]> Git Repo - qemu.git/blobdiff - dump.c
hw/pci-host/apb.c: Avoid shifting left into sign bit
[qemu.git] / dump.c
diff --git a/dump.c b/dump.c
index 80a9116c778830728e0b541383b2fbd52d8e2e99..14b3d1d6aea64151b4ac4951c2f81f6ce1e24ecd 100644 (file)
--- a/dump.c
+++ b/dump.c
 #include "qapi/error.h"
 #include "qmp-commands.h"
 
+#include <zlib.h>
+#ifdef CONFIG_LZO
+#include <lzo/lzo1x.h>
+#endif
+#ifdef CONFIG_SNAPPY
+#include <snappy-c.h>
+#endif
+#ifndef ELF_MACHINE_UNAME
+#define ELF_MACHINE_UNAME "Unknown"
+#endif
+
 static uint16_t cpu_convert_to_target16(uint16_t val, int endian)
 {
     if (endian == ELFDATA2LSB) {
@@ -76,6 +87,19 @@ typedef struct DumpState {
     int64_t begin;
     int64_t length;
     Error **errp;
+
+    uint8_t *note_buf;          /* buffer for notes */
+    size_t note_buf_offset;     /* the writing place in note_buf */
+    uint32_t nr_cpus;           /* number of guest's cpu */
+    size_t page_size;           /* guest's page size */
+    uint32_t page_shift;        /* guest's page shift */
+    uint64_t max_mapnr;         /* the biggest guest's phys-mem's number */
+    size_t len_dump_bitmap;     /* the size of the place used to store
+                                   dump_bitmap in vmcore */
+    off_t offset_dump_bitmap;   /* offset of dump_bitmap part in vmcore */
+    off_t offset_page;          /* offset of page part in vmcore */
+    size_t num_dumpable;        /* number of page that can be dumped */
+    uint32_t flag_compress;     /* indicate the compression format */
 } DumpState;
 
 static int dump_cleanup(DumpState *s)
@@ -99,7 +123,7 @@ static void dump_error(DumpState *s, const char *reason)
     dump_cleanup(s);
 }
 
-static int fd_write_vmcore(void *buf, size_t size, void *opaque)
+static int fd_write_vmcore(const void *buf, size_t size, void *opaque)
 {
     DumpState *s = opaque;
     size_t written_size;
@@ -271,7 +295,7 @@ static inline int cpu_index(CPUState *cpu)
     return cpu->cpu_index + 1;
 }
 
-static int write_elf64_notes(DumpState *s)
+static int write_elf64_notes(WriteCoreDumpFunction f, DumpState *s)
 {
     CPUState *cpu;
     int ret;
@@ -279,7 +303,7 @@ static int write_elf64_notes(DumpState *s)
 
     CPU_FOREACH(cpu) {
         id = cpu_index(cpu);
-        ret = cpu_write_elf64_note(fd_write_vmcore, cpu, id, s);
+        ret = cpu_write_elf64_note(f, cpu, id, s);
         if (ret < 0) {
             dump_error(s, "dump: failed to write elf notes.\n");
             return -1;
@@ -287,7 +311,7 @@ static int write_elf64_notes(DumpState *s)
     }
 
     CPU_FOREACH(cpu) {
-        ret = cpu_write_elf64_qemunote(fd_write_vmcore, cpu, s);
+        ret = cpu_write_elf64_qemunote(f, cpu, s);
         if (ret < 0) {
             dump_error(s, "dump: failed to write CPU status.\n");
             return -1;
@@ -321,7 +345,7 @@ static int write_elf32_note(DumpState *s)
     return 0;
 }
 
-static int write_elf32_notes(DumpState *s)
+static int write_elf32_notes(WriteCoreDumpFunction f, DumpState *s)
 {
     CPUState *cpu;
     int ret;
@@ -329,7 +353,7 @@ static int write_elf32_notes(DumpState *s)
 
     CPU_FOREACH(cpu) {
         id = cpu_index(cpu);
-        ret = cpu_write_elf32_note(fd_write_vmcore, cpu, id, s);
+        ret = cpu_write_elf32_note(f, cpu, id, s);
         if (ret < 0) {
             dump_error(s, "dump: failed to write elf notes.\n");
             return -1;
@@ -337,7 +361,7 @@ static int write_elf32_notes(DumpState *s)
     }
 
     CPU_FOREACH(cpu) {
-        ret = cpu_write_elf32_qemunote(fd_write_vmcore, cpu, s);
+        ret = cpu_write_elf32_qemunote(f, cpu, s);
         if (ret < 0) {
             dump_error(s, "dump: failed to write CPU status.\n");
             return -1;
@@ -574,7 +598,7 @@ static int dump_begin(DumpState *s)
         }
 
