size);
+ b[BLOB_NAME_START] = 'b'; /* Make sure we start with a alpha */
+ NSSBase64_EncodeItem(NULL, &b[BLOB_NAME_START + 1], BLOB_NAME_LEN - 1, &sha1Item);
+ b[BLOB_BUF_LEN - 1] = 0;
+ dbs_replaceSlash(&b[BLOB_NAME_START + 1], BLOB_NAME_LEN - 1);
+ blobData->data = b;
+ blobData->size = BLOB_BUF_LEN;
+ return;
+}
+
+/*
+ * construct a path to the actual blob. The string returned must be
+ * freed by the caller with PR_smprintf_free.
+ *
+ * Note: this file does lots of consistancy checks on the DBT. The
+ * routines that call this depend on these checks, so they don't worry
+ * about them (success of this routine implies a good blobdata record).
+ */
+static char *
+dbs_getBlobFilePath(char *blobdir, DBT *blobData)
+{
+ const char *name;
+
+ if (blobdir == NULL) {
+ PR_SetError(SEC_ERROR_BAD_DATABASE, 0);
+ return NULL;
+ }
+ if (!dbs_IsBlob(blobData)) {
+ PR_SetError(SEC_ERROR_BAD_DATABASE, 0);
+ return NULL;
+ }
+ name = dbs_getBlobFileName(blobData);
+ if (!name || *name == 0) {
+ PR_SetError(SEC_ERROR_BAD_DATABASE, 0);
+ return NULL;
+ }
+ return PR_smprintf("%s" PATH_SEPARATOR "%s", blobdir, name);
+}
+
+/*
+ * Delete a blob file pointed to by the blob record.
+ */
+static void
+dbs_removeBlob(DBS *dbsp, DBT *blobData)
+{
+ char *file;
+
+ file = dbs_getBlobFilePath(dbsp->blobdir, blobData);
+ if (!file) {
+ return;
+ }
+ PR_Delete(file);
+ PR_smprintf_free(file);
+}
+
+/*
+ * Directory modes are slightly different, the 'x' bit needs to be on to
+ * access them. Copy all the read bits to 'x' bits
+ */
+static int
+dbs_DirMode(int mode)
+{
+ int x_bits = (mode >> 2) & 0111;
+ return mode | x_bits;
+}
+
+/*
+ * write a data blob to it's file. blobdData is the blob record that will be
+ * stored in the database. data is the actual data to go out on disk.
+ */
+static int
+dbs_writeBlob(DBS *dbsp, int mode, DBT *blobData, const DBT *data)
+{
+ char *file = NULL;
+ PRFileDesc *filed;
+ PRStatus status;
+ int len;
+ int error = 0;
+
+ file = dbs_getBlobFilePath(dbsp->blobdir, blobData);
+ if (!file) {
+ goto loser;
+ }
+ if (PR_Access(dbsp->blobdir, PR_ACCESS_EXISTS) != PR_SUCCESS) {
+ status = PR_MkDir(dbsp->blobdir, dbs_DirMode(mode));
+ if (status != PR_SUCCESS) {
+ goto loser;
+ }
+ }
+ filed = PR_OpenFile(file, PR_CREATE_FILE | PR_TRUNCATE | PR_WRONLY, mode);
+ if (filed == NULL) {
+ error = PR_GetError();
+ goto loser;
+ }
+ len = PR_Write(filed, data->data, data->size);
+ error = PR_GetError();
+ PR_Close(filed);
+ if (len < (int)data->size) {
+ goto loser;
+ }
+ PR_smprintf_free(file);
+ return 0;
+
+loser:
+ if (file) {
+ PR_Delete(file);
+ PR_smprintf_free(file);
+ }
+ /* don't let close or delete reset the error */
+ PR_SetError(error, 0);
+ return -1;
+}
+
+/*
+ * we need to keep a address map in memory between calls to DBM.
