From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from userp1040.oracle.com ([156.151.31.81]:19856 "EHLO userp1040.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751051AbcGNXBD (ORCPT ); Thu, 14 Jul 2016 19:01:03 -0400 Date: Thu, 14 Jul 2016 16:05:00 -0700 From: Liu Bo To: kreijack@inwind.it Cc: linux-btrfs , David Sterba Subject: Re: New btrfs sub command: btrfs inspect physical-find Message-ID: <20160714230500.GA16232@localhost.localdomain> Reply-To: bo.li.liu@oracle.com References: <7c20be09-1ac2-1d1f-f545-e2d56578a77b@libero.it> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: <7c20be09-1ac2-1d1f-f545-e2d56578a77b@libero.it> Sender: linux-btrfs-owner@vger.kernel.org List-ID: On Tue, Jul 12, 2016 at 11:40:13PM +0200, Goffredo Baroncelli wrote: > Hi All, > > the enclosed patch adds a new btrfs sub command: "btrfs inspect physical-find". The aim of this new command is to show the physical placement on the disk of a file. Currently it handles all the profiles (single, dup, raid1/10/5/6). > I develop this command in order to show some bug in btrfs RAID5 profile (see next email). > > You can pull the code from: > > https://github.com/kreijack/btrfs-progs.git > > branch > > insp-phy The tool looks similar to the existing 'btrfs-map-logical', yes, btrfs-map-logical has some problems on raid56, and a quick glance shows that it's due to that btrfs_num_copies() can only return 2 copies for raid5 and 3 copies for raid6. Thanks, -liubo > > The syntax of this new command is simple: > > # btrfs inspect physical-find [] > > where: > is the file to inspect > is the offset of the file to inspect (default 0) > > Below some examples: > > ** Single > > $ sudo mkfs.btrfs -f -d single -m single /dev/loop0 > $ sudo mount /dev/loop0 mnt/ > $ python -c "print 'ad'+'a'*65534+'bd'+'b'*65533" | sudo tee mnt/out.txt >/dev/null > $ sudo ../btrfs-progs/btrfs inspect physical-find mnt/out.txt > mnt/out.txt: 0 > devid 1, /dev/loop0 : 12582912 LINEAR > $ dd 2>/dev/null if=/dev/loop0 skip=12582912 bs=1 count=5; echo > adaaa > > ** Dup > > The command shows both the copies > > $ sudo mkfs.btrfs -f -d single -m single /dev/loop0 > $ sudo mount /dev/loop0 mnt/ > $ python -c "print 'ad'+'a'*65534+'bd'+'b'*65533" | sudo tee mnt/out.txt >/dev/null > $ sudo ../btrfs-progs/btrfs inspect physical-find mnt/out.txt > $ sudo ../btrfs-progs/btrfs inspect physical-find mnt/out.txt > mnt/out.txt: 0 > devid 1, /dev/loop0 : 71303168 DUP > devid 1, /dev/loop0 : 104857600 DUP > $ dd 2>/dev/null if=/dev/loop0 skip=104857600 bs=1 count=5 ; echo > adaaa > > > ** Raid1 > > The command shows both the copies > > $ sudo mkfs.btrfs -f -d raid1 -m raid1 /dev/loop0 /dev/loop1 > $ sudo mount /dev/loop0 mnt/ > $ python -c "print 'ad'+'a'*65534+'bd'+'b'*65533" | sudo tee mnt/out.txt >/dev/null > $ sudo ../btrfs-progs/btrfs inspect physical-find mnt/out.txt mnt/out.txt: 0 > devid 2, /dev/loop1 : 61865984 RAID1 > devid 1, /dev/loop0 : 81788928 RAID1 > $ dd 2>/dev/null if=/dev/loop0 skip=81788928 bs=1 count=5; echo > adaaa > > > ** Raid10 > > The command show both the copies; if you set an offset to the next disk-stripe, you can see the next pair of disk-stripe > > $ sudo mkfs.btrfs -f -d raid10 -m raid10 /dev/loop[0123] > $ sudo mount /dev/loop0 mnt/ > $ python -c "print 'ad'+'a'*65534+'bd'+'b'*65533" | sudo tee mnt/out.txt >/dev/null > $ sudo ../btrfs-progs/btrfs inspect physical-find mnt/out.txt mnt/out.txt: 0 > devid 4, /dev/loop3 : 61931520 RAID10 > devid 3, /dev/loop2 : 61931520 RAID10 > $ dd 2>/dev/null if=/dev/loop2 