         /* write notes to vmcore */
-        if (write_elf64_notes(s) < 0) {
+        if (write_elf64_notes(fd_write_vmcore, s) < 0) {
             return -1;
         }
 
@@ -597,7 +621,7 @@ static int dump_begin(DumpState *s)
         }
 
         /* write notes to vmcore */
-        if (write_elf32_notes(s) < 0) {
+        if (write_elf32_notes(fd_write_vmcore, s) < 0) {
             return -1;
         }
     }
@@ -686,6 +710,800 @@ static int create_vmcore(DumpState *s)
     return 0;
 }
 
+static int write_start_flat_header(int fd)
+{
+    uint8_t *buf;
+    MakedumpfileHeader mh;
+    int ret = 0;
+
+    memset(&mh, 0, sizeof(mh));
+    strncpy(mh.signature, MAKEDUMPFILE_SIGNATURE,
+            strlen(MAKEDUMPFILE_SIGNATURE));
+
+    mh.type = cpu_to_be64(TYPE_FLAT_HEADER);
+    mh.version = cpu_to_be64(VERSION_FLAT_HEADER);
+
+    buf = g_malloc0(MAX_SIZE_MDF_HEADER);
+    memcpy(buf, &mh, sizeof(mh));
+
+    size_t written_size;
+    written_size = qemu_write_full(fd, buf, MAX_SIZE_MDF_HEADER);
+    if (written_size != MAX_SIZE_MDF_HEADER) {
+        ret = -1;
+    }
+
+    g_free(buf);
+    return ret;
+}
+
+static int write_end_flat_header(int fd)
+{
+    MakedumpfileDataHeader mdh;
+
+    mdh.offset = END_FLAG_FLAT_HEADER;
+    mdh.buf_size = END_FLAG_FLAT_HEADER;
+
+    size_t written_size;
+    written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
+    if (written_size != sizeof(mdh)) {
+        return -1;
+    }
+
+    return 0;
+}
+
+static int write_buffer(int fd, off_t offset, const void *buf, size_t size)
+{
+    size_t written_size;
+    MakedumpfileDataHeader mdh;
+
+    mdh.offset = cpu_to_be64(offset);
+    mdh.buf_size = cpu_to_be64(size);
+
+    written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
+    if (written_size != sizeof(mdh)) {
+        return -1;
+    }
+
+    written_size = qemu_write_full(fd, buf, size);
+    if (written_size != size) {
+        return -1;
+    }
+
+    return 0;
+}
+
+static int buf_write_note(const void *buf, size_t size, void *opaque)
+{
+    DumpState *s = opaque;
+
+    /* note_buf is not enough */
+    if (s->note_buf_offset + size > s->note_size) {
+        return -1;
+    }
+
+    memcpy(s->note_buf + s->note_buf_offset, buf, size);
+
+    s->note_buf_offset += size;
+
+    return 0;
+}
+
+/* write common header, sub header and elf note to vmcore */
+static int create_header32(DumpState *s)
+{
+    int ret = 0;
+    DiskDumpHeader32 *dh = NULL;
+    KdumpSubHeader32 *kh = NULL;
+    size_t size;
+    int endian = s->dump_info.d_endian;
+    uint32_t block_size;
+    uint32_t sub_hdr_size;
+    uint32_t bitmap_blocks;
+    uint32_t status = 0;
+    uint64_t offset_note;
+
+    /* write common header, the version of kdump-compressed format is 6th */
+    size = sizeof(DiskDumpHeader32);
+    dh = g_malloc0(size);
+
+    strncpy(dh->signature, KDUMP_SIGNATURE, strlen(KDUMP_SIGNATURE));
+    dh->header_version = cpu_convert_to_target32(6, endian);
+    block_size = s->page_size;
+    dh->block_size = cpu_convert_to_target32(block_size, endian);
+    sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
+    sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
+    dh->sub_hdr_size = cpu_convert_to_target32(sub_hdr_size, endian);
+    /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
+    dh->max_mapnr = cpu_convert_to_target32(MIN(s->max_mapnr, UINT_MAX),
+                                            endian);
+    dh->nr_cpus = cpu_convert_to_target32(s->nr_cpus, endian);
+    bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
+    dh->bitmap_blocks = cpu_convert_to_target32(bitmap_blocks, endian);
+    strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
+
+    if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
+        status |= DUMP_DH_COMPRESSED_ZLIB;
+    }
+#ifdef CONFIG_LZO
+    if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
+        status |= DUMP_DH_COMPRESSED_LZO;
+    }
+#endif
+#ifdef CONFIG_SNAPPY
+    if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
+        status |= DUMP_DH_COMPRESSED_SNAPPY;
+    }
+#endif
+    dh->status = cpu_convert_to_target32(status, endian);
+
+    if (write_buffer(s->fd, 0, dh, size) < 0) {
+        dump_error(s, "dump: failed to write disk dump header.\n");
+        ret = -1;
+        goto out;
+    }
+
+    /* write sub header */
+    size = sizeof(KdumpSubHeader32);
+    kh = g_malloc0(size);
+
+    /* 64bit max_mapnr_64 */
+    kh->max_mapnr_64 = cpu_convert_to_target64(s->max_mapnr, endian);
+    kh->phys_base = cpu_convert_to_target32(PHYS_BASE, endian);
+    kh->dump_level = cpu_convert_to_target32(DUMP_LEVEL, endian);
+
+    offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
+    kh->offset_note = cpu_convert_to_target64(offset_note, endian);
+    kh->note_size = cpu_convert_to_target32(s->note_size, endian);
+
+    if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
+                     block_size, kh, size) < 0) {
+        dump_error(s, "dump: failed to write kdump sub header.\n");
+        ret = -1;
+        goto out;
+    }
+
+    /* write note */
+    s->note_buf = g_malloc0(s->note_size);
+    s->note_buf_offset = 0;
+