+ * remember what we have mapped can close it when we get another dbm
+ * call.
+ *
+ * NOTE: Not all platforms support mapped files. This code is designed to
+ * detect this at runtime. If map files aren't supported the OS will indicate
+ * this by failing the PR_Memmap call. In this case we emulate mapped files
+ * by just reading in the file into regular memory. We signal this state by
+ * making dbs_mapfile NULL and dbs_addr non-NULL.
+ */
+
+static void
+dbs_freemap(DBS *dbsp)
+{
+ if (dbsp->dbs_mapfile) {
+ PR_MemUnmap(dbsp->dbs_addr, dbsp->dbs_len);
+ PR_CloseFileMap(dbsp->dbs_mapfile);
+ dbsp->dbs_mapfile = NULL;
+ dbsp->dbs_addr = NULL;
+ dbsp->dbs_len = 0;
+ } else if (dbsp->dbs_addr) {
+ PORT_Free(dbsp->dbs_addr);
+ dbsp->dbs_addr = NULL;
+ dbsp->dbs_len = 0;
+ }
+ return;
+}
+
+static void
+dbs_setmap(DBS *dbsp, PRFileMap *mapfile, unsigned char *addr, PRUint32 len)
+{
+ dbsp->dbs_mapfile = mapfile;
+ dbsp->dbs_addr = addr;
+ dbsp->dbs_len = len;
+}
+
+/*
+ * platforms that cannot map the file need to read it into a temp buffer.
+ */
+static unsigned char *
+dbs_EmulateMap(PRFileDesc *filed, int len)
+{
+ unsigned char *addr;
+ PRInt32 dataRead;
+
+ addr = PORT_Alloc(len);
+ if (addr == NULL) {
+ return NULL;
+ }
+
+ dataRead = PR_Read(filed, addr, len);
+ if (dataRead != len) {
+ PORT_Free(addr);
+ if (dataRead > 0) {
+ /* PR_Read didn't set an error, we need to */
+ PR_SetError(SEC_ERROR_BAD_DATABASE, 0);
+ }
+ return NULL;
+ }
+
+ return addr;
+}
+
+/*
+ * pull a database record off the disk
+ * data points to the blob record on input and the real record (if we could
+ * read it) on output. if there is an error data is not modified.
+ */
+static int
+dbs_readBlob(DBS *dbsp, DBT *data)
+{
+ char *file = NULL;
+ PRFileDesc *filed = NULL;
+ PRFileMap *mapfile = NULL;
+ unsigned char *addr = NULL;
+ int error;
+ int len = -1;
+
+ file = dbs_getBlobFilePath(dbsp->blobdir, data);
+ if (!file) {
+ goto loser;
+ }
+ filed = PR_OpenFile(file, PR_RDONLY, 0);
+ PR_smprintf_free(file);
+ file = NULL;
+ if (filed == NULL) {
+ goto loser;
+ }
+
+ len = dbs_getBlobSize(data);
+ /* Bug 1323150
+ * PR_MemMap fails on Windows for larger certificates.
+ * https://msdn.microsoft.com/en-us/library/windows/desktop/aa366761(v=vs.85).aspx
+ * Let's always use the emulated map, i.e. read the file.