skip=61931520 bs=1 count=5; echo > adaaa > $ sudo ../btrfs-progs/btrfs inspect physical-find mnt/out.txt 65536 > mnt/out.txt: 65536 > devid 2, /dev/loop1 : 61931520 RAID10 > devid 1, /dev/loop0 : 81854464 RAID10 > $ dd 2>/dev/null if=/dev/loop0 skip=81854464 bs=1 count=5; echo > bdbbb > > > ** Raid5 > > Depending by the offset, you can see which disk-stripe is used. > > $ sudo mkfs.btrfs -f -d raid5 -m raid5 /dev/loop[012] > $ sudo mount /dev/loop0 mnt/ > $ python -c "print 'ad'+'a'*65534+'bd'+'b'*65533" | sudo tee mnt/out.txt >/dev/null > $ sudo ../btrfs-progs/btrfs inspect physical-find mnt/out.txt > mnt/out.txt: 0 > devid 2, /dev/loop1 : 61931520 DATA > devid 1, /dev/loop0 : 81854464 OTHER > devid 3, /dev/loop2 : 61931520 PARITY > $ sudo ../btrfs-progs/btrfs inspect physical-find mnt/out.txt 65536mnt/out.txt: 65536 > devid 2, /dev/loop1 : 61931520 OTHER > devid 1, /dev/loop0 : 81854464 DATA > devid 3, /dev/loop2 : 61931520 PARITY > $ dd 2>/dev/null if=/dev/loop1 skip=61931520 bs=1 count=5; echo > adaaa > $ dd 2>/dev/null if=/dev/loop0 skip=81854464 bs=1 count=5; echo > bdbbb > $ dd 2>/dev/null if=/dev/loop2 skip=61931520 bs=1 count=5 | xxd > 00000000: 0300 0303 03 ..... > > The parity is computed as: parity=disk1^disk2. So "adaa" ^ "bdbb" == "\x03\x00\x03\x03 > > ** Raid6 > $ sudo mkfs.btrfs -f -mraid6 -draid6 /dev/loop[0-4]^C > $ sudo mount /dev/loop0 mnt/ > $ python -c "print 'ad'+'a'*65534+'bd'+'b'*65533" | sudo tee mnt/out.txt >/dev/null > $ sudo ../btrfs-progs/btrfs inspect physical-find mnt/out.txt > mnt/out.txt: 0 > devid 3, /dev/loop2 : 61931520 DATA > devid 2, /dev/loop1 : 61931520 OTHER > devid 1, /dev/loop0 : 81854464 PARITY > devid 4, /dev/loop3 : 61931520 PARITY > > $ dd 2>/dev/null if=/dev/loop2 skip=61931520 bs=1 count=5 ; echo > adaaa > > -- > > diff --git a/cmds-inspect.c b/cmds-inspect.c > index dd7b9dd..a604c2b 100644 > --- a/cmds-inspect.c > +++ b/cmds-inspect.c > @@ -22,6 +22,11 @@ > #include > #include > #include > +#include > +#include > +#include > +#include > +#include > > #include "kerncompat.h" > #include "ioctl.h" > @@ -623,6 +628,450 @@ out: > return !!ret; > } > > + > +static const char* const cmd_inspect_physical_find_usage[] = { > + "btrfs inspect-internal physical-find [options] [...]", > + "Show the physical address of each blocks", > + "-m the output is machine readable", > + NULL > +}; > + > +static void dump_stripes(int ndisks, struct btrfs_ioctl_dev_info_args *disks, > + struct btrfs_chunk *chunk, u64 logical_start) { > + struct btrfs_stripe *stripes; > + stripes = &chunk->stripe; > + > + if ((chunk->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ) { > + /* LINEAR: each chunk has (should have) only one disk */ > + int j; > + char *dname = ""; > + > + assert(chunk->num_stripes == 1); > + > + u64 phy_start = stripes[0].offset + > + +logical_start; > + for (j = 0 ; j < ndisks ; j++) > + if (stripes[0].devid == disks[j].devid) { > + dname = (char*)disks[j].path; > + break; > + } > + printf("\tdevid %llu, %s : %llu LINEAR\n", > + stripes[0].devid, dname, phy_start); > + } else if (chunk->type & BTRFS_BLOCK_GROUP_RAID0) { > + /* > + * RAID0: each chunk is composed by more disks; > + * each stripe_len bytes are in a different disk: > + * > + * file: ABC...NMOP.... > + * > + * disk1 disk2 disk3 .... disksN > + * > + * A B C .... N > + * M O P .... > + * > + */ > + u64 disks_number = chunk->num_stripes; > + u64 disk_stripe_size = chunk->stripe_len; > + u64 stripe_capacity ; > + u64 stripe_nr; > + u64 disk_stripe_start; > + int