+    /* use s->note_buf to store notes temporarily */
+    if (write_elf32_notes(buf_write_note, s) < 0) {
+        ret = -1;
+        goto out;
+    }
+
+    if (write_buffer(s->fd, offset_note, s->note_buf,
+                     s->note_size) < 0) {
+        dump_error(s, "dump: failed to write notes");
+        ret = -1;
+        goto out;
+    }
+
+    /* get offset of dump_bitmap */
+    s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
+                             block_size;
+
+    /* get offset of page */
+    s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
+                     block_size;
+
+out:
+    g_free(dh);
+    g_free(kh);
+    g_free(s->note_buf);
+
+    return ret;
+}
+
+/* write common header, sub header and elf note to vmcore */
+static int create_header64(DumpState *s)
+{
+    int ret = 0;
+    DiskDumpHeader64 *dh = NULL;
+    KdumpSubHeader64 *kh = NULL;
+    size_t size;
+    int endian = s->dump_info.d_endian;
+    uint32_t block_size;
+    uint32_t sub_hdr_size;
+    uint32_t bitmap_blocks;
+    uint32_t status = 0;
+    uint64_t offset_note;
+
+    /* write common header, the version of kdump-compressed format is 6th */
+    size = sizeof(DiskDumpHeader64);
+    dh = g_malloc0(size);
+
+    strncpy(dh->signature, KDUMP_SIGNATURE, strlen(KDUMP_SIGNATURE));
+    dh->header_version = cpu_convert_to_target32(6, endian);
+    block_size = s->page_size;
+    dh->block_size = cpu_convert_to_target32(block_size, endian);
+    sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
+    sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
+    dh->sub_hdr_size = cpu_convert_to_target32(sub_hdr_size, endian);
+    /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
+    dh->max_mapnr = cpu_convert_to_target32(MIN(s->max_mapnr, UINT_MAX),
+                                            endian);
+    dh->nr_cpus = cpu_convert_to_target32(s->nr_cpus, endian);
+    bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
+    dh->bitmap_blocks = cpu_convert_to_target32(bitmap_blocks, endian);
+    strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
+
+    if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
+        status |= DUMP_DH_COMPRESSED_ZLIB;
+    }
+#ifdef CONFIG_LZO
+    if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
+        status |= DUMP_DH_COMPRESSED_LZO;
+    }
+#endif
+#ifdef CONFIG_SNAPPY
+    if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
+        status |= DUMP_DH_COMPRESSED_SNAPPY;
+    }
+#endif
+    dh->status = cpu_convert_to_target32(status, endian);
+
+    if (write_buffer(s->fd, 0, dh, size) < 0) {
+        dump_error(s, "dump: failed to write disk dump header.\n");
+        ret = -1;
+        goto out;
+    }
+
+    /* write sub header */
+    size = sizeof(KdumpSubHeader64);
+    kh = g_malloc0(size);
+
+    /* 64bit max_mapnr_64 */
+    kh->max_mapnr_64 = cpu_convert_to_target64(s->max_mapnr, endian);
+    kh->phys_base = cpu_convert_to_target64(PHYS_BASE, endian);
+    kh->dump_level = cpu_convert_to_target32(DUMP_LEVEL, endian);
+
+    offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
+    kh->offset_note = cpu_convert_to_target64(offset_note, endian);
+    kh->note_size = cpu_convert_to_target64(s->note_size, endian);
+
+    if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
+                     block_size, kh, size) < 0) {
+        dump_error(s, "dump: failed to write kdump sub header.\n");
+        ret = -1;
+        goto out;
+    }
+
+    /* write note */
+    s->note_buf = g_malloc0(s->note_size);
+    s->note_buf_offset = 0;
+
+    /* use s->note_buf to store notes temporarily */
+    if (write_elf64_notes(buf_write_note, s) < 0) {
+        ret = -1;
+        goto out;
+    }
+
+    if (write_buffer(s->fd, offset_note, s->note_buf,
+                     s->note_size) < 0) {
+        dump_error(s, "dump: failed to write notes");
+        ret = -1;
+        goto out;
+    }
+
+    /* get offset of dump_bitmap */
+    s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
+                             block_size;
+
+    /* get offset of page */
+    s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
+                     block_size;
+
+out:
+    g_free(dh);
+    g_free(kh);
+    g_free(s->note_buf);
+
+    return ret;
+}
+
+static int write_dump_header(DumpState *s)
+{
+    if (s->dump_info.d_machine == EM_386) {
+        return create_header32(s);
+    } else {
+        return create_header64(s);
+    }
+}
+
+/*
+ * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
+ * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
+ * set_dump_bitmap will always leave the recently set bit un-sync. And setting
+ * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
+ * vmcore, ie. synchronizing un-sync bit into vmcore.
+ */
+static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
+                           uint8_t *buf, DumpState *s)
+{
+    off_t old_offset, new_offset;
+    off_t offset_bitmap1, offset_bitmap2;
+    uint32_t byte, bit;
+
+    /* should not set the previous place */
+    assert(last_pfn <= pfn);
+
+    /*
+     * if the bit needed to be set is not cached in buf, flush the data in buf