+ */
+ addr = dbs_EmulateMap(filed, len);
+ if (addr == NULL) {
+ goto loser;
+ }
+ PR_Close(filed);
+ dbs_setmap(dbsp, mapfile, addr, len);
+
+ data->data = addr;
+ data->size = len;
+ return 0;
+
+loser:
+ /* preserve the error code */
+ error = PR_GetError();
+ if (mapfile) {
+ PR_CloseFileMap(mapfile);
+ }
+ if (filed) {
+ PR_Close(filed);
+ }
+ PR_SetError(error, 0);
+ return -1;
+}
+
+/*
+ * actual DBM shims
+ */
+static int
+dbs_get(const DB *dbs, const DBT *key, DBT *data, unsigned int flags)
+{
+ int ret;
+ DBS *dbsp = (DBS *)dbs;
+ DB *db = (DB *)dbs->internal;
+
+ dbs_freemap(dbsp);
+
+ ret = (*db->get)(db, key, data, flags);
+ if ((ret == 0) && dbs_IsBlob(data)) {
+ ret = dbs_readBlob(dbsp, data);
+ }
+
+ return (ret);
+}
+
+static int
+dbs_put(const DB *dbs, DBT *key, const DBT *data, unsigned int flags)
+{
+ DBT blob;
+ int ret = 0;
+ DBS *dbsp = (DBS *)dbs;
+ DB *db = (DB *)dbs->internal;
+
+ dbs_freemap(dbsp);
+
+ /* If the db is readonly, just pass the data down to rdb and let it fail */
+ if (!dbsp->readOnly) {
+ DBT oldData;
+ int ret1;
+
+ /* make sure the current record is deleted if it's a blob */
+ ret1 = (*db->get)(db, key, &oldData, 0);
+ if ((ret1 == 0) && flags == R_NOOVERWRITE) {
+ /* let DBM return the error to maintain consistancy */
+ return (*db->put)(db, key, data, flags);
+ }
+ if ((ret1 == 0) && dbs_IsBlob(&oldData)) {
+ dbs_removeBlob(dbsp, &oldData);
+ }
+
+ if (data->size > DBS_MAX_ENTRY_SIZE) {
+ dbs_mkBlob(dbsp, key, data, &blob);
+ ret = dbs_writeBlob(dbsp, dbsp->mode, &blob, data);
+ data = &blob;
+ }
+ }
+
+ if (ret == 0) {
+ ret = (*db->put)(db, key, data, flags);
+ }
+ return (ret);
+}
+
+static int
+dbs_sync(const DB *dbs, unsigned int flags)
+{
+ DB *db = (DB *)dbs->internal;
+ DBS *dbsp = (DBS *)dbs;
+
+ dbs_freemap(dbsp);
+
+ return (*db->sync)(db, flags);
+}
+
+static int
+dbs_del(const DB *dbs, const DBT *key, unsigned int flags)
+{
+ int ret;
+ DBS *dbsp = (DBS *)dbs;
+ DB *db = (DB *)dbs->internal;
+
+ dbs_freemap(dbsp);
+
+ if (!dbsp->readOnly) {
+ DBT oldData;
+ ret = (*db->get)(db, key, &oldData, 0);
+ if ((ret == 0) && dbs_IsBlob(&oldData)) {
+ dbs_removeBlob(dbsp, &oldData);
+ }
+ }
+
+ return (*db->del)(db, key, flags);
+}
+
+static int
+dbs_seq(const DB *dbs, DBT *key, DBT *data, unsigned int flags)
+{
+ int ret;
+ DBS *dbsp = (DBS *)dbs;
+ DB *db = (DB *)dbs->internal;
+
+ dbs_freemap(dbsp);
+
+ ret = (*db->seq)(db, key, data, flags);
+ if ((ret == 0) && dbs_IsBlob(data)) {
+ /* don't return a blob read as an error so traversals keep going */
+ (void)dbs_readBlob(dbsp, data);
+ }
+
+ return (ret);
+}
+
+static int
+dbs_close(DB *dbs)
+{
+ DBS *dbsp = (DBS *)dbs;
+ DB *db = (DB *)dbs->internal;
+ int ret;
+
+ dbs_freemap(dbsp);
+ ret = (*db->close)(db);
+ PORT_Free(dbsp->blobdir);
+ PORT_Free(dbsp);
+ return ret;
+}
+
+static int
+dbs_fd(const DB *dbs)
+{
+ DB *db = (DB *)dbs->internal;
+
+ return (*db->fd)(db);
+}
+
+/*
+ * the naming convention we use is
+ * change the .xxx into .dir. (for nss it's always .db);
+ * if no .extension exists or is equal to .dir, add a .dir
+ * the returned data must be freed.