sidx; > + int j; > + char *dname = ""; > + > + stripe_capacity = disks_number * disk_stripe_size; > + stripe_nr = logical_start / stripe_capacity; > + disk_stripe_start = logical_start % disk_stripe_size; > + > + sidx = (logical_start / disk_stripe_size) % disks_number; > + > + u64 phy_start = stripes[sidx].offset + > + stripe_nr * disk_stripe_size + > + disk_stripe_start; > + > + for (j = 0 ; j < ndisks ; j++) > + if (stripes[sidx].devid == disks[j].devid) { > + dname = (char*)disks[j].path; > + break; > + } > + printf("\tdevid %llu, %s : %llu RAID0\n", > + stripes[sidx].devid, dname, phy_start); > + > + } else if (chunk->type & BTRFS_BLOCK_GROUP_RAID1) { > + /* > + * RAID0: each chunk is composed by more disks; > + * each stripe_len bytes are in a different disk: > + * > + * file: ABC... > + * > + * disk1 disk2 disk3 .... > + * > + * A A > + * B B > + * C C > + * > + */ > + int sidx; > + for (sidx = 0; sidx < chunk->num_stripes; sidx++) { > + int j; > + char *dname = ""; > + u64 phy_start = stripes[sidx].offset + > + +logical_start; > + > + for (j = 0 ; j < ndisks ; j++) > + if (stripes[sidx].devid == disks[j].devid) { > + dname = (char*)disks[j].path; > + break; > + } > + printf("\tdevid %llu, %s : %llu RAID1\n", > + stripes[sidx].devid, dname, phy_start); > + } > + } else if (chunk->type & BTRFS_BLOCK_GROUP_DUP) { > + /* > + * DUP: each chunk has 'num_stripes' disk_stripe. Heach > + * disk_stripe has its own copy of data > + * > + * file: ABCD.... > + * > + * disk1 disk2 disk3 > + * > + * A > + * B > + * C > + * [...] > + * A > + * B > + * C > + * > + * > + * NOTE: the difference between DUP and RAID1 is that > + * in RAID1 each disk_stripe is in a different disk, in DUP > + * each disk chunk is in the same disk > + */ > + int sidx; > + /* TBD: check what happens with the stripes */ > + for (sidx = 0; sidx < chunk->num_stripes; sidx++) { > + int j; > + char *dname = ""; > + u64 phy_start = stripes[sidx].offset + > + +logical_start; > + > + for (j = 0 ; j < ndisks ; j++) > + if (stripes[sidx].devid == disks[j].devid) { > + dname = (char*)disks[j].path; > + break; > + } > + printf("\tdevid %llu, %s : %llu DUP\n", > + stripes[sidx].devid, dname, phy_start); > + } > + } else if (chunk->type & BTRFS_BLOCK_GROUP_RAID10) { > + /* > + * RAID10: each chunk is composed by more disks; > + * each stripe_len bytes are in a different disk: > + * > + * file: ABCD.... > + * > + * disk1 disk2 disk3 disk4 > + * > + * A A B B > + * C C D D > + * > + * > + */ > + int i; > + u64 disks_number = chunk->num_stripes; > + u64 disk_stripe_size = chunk->stripe_len; > + u64 stripe_capacity ; > + u64 stripe_nr; > + u64 stripe_start; > + u64 disk_stripe_start; > + > + stripe_capacity = disks_number * disk_stripe_size / chunk->sub_stripes; > + stripe_nr = logical_start / stripe_capacity; > + stripe_start = logical_start % stripe_capacity; > + disk_stripe_start = logical_start % disk_stripe_size; > + > + for (i = 0; i < chunk->sub_stripes; i++) { > + int j; > + char *dname = ""; > + int sidx = (i + > + stripe_start/disk_stripe_size*chunk->sub_stripes) % > + disks_number; > + > + u64 phy_start = stripes[sidx].offset + > + +stripe_nr*disk_stripe_size + disk_stripe_start; > + > + for (j = 0 ; j < ndisks ; j++) > + if (stripes[sidx].devid == disks[j].devid) { > + dname = (char*)disks[j].path; > + break; > + } > + printf("\tdevid %llu, %s : %llu RAID10\n", > + stripes[sidx].devid, dname, phy_start); > + } > + } else if (chunk->type & BTRFS_BLOCK_GROUP_RAID5 || > + chunk->type & BTRFS_BLOCK_GROUP_RAID6 ) { > + /* > + * RAID5: each chunk is spread on a different disk; however one > + * disk is used for parity > + * > + * file: ABCDEFGHIJK.... > + * > + * disk1 disk2 disk3 disk4 disk5 > + * > + * A B C D P > + * P D E F G > + * H P I J K > + * > + * Note: P == parity > + * > + * RAID6: each chunk is spread on a different disk; however two > + * disks are used for parity > + * > + * file: ABCDEFGHI... > + * > + * disk1 disk2 disk3 disk4 disk5 > + * > + * A B C P Q > + * Q D E F P > + * P Q G H I > + * > + * Note: P,Q == parity > + * > + */ > + int parities_nr = 1; > + u64 disks_number = chunk->num_stripes; > + u64 disk_stripe_size = chunk->stripe_len; > + u64 stripe_capacity ; > + u64 stripe_nr; > + u64 stripe_start; > + u64 pos = 0; > + u64 disk_stripe_start; > + int sidx; > + > + if (chunk->type & BTRFS_BLOCK_GROUP_RAID6) > + parities_nr = 2; > + > + stripe_capacity = (disks_number - parities_nr) * > + disk_stripe_size; > + stripe_nr = logical_start / stripe_capacity; > + stripe_start = logical_start % stripe_capacity; > + disk_stripe_start = logical_start % disk_stripe_size; > + > + for (sidx = 0; sidx < disks_number ; sidx++) { > + int j; > + char *dname = ""; > + u64 stripe_index = (sidx + stripe_nr) % disks_number; > + u64 phy_start = stripes[stripe_index].offset + /* chunk start */ > + + stripe_nr*disk_stripe_size + /* stripe start */ > + + disk_stripe_start; > + > + for (j = 0 ; j < ndisks ; j++) > + if (stripes[stripe_index].devid == disks[j].devid) { > + dname = (char*)disks[j].path; > + break; > + } > + > + if (sidx >= (disks_number - parities_nr)) { > + printf("\tdevid %llu, %s : %llu PARITY\n", > + stripes[stripe_index].devid, dname, > + phy_start); > + continue; > + } > + > + if (stripe_start >= pos && stripe_start < (pos+disk_stripe_size)) { > + printf("\tdevid %llu, %s : %llu DATA\n", > + stripes[stripe_index].devid, > + dname, phy_start); > + } else { > + printf("\tdevid %llu, %s : %llu OTHER\n", > + stripes[stripe_index].devid, > + dname, phy_start); > + } > + > + pos += disk_stripe_size; > + } > + assert(pos == stripe_capacity); > + } else { > + error("Unknown chunk type = 0x%016llx\n", chunk->type); > + return; > + } > + > +} > + > +static int dump_extent(char *fname, int fd, u64 logical_start) { > + > + struct btrfs_ioctl_search_args args; > + struct btrfs_ioctl_search_key *sk = &args.key; > + struct btrfs_ioctl_search_header sh; > + unsigned long off = 0; > + int i; > + int e; > + struct btrfs_ioctl_dev_info_args *disks = NULL; > + struct btrfs_ioctl_fs_info_args fi_args = {0}; > + > + e = get_fs_info(fname, &fi_args, &disks); > + if ( e< 0) { > + error("Cannot get info for the filesystem: may be it is not a btrfs filesystem ?\n"); > + free(disks); > + return -1; > + } > + > + memset(&args, 0, sizeof(args)); > + sk->tree_id = BTRFS_CHUNK_TREE_OBJECTID; > + sk->min_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; > + sk->max_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; > + sk->min_type = BTRFS_CHUNK_ITEM_KEY; > + sk->max_type = BTRFS_CHUNK_ITEM_KEY; > + sk->max_offset = (u64)-1; > + sk->min_offset = 0; > + sk->max_transid = (u64)-1; > + > + while (1) { > + int ret; > + > + sk->nr_items = 1; > + ret = ioctl(fd, BTRFS_IOC_TREE_SEARCH, &args); > + e = errno; > + if (ret < 0) { > + error("cannot perform the search: %s", strerror(e)); > + free(disks); > + return -1; > + } > + if (sk->nr_items == 0) > + break; > + > + off = 0; > + for (i = 0; i < sk->nr_items; i++) { > + struct btrfs_chunk *item; > + > + memcpy(&sh, args.buf + off, sizeof(sh)); > + off += sizeof(sh); > + item = (struct btrfs_chunk*)(args.buf + off); > + off += sh.len; > + > + if (logical_start >= sh.offset && > + logical_start <= sh.offset+item->length) { > + dump_stripes(fi_args.num_devices, disks, > + item, > + logical_start-sh.offset); > + free(disks); > + return 0; > + } > + > + > + sk->min_objectid = sh.objectid; > + sk->min_type = sh.type; > + sk->min_offset = sh.offset; > + } > + > + if (sk->min_offset < (u64)-1) > + sk->min_offset++; > + else > + break; > + } > + > + free(disks); > + return 0; > +} > + > +/* > + * Inline extents are skipped because they do not take data space, > + * delalloc and unknown are skipped because we do not know how much > + * space they will use yet. > + */ > +#define SKIP_FLAGS (FIEMAP_EXTENT_UNKNOWN|FIEMAP_EXTENT_DELALLOC| \ > + FIEMAP_EXTENT_DATA_INLINE) > +static int cmd_inspect_physical_find(int argc, char **argv) > +{ > + int ret = 0; > + u64 logical = 0ull; > + int fd; > + int last = 0; > + char buf[16384]; > + char *fname; > + int found = 0; > + struct fiemap *fiemap = (struct fiemap*)buf; > + struct fiemap_extent *fm_ext = &fiemap->fm_extents[0]; > + const int count = (sizeof(buf) - sizeof(*fiemap)) / > + sizeof(struct fiemap_extent); > + > + int minargc = 1; > + > + memset(fiemap, 0, sizeof(struct fiemap)); > + > + if (check_argc_min(argc - minargc, 1) || > + check_argc_max(argc - minargc, 2) ) > + usage(cmd_inspect_physical_find_usage); > + > + if (argc - minargc == 2) > + logical = strtoull(argv[minargc+1], NULL, 0); > + fname = argv[minargc]; > + > + printf("%s: %llu\n", fname, logical); > + > + fd = open(fname, O_RDONLY); > + if (fd < 0) { > + error("Can't open '%s' for reading\n", fname); > + ret = -errno; > + goto out; > + } > + > + do { > + > + int rc; > + int j; > + > + fiemap->fm_length = ~0ULL; > + fiemap->fm_extent_count = count; > + fiemap->fm_flags = FIEMAP_FLAG_SYNC; > + rc = ioctl(fd, FS_IOC_FIEMAP, (unsigned long) fiemap); > + if (rc < 0) { > + error("Can't do ioctl()\n"); > + close(fd); > + ret = -errno; > + goto out; > + } > + > + for (j = 0; j < fiemap->fm_mapped_extents; j++) { > + u32 flags = fm_ext[j].fe_flags; > + > + fiemap->fm_start = (fm_ext[j].fe_logical + > + fm_ext[j].fe_length); > + > + if (flags & FIEMAP_EXTENT_LAST) > + last = 1; > + > + if (flags & SKIP_FLAGS) > + continue; > + > + if (logical > fm_ext[j].fe_logical + > + fm_ext[j].fe_length) > + continue; > + > + found = 1; > + > + rc = dump_extent(fname, fd, > + fm_ext[j].fe_physical + logical - > + fm_ext[j].fe_logical); > + if (rc < 0) > + ret = -errno; > + last = 1; > + break; > + } > + } while (last == 0); > + > + close(fd); > + > + if (!found) { > + error("Can't find the extent: the file is too short, or the file is stored in a leaf.\n"); > + ret = 10; > + } > + > +out: > + return ret; > +} > + > static const char inspect_cmd_group_info[] = > "query various internal information"; > > @@ -644,6 +1093,8 @@ const struct cmd_group inspect_cmd_group = { > cmd_inspect_dump_super_usage, NULL, 0 }, > { "tree-stats", cmd_inspect_tree_stats, > cmd_inspect_tree_stats_usage, NULL, 0 }, > + { "physical-find", cmd_inspect_physical_find, > + cmd_inspect_physical_find_usage, NULL, 0 }, > NULL_CMD_STRUCT > } > }; > > > > > -- > gpg @keyserver.linux.it: Goffredo Baroncelli > Key fingerprint BBF5 1610 0B64 DAC6 5F7D 17B2 0EDA 9B37 8B82 E0B5 > > -- > gpg @keyserver.linux.it: Goffredo Baroncelli > Key fingerprint BBF5 1610 0B64 DAC6 5F7D 17B2 0EDA 9B37 8B82 E0B5 > -- > To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html