+     * to vmcore firstly.
+     * making new_offset be bigger than old_offset can also sync remained data
+     * into vmcore.
+     */
+    old_offset = BUFSIZE_BITMAP * (last_pfn / PFN_BUFBITMAP);
+    new_offset = BUFSIZE_BITMAP * (pfn / PFN_BUFBITMAP);
+
+    while (old_offset < new_offset) {
+        /* calculate the offset and write dump_bitmap */
+        offset_bitmap1 = s->offset_dump_bitmap + old_offset;
+        if (write_buffer(s->fd, offset_bitmap1, buf,
+                         BUFSIZE_BITMAP) < 0) {
+            return -1;
+        }
+
+        /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
+        offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
+                         old_offset;
+        if (write_buffer(s->fd, offset_bitmap2, buf,
+                         BUFSIZE_BITMAP) < 0) {
+            return -1;
+        }
+
+        memset(buf, 0, BUFSIZE_BITMAP);
+        old_offset += BUFSIZE_BITMAP;
+    }
+
+    /* get the exact place of the bit in the buf, and set it */
+    byte = (pfn % PFN_BUFBITMAP) / CHAR_BIT;
+    bit = (pfn % PFN_BUFBITMAP) % CHAR_BIT;
+    if (value) {
+        buf[byte] |= 1u << bit;
+    } else {
+        buf[byte] &= ~(1u << bit);
+    }
+
+    return 0;
+}
+
+/*
+ * exam every page and return the page frame number and the address of the page.
+ * bufptr can be NULL. note: the blocks here is supposed to reflect guest-phys
+ * blocks, so block->target_start and block->target_end should be interal
+ * multiples of the target page size.
+ */
+static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
+                          uint8_t **bufptr, DumpState *s)
+{
+    GuestPhysBlock *block = *blockptr;
+    hwaddr addr;
+    uint8_t *buf;
+
+    /* block == NULL means the start of the iteration */
+    if (!block) {
+        block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
+        *blockptr = block;
+        assert(block->target_start % s->page_size == 0);
+        assert(block->target_end % s->page_size == 0);
+        *pfnptr = paddr_to_pfn(block->target_start, s->page_shift);
+        if (bufptr) {
+            *bufptr = block->host_addr;
+        }
+        return true;
+    }
+
+    *pfnptr = *pfnptr + 1;
+    addr = pfn_to_paddr(*pfnptr, s->page_shift);
+
+    if ((addr >= block->target_start) &&
+        (addr + s->page_size <= block->target_end)) {
+        buf = block->host_addr + (addr - block->target_start);
+    } else {
+        /* the next page is in the next block */
+        block = QTAILQ_NEXT(block, next);
+        *blockptr = block;
+        if (!block) {
+            return false;
+        }
+        assert(block->target_start % s->page_size == 0);
+        assert(block->target_end % s->page_size == 0);
+        *pfnptr = paddr_to_pfn(block->target_start, s->page_shift);
+        buf = block->host_addr;
+    }
+
+    if (bufptr) {
+        *bufptr = buf;
+    }
+
+    return true;
+}
+
+static int write_dump_bitmap(DumpState *s)
+{
+    int ret = 0;
+    uint64_t last_pfn, pfn;
+    void *dump_bitmap_buf;
+    size_t num_dumpable;
+    GuestPhysBlock *block_iter = NULL;
+
+    /* dump_bitmap_buf is used to store dump_bitmap temporarily */
+    dump_bitmap_buf = g_malloc0(BUFSIZE_BITMAP);
+
+    num_dumpable = 0;
+    last_pfn = 0;
+
+    /*
+     * exam memory page by page, and set the bit in dump_bitmap corresponded
+     * to the existing page.
+     */
+    while (get_next_page(&block_iter, &pfn, NULL, s)) {
+        ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
+        if (ret < 0) {
+            dump_error(s, "dump: failed to set dump_bitmap.\n");
+            ret = -1;
+            goto out;
+        }
+
+        last_pfn = pfn;
+        num_dumpable++;
+    }
+
+    /*
+     * set_dump_bitmap will always leave the recently set bit un-sync. Here we
+     * set last_pfn + PFN_BUFBITMAP to 0 and those set but un-sync bit will be
+     * synchronized into vmcore.
+     */
+    if (num_dumpable > 0) {
+        ret = set_dump_bitmap(last_pfn, last_pfn + PFN_BUFBITMAP, false,
+                              dump_bitmap_buf, s);
+        if (ret < 0) {
+            dump_error(s, "dump: failed to sync dump_bitmap.\n");
+            ret = -1;
+            goto out;
+        }
+    }
+
+    /* number of dumpable pages that will be dumped later */
+    s->num_dumpable = num_dumpable;
+
+out:
+    g_free(dump_bitmap_buf);
+
+    return ret;
+}
+
+static void prepare_data_cache(DataCache *data_cache, DumpState *s,
+                               off_t offset)
+{
+    data_cache->fd = s->fd;
+    data_cache->data_size = 0;
+    data_cache->buf_size = BUFSIZE_DATA_CACHE;
+    data_cache->buf = g_malloc0(BUFSIZE_DATA_CACHE);
+    data_cache->offset = offset;
+}
+
+static int write_cache(DataCache *dc, const void *buf, size_t size,
+                       bool flag_sync)
+{
+    /*
+     * dc->buf_size should not be less than size, otherwise dc will never be
+     * enough
+     */
+    assert(size <= dc->buf_size);
+
+    /*
+     * if flag_sync is set, synchronize data in dc->buf into vmcore.
+     * otherwise check if the space is enough for caching data in buf, if not,
+     * write the data in dc->buf to dc->fd and reset dc->buf