+ */
+#define DIRSUFFIX ".dir"
+static char *
+dbs_mkBlobDirName(const char *dbname)
+{
+ int dbname_len = PORT_Strlen(dbname);
+ int dbname_end = dbname_len;
+ const char *cp;
+ char *blobDir = NULL;
+
+ /* scan back from the end looking for either a directory separator, a '.',
+ * or the end of the string. NOTE: Windows should check for both separators
+ * here. For now this is safe because we know NSS always uses a '.'
+ */
+ for (cp = &dbname[dbname_len];
+ (cp > dbname) && (*cp != '.') && (*cp != *PATH_SEPARATOR);
+ cp--)
+ /* Empty */;
+ if (*cp == '.') {
+ dbname_end = cp - dbname;
+ if (PORT_Strcmp(cp, DIRSUFFIX) == 0) {
+ dbname_end = dbname_len;
+ }
+ }
+ blobDir = PORT_ZAlloc(dbname_end + sizeof(DIRSUFFIX));
+ if (blobDir == NULL) {
+ return NULL;
+ }
+ PORT_Memcpy(blobDir, dbname, dbname_end);
+ PORT_Memcpy(&blobDir[dbname_end], DIRSUFFIX, sizeof(DIRSUFFIX));
+ return blobDir;
+}
+
+#define DBM_DEFAULT 0
+static const HASHINFO dbs_hashInfo = {
+ DBS_BLOCK_SIZE, /* bucket size, must be greater than = to
+ * or maximum entry size (+ header)
+ * we allow before blobing */
+ DBM_DEFAULT, /* Fill Factor */
+ DBM_DEFAULT, /* number of elements */
+ DBS_CACHE_SIZE, /* cache size */
+ DBM_DEFAULT, /* hash function */
+ DBM_DEFAULT, /* byte order */
+};
+
+/*
+ * the open function. NOTE: this is the only exposed function in this file.
+ * everything else is called through the function table pointer.
+ */
+DB *
+dbsopen(const char *dbname, int flags, int mode, DBTYPE type,
+ const void *userData)
+{
+ DB *db = NULL, *dbs = NULL;
+ DBS *dbsp = NULL;
+
+ /* NOTE: we are overriding userData with dbs_hashInfo. since all known
+ * callers pass 0, this is ok, otherwise we should merge the two */
+
+ dbsp = (DBS *)PORT_ZAlloc(sizeof(DBS));
+ if (!dbsp) {
+ return NULL;
+ }
+ dbs = &dbsp->db;
+
+ dbsp->blobdir = dbs_mkBlobDirName(dbname);
+ if (dbsp->blobdir == NULL) {
+ goto loser;
+ }
+ dbsp->mode = mode;
+ dbsp->readOnly = (PRBool)(flags == NO_RDONLY);
+ dbsp->dbs_mapfile = NULL;
+ dbsp->dbs_addr = NULL;
+ dbsp->dbs_len = 0;
+
+ /* the real dbm call */
+ db = dbopen(dbname, flags, mode, type, &dbs_hashInfo);
+ if (db == NULL) {
+ goto loser;
+ }
+ dbs->internal = (void *)db;
+ dbs->type = type;
+ dbs->close = dbs_close;
+ dbs->get = dbs_get;
+ dbs->del = dbs_del;
+ dbs->put = dbs_put;
+ dbs->seq = dbs_seq;
+ dbs->sync = dbs_sync;
+ dbs->fd = dbs_fd;
+
+ return dbs;
+loser:
+ if (db) {
+ (*db->close)(db);
+ }
+ if (dbsp->blobdir) {
+ PORT_Free(dbsp->blobdir);
+ }
+ PORT_Free(dbsp);
+ return NULL;
+}