+     */
+    if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
+        (flag_sync && dc->data_size > 0)) {
+        if (write_buffer(dc->fd, dc->offset, dc->buf, dc->data_size) < 0) {
+            return -1;
+        }
+
+        dc->offset += dc->data_size;
+        dc->data_size = 0;
+    }
+
+    if (!flag_sync) {
+        memcpy(dc->buf + dc->data_size, buf, size);
+        dc->data_size += size;
+    }
+
+    return 0;
+}
+
+static void free_data_cache(DataCache *data_cache)
+{
+    g_free(data_cache->buf);
+}
+
+static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
+{
+    size_t len_buf_out_zlib, len_buf_out_lzo, len_buf_out_snappy;
+    size_t len_buf_out;
+
+    /* init buf_out */
+    len_buf_out_zlib = len_buf_out_lzo = len_buf_out_snappy = 0;
+
+    /* buf size for zlib */
+    len_buf_out_zlib = compressBound(page_size);
+
+    /* buf size for lzo */
+#ifdef CONFIG_LZO
+    if (flag_compress & DUMP_DH_COMPRESSED_LZO) {
+        if (lzo_init() != LZO_E_OK) {
+            /* return 0 to indicate lzo is unavailable */
+            return 0;
+        }
+    }
+
+    /*
+     * LZO will expand incompressible data by a little amount. please check the
+     * following URL to see the expansion calculation:
+     * http://www.oberhumer.com/opensource/lzo/lzofaq.php
+     */
+    len_buf_out_lzo = page_size + page_size / 16 + 64 + 3;
+#endif
+
+#ifdef CONFIG_SNAPPY
+    /* buf size for snappy */
+    len_buf_out_snappy = snappy_max_compressed_length(page_size);
+#endif
+
+    /* get the biggest that can store all kinds of compressed page */
+    len_buf_out = MAX(len_buf_out_zlib,
+                      MAX(len_buf_out_lzo, len_buf_out_snappy));
+
+    return len_buf_out;
+}
+
+/*
+ * check if the page is all 0
+ */
+static inline bool is_zero_page(const uint8_t *buf, size_t page_size)
+{
+    return buffer_is_zero(buf, page_size);
+}
+
+static int write_dump_pages(DumpState *s)
+{
+    int ret = 0;
+    DataCache page_desc, page_data;
+    size_t len_buf_out, size_out;
+#ifdef CONFIG_LZO
+    lzo_bytep wrkmem = NULL;
+#endif
+    uint8_t *buf_out = NULL;
+    off_t offset_desc, offset_data;
+    PageDescriptor pd, pd_zero;
+    uint8_t *buf;
+    int endian = s->dump_info.d_endian;
+    GuestPhysBlock *block_iter = NULL;
+    uint64_t pfn_iter;
+
+    /* get offset of page_desc and page_data in dump file */
+    offset_desc = s->offset_page;
+    offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
+
+    prepare_data_cache(&page_desc, s, offset_desc);
+    prepare_data_cache(&page_data, s, offset_data);
+
+    /* prepare buffer to store compressed data */
+    len_buf_out = get_len_buf_out(s->page_size, s->flag_compress);
+    if (len_buf_out == 0) {
+        dump_error(s, "dump: failed to get length of output buffer.\n");
+        goto out;
+    }
+
+#ifdef CONFIG_LZO
+    wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
+#endif
+
+    buf_out = g_malloc(len_buf_out);
+
+    /*
+     * init zero page's page_desc and page_data, because every zero page
+     * uses the same page_data
+     */
+    pd_zero.size = cpu_convert_to_target32(s->page_size, endian);
+    pd_zero.flags = cpu_convert_to_target32(0, endian);
+    pd_zero.offset = cpu_convert_to_target64(offset_data, endian);
+    pd_zero.page_flags = cpu_convert_to_target64(0, endian);
+    buf = g_malloc0(s->page_size);
+    ret = write_cache(&page_data, buf, s->page_size, false);
+    g_free(buf);
+    if (ret < 0) {
+        dump_error(s, "dump: failed to write page data(zero page).\n");
+        goto out;
+    }
+
+    offset_data += s->page_size;
+
+    /*
+     * dump memory to vmcore page by page. zero page will all be resided in the
+     * first page of page section
+     */
+    while (get_next_page(&block_iter, &pfn_iter, &buf, s)) {
+        /* check zero page */
+        if (is_zero_page(buf, s->page_size)) {
+            ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
+                              false);
+            if (ret < 0) {
+                dump_error(s, "dump: failed to write page desc.\n");
+                goto out;
+            }
+        } else {
+            /*
+             * not zero page, then:
+             * 1. compress the page
+             * 2. write the compressed page into the cache of page_data
+             * 3. get page desc of the compressed page and write it into the
+             *    cache of page_desc
+             *
+             * only one compression format will be used here, for
+             * s->flag_compress is set. But when compression fails to work,
+             * we fall back to save in plaintext.
+             */
+             size_out = len_buf_out;
+             if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
+                    (compress2(buf_out, (uLongf *)&size_out, buf, s->page_size,
+                    Z_BEST_SPEED) == Z_OK) && (size_out < s->page_size)) {
+                pd.flags = cpu_convert_to_target32(DUMP_DH_COMPRESSED_ZLIB,
+                                                   endian);
+                pd.size  = cpu_convert_to_target32(size_out, endian);
+
+                ret = write_cache(&page_data, buf_out, size_out, false);
+                if (ret < 0) {
+                    dump_error(s, "dump: failed to write page data.\n");
+                    goto out;
+                }
+#ifdef CONFIG_LZO
+            } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
+                    (lzo1x_1_compress(buf, s->page_size, buf_out,
+                    (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
+                    (size_out < s->page_size)) {
+                pd.flags = cpu_convert_to_target32(DUMP_DH_COMPRESSED_LZO,
+                                                   endian);
+                pd.size  = cpu_convert_to_target32(size_out, endian);
+
+                ret = write_cache(&page_data, buf_out, size_out, false);
+                if (ret < 0) {
+                    dump_error(s, "dump: failed to write page data.\n");
+                    goto out;
+                }
+#endif
+#ifdef CONFIG_SNAPPY
+            } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
+                    (snappy_compress((char *)buf, s->page_size,
+                    (char *)buf_out, &size_out) == SNAPPY_OK) &&
+                    (size_out < s->page_size)) {
+                pd.flags = cpu_convert_to_target32(
+                                        DUMP_DH_COMPRESSED_SNAPPY, endian);
+                pd.size  = cpu_convert_to_target32(size_out, endian);
+
+                ret = write_cache(&page_data, buf_out, size_out, false);
+                if (ret < 0) {
+                    dump_error(s, "dump: failed to write page data.\n");
+                    goto out;
+                }
+#endif
+            } else {
+                /*
+                 * fall back to save in plaintext, size_out should be
+                 * assigned to s->page_size
+                 */
+                pd.flags = cpu_convert_to_target32(0, endian);
+                size_out = s->page_size;
+                pd.size = cpu_convert_to_target32(size_out, endian);
+
+                ret = write_cache(&page_data, buf, s->page_size, false);
+                if (ret < 0) {
+                    dump_error(s, "dump: failed to write page data.\n");
+                    goto out;
+                }
+            }
+
+            /* get and write page desc here */
+            pd.page_flags = cpu_convert_to_target64(0, endian);
+            pd.offset = cpu_convert_to_target64(offset_data, endian);
+            offset_data += size_out;
+
+            ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
+            if (ret < 0) {
+                dump_error(s, "dump: failed to write page desc.\n");
+                goto out;
+            }
+        }
+    }
+
+    ret = write_cache(&page_desc, NULL, 0, true);
+    if (ret < 0) {
+        dump_error(s, "dump: failed to sync cache for page_desc.\n");
+        goto out;
+    }
+    ret = write_cache(&page_data, NULL, 0, true);
+    if (ret < 0) {
+        dump_error(s, "dump: failed to sync cache for page_data.\n");
+        goto out;
+    }
+
+out:
+    free_data_cache(&page_desc);
+    free_data_cache(&page_data);
+
+#ifdef CONFIG_LZO
+    g_free(wrkmem);
+#endif
+
+    g_free(buf_out);
+
+    return ret;
+}
+
+static int create_kdump_vmcore(DumpState *s)
+{
+    int ret;
+
+    /*
+     * the kdump-compressed format is:
+     *                                               File offset
+     *  +------------------------------------------+ 0x0
+     *  |    main header (struct disk_dump_header) |
+     *  |------------------------------------------+ block 1
+     *  |    sub header (struct kdump_sub_header)  |
+     *  |------------------------------------------+ block 2
+     *  |            1st-dump_bitmap               |
+     *  |------------------------------------------+ block 2 + X blocks
+     *  |            2nd-dump_bitmap               | (aligned by block)
+     *  |------------------------------------------+ block 2 + 2 * X blocks
+     *  |  page desc for pfn 0 (struct page_desc)  | (aligned by block)
+     *  |  page desc for pfn 1 (struct page_desc)  |
+     *  |                    :                     |
+     *  |------------------------------------------| (not aligned by block)
+     *  |         page data (pfn 0)                |
+     *  |         page data (pfn 1)                |
+     *  |                    :                     |
+     *  +------------------------------------------+
+     */
+
+    ret = write_start_flat_header(s->fd);
+    if (ret < 0) {
+        dump_error(s, "dump: failed to write start flat header.\n");
+        return -1;
+    }
+
+    ret = write_dump_header(s);
+    if (ret < 0) {
+        return -1;
+    }
+
+    ret = write_dump_bitmap(s);
+    if (ret < 0) {
+        return -1;
+    }
+
+    ret = write_dump_pages(s);
+    if (ret < 0) {
+        return -1;
+    }
+
+    ret = write_end_flat_header(s->fd);
+    if (ret < 0) {
+        dump_error(s, "dump: failed to write end flat header.\n");
+        return -1;
+    }
+
+    dump_completed(s);
+
+    return 0;
+}
+
 static ram_addr_t get_start_block(DumpState *s)
 {
     GuestPhysBlock *block;
@@ -714,7 +1532,16 @@ static ram_addr_t get_start_block(DumpState *s)
     return -1;
 }
 
-static int dump_init(DumpState *s, int fd, bool paging, bool has_filter,
+static void get_max_mapnr(DumpState *s)
+{
+    GuestPhysBlock *last_block;
+
+    last_block = QTAILQ_LAST(&s->guest_phys_blocks.head, GuestPhysBlockHead);
+    s->max_mapnr = paddr_to_pfn(last_block->target_end, s->page_shift);
+}
+
+static int dump_init(DumpState *s, int fd, bool has_format,
+                     DumpGuestMemoryFormat format, bool paging, bool has_filter,
                      int64_t begin, int64_t length, Error **errp)
 {
     CPUState *cpu;
@@ -722,6 +1549,11 @@ static int dump_init(DumpState *s, int fd, bool paging, bool has_filter,
     Error *err = NULL;
     int ret;
 
+    /* kdump-compressed is conflict with paging and filter */
+    if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
+        assert(!paging && !has_filter);
+    }
+
     if (runstate_is_running()) {
         vm_stop(RUN_STATE_SAVE_VM);
         s->resume = true;
@@ -782,6 +1614,38 @@ static int dump_init(DumpState *s, int fd, bool paging, bool has_filter,
         qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
     }
 
+    s->nr_cpus = nr_cpus;
+    s->page_size = TARGET_PAGE_SIZE;
+    s->page_shift = ffs(s->page_size) - 1;
+
+    get_max_mapnr(s);
+
+    uint64_t tmp;
+    tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT), s->page_size);
+    s->len_dump_bitmap = tmp * s->page_size;
+
+    /* init for kdump-compressed format */
+    if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
+        switch (format) {
+        case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
+            s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
+            break;
+
+        case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
+            s->flag_compress = DUMP_DH_COMPRESSED_LZO;
+            break;
+
+        case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
+            s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
+            break;
+
+        default:
+            s->flag_compress = 0;
+        }
+
+        return 0;
+    }
+
     if (s->has_filter) {
         memory_mapping_filter(&s->list, s->begin, s->length);
     }
@@ -841,14 +1705,25 @@ cleanup:
 }
 
 void qmp_dump_guest_memory(bool paging, const char *file, bool has_begin,
-                           int64_t begin, bool has_length, int64_t length,
-                           Error **errp)
+                           int64_t begin, bool has_length,
+                           int64_t length, bool has_format,
+                           DumpGuestMemoryFormat format, Error **errp)
 {
     const char *p;
     int fd = -1;
     DumpState *s;
     int ret;
 
+    /*
+     * kdump-compressed format need the whole memory dumped, so paging or
+     * filter is not supported here.
+     */
+    if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
+        (paging || has_begin || has_length)) {
+        error_setg(errp, "kdump-compressed format doesn't support paging or "
+                         "filter");
+        return;
+    }
     if (has_begin && !has_length) {
         error_set(errp, QERR_MISSING_PARAMETER, "length");
         return;
@@ -858,6 +1733,21 @@ void qmp_dump_guest_memory(bool paging, const char *file, bool has_begin,
         return;
     }
 
+    /* check whether lzo/snappy is supported */
+#ifndef CONFIG_LZO
+    if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
+        error_setg(errp, "kdump-lzo is not available now");
+        return;
+    }
+#endif
+
+#ifndef CONFIG_SNAPPY
+    if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
+        error_setg(errp, "kdump-snappy is not available now");
+        return;
+    }
+#endif
+
 #if !defined(WIN32)
     if (strstart(file, "fd:", &p)) {
         fd = monitor_get_fd(cur_mon, p, errp);
@@ -882,15 +1772,55 @@ void qmp_dump_guest_memory(bool paging, const char *file, bool has_begin,
 
     s = g_malloc0(sizeof(DumpState));
 
-    ret = dump_init(s, fd, paging, has_begin, begin, length, errp);
+    ret = dump_init(s, fd, has_format, format, paging, has_begin,
+                    begin, length, errp);
     if (ret < 0) {
         g_free(s);
         return;
     }
 
-    if (create_vmcore(s) < 0 && !error_is_set(s->errp)) {
-        error_set(errp, QERR_IO_ERROR);
+    if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
+        if (create_kdump_vmcore(s) < 0 && !error_is_set(s->errp)) {
+            error_set(errp, QERR_IO_ERROR);
+        }
+    } else {
+        if (create_vmcore(s) < 0 && !error_is_set(s->errp)) {
+            error_set(errp, QERR_IO_ERROR);
+        }
     }
 
     g_free(s);
 }
+
+DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
+{
+    DumpGuestMemoryFormatList *item;
+    DumpGuestMemoryCapability *cap =
+                                  g_malloc0(sizeof(DumpGuestMemoryCapability));
+
+    /* elf is always available */
+    item = g_malloc0(sizeof(DumpGuestMemoryFormatList));
+    cap->formats = item;
+    item->value = DUMP_GUEST_MEMORY_FORMAT_ELF;
+
+    /* kdump-zlib is always available */
+    item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
+    item = item->next;
+    item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB;
+
+    /* add new item if kdump-lzo is available */
+#ifdef CONFIG_LZO
+    item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
+    item = item->next;
+    item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO;
+#endif
+
+    /* add new item if kdump-snappy is available */
+#ifdef CONFIG_SNAPPY
+    item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
+    item = item->next;
+    item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY;
+#endif
+
+    return cap;
+}
This page took 0.047102 seconds and 4 git commands to generate.