diff options
Diffstat (limited to 'sbin/recoverdisk/recoverdisk.1')
| -rw-r--r-- | sbin/recoverdisk/recoverdisk.1 | 258 |
1 files changed, 169 insertions, 89 deletions
diff --git a/sbin/recoverdisk/recoverdisk.1 b/sbin/recoverdisk/recoverdisk.1 index 2999ac6ec409..9f1deb4c0c23 100644 --- a/sbin/recoverdisk/recoverdisk.1 +++ b/sbin/recoverdisk/recoverdisk.1 @@ -27,7 +27,7 @@ .Os .Sh NAME .Nm recoverdisk -.Nd recover data from hard disk or optical media +.Nd recover data from disk-like devices. .Sh SYNOPSIS .Nm .Op Fl b Ar bigsize @@ -41,79 +41,101 @@ .Sh DESCRIPTION The .Nm -utility reads data from the +utility reads all data from the .Ar source -file until all blocks could be successfully read. +and retries read operations until they succeed. If .Ar destination -was specified all data is being written to that file. -It starts reading in multiples of the sector size. -Whenever a block fails, it is put to the end of the working queue and will be -read again, possibly with a smaller read size. +is specified all data read be written there. .Pp -By default it uses block sizes of roughly 1 MB, 32kB, and the native -sector size (usually 512 bytes). -These figures are adjusted slightly, for devices whose sectorsize is not a -power of 2, e.g., audio CDs with a sector size of 2352 bytes. +The internal work-list can be saved and loaded so that +.Nm +sessions can be resumed, for instance when a marginal +source hard-disk shuts down. +.Pp +The work-list is initialized with a single item which covers the entire +.Ar source +and +.Nm +always chips away at the first item on the work-list. + +When a read succeeds, that part of the current chunk is eliminated +from the work-list. + +When a read fails, that part of the item is appended to the worklist +as a separate item, and will be retried in due order. +If +.Ar destination +is specified, the corresponding range is filled with '_UNREAD_'. +.Pp +The first pass attempts to read everything in "big-size" chunks, +the second pass reads in "medium-size" chunks and third and subsequent +passes read in "small-size" chunks. This three stage process is +an attempt to optimize the case where only a few bad blocks exist +on +.Ar source . +If too many read-errors are encountered, +.Nm +will fall back to smaller sizes sooner. +.Pp +The three sizes default to 128kB (or less if the sector size does +not divide 128kB cleanly, for instance audio CD media), and the +reported +.Dv DIOCGSTRIPESIZE +and +.Dv DIOCGSECTORSIZE +respectively. .Pp The options are as follows: .Bl -tag -width indent .It Fl b Ar bigsize -The size of reads attempted first. -The middle pass is roughly the logarithmic average of the bigsize and -the sectorsize. -.It Fl r Ar readlist -Read the list of blocks and block sizes to read from the specified file. -.It Fl s Ar interval -How often we should update the writelist file while things go OK. -The default is 60 and the unit is "progress messages" so if things -go well, this is the same as once per minute. +The size of reads attempted in first pass. +.It Fl m Ar mediumsize +The size of reads attempted in second pass. +.It Fl s Ar smallsize +The size of reads attempted in third and subsequent passes. +.It Fl r Ar work-list-file +Read the work-list from a file. +.It Fl w Ar work-list-file +Write the work-list to a file when a read succeed, but at most once +every minute. +.It Fl l Ar log-file +Each successful read is logged with timestamp, offset and length. +.It Fl t Ar totalsize +How many bytes should be recovered. The default is what +.Dv DIOCGMEDIASIZE +reports for character and block devices or +.Dv st_size +if +.Ar source +is a regular file. +.It Fl p Ar pause +.Xr sleep 3 +this long whenever a read fails. This makes the +.Ar source +device look less sick to the operating system. .It Fl u Ar pattern -By default blocks which encounter read errors will be filled with -the pattern +By default blocks which cannot be read are filled with the pattern .Ql _UNREAD_ -in the output file. -This option can be -used to specify another pattern. -Nothing gets written if the string is empty. +in the output file. This option can be used to specify a different +pattern. If the pattern is the empty string, nothing is written. .It Fl v -Enables nicer status report using ANSI escapes and UTF-8. -.It Fl w Ar writelist -Write the list of remaining blocks to read to the specified file if -.Nm -is aborted via -.Dv SIGINT . +Produce a detailed progress report with ANSI escapes and UTF-8. .El .Pp -The -.Fl r -and -.Fl w -options can be specified together. -Especially, they can point to the same file, which will be updated on abort. -.Sh OUTPUT -The .Nm -utility -prints several columns, detailing the progress -.Bl -tag -width remaining -.It Va start -Starting offset of the current block. -.It Va size -Read size of the current block. -.It Va len -Length of the current block. -.It Va state -Is increased for every failed read. -.It Va done -Number of bytes already read. -.It Va remaining -Number of bytes remaining. -.It Va "% done" -Percent complete. -.El +can be aborted with +.Dv SIGINT , +but with a sick +.Ar source +it may take up to several minutes before the current read operation +returns from the kernel. +.Pp .Sh EXAMPLES .Bd -literal +# check if all sectors can be read on a USB stick: +recoverdisk /dev/da0 + # recover data from failing hard drive ada3 recoverdisk /dev/ada3 /data/disk.img @@ -129,10 +151,72 @@ recoverdisk -r worklist -w worklist /dev/cd0 /data/cd.iso # recover a single file from the unreadable media recoverdisk /cdrom/file.avi file.avi -# If the disk hangs the system on read-errors try: -recoverdisk -b 0 /dev/ada3 /somewhere - .Ed +.Sh PRACTICAL ADVICE +In Datamuseum.dk +.Nm +has been used to recover all sorts of data-media for two decades, +here are some things we have learned: +.Bl -bullet +.It +Interacting with failing hardware has a tendency to crash machines, +so it is always a good idea to use the +.Fl -w work-list-file +so that it is possible to continue. +.It +When attempting to recover hard to read data from failing hard disks, +it pays to pamper the drive as much as possible: +.It +It is generally best to keep the drive in it's usual physical orientation, +but it can also help to try other orientations. +.It +Insulate the drive from external vibrations. +.It +Keep the drive cool with a fan. +.It +If possible, power the drive from a laboratory power supply. +.It +Do not loose patience: Let +.Nm +run as long as possible. +.It +(S)ATA controllers do not handle failing disks well, if this +is a problem, use a USB-(S)ATA adapter instead. +.It +The +.Nm +source code is deliberately written to be easily portable to +older versions of +.Fx +and to other operating systems. +.It +If you need to read ST-506, RLL or ESDI drives +.Fx 3.5.1 +is a good compromise. +.It +Sometimes forcing the disk to step between reads helps. +Since +.Nm +process the work-list in the order it is read, this +can be accomplished by sorting the work-list with +something like: +.Dl % sort +0.5 +.It +By default the +.Xr CAM +layer will retry failing read operations, but that +will get stuck on the bad sectors for long time +and delay recovering what actually can be read from +a rapidly failing drive. +In that situation, set the appropriate +.Dl kern.cam.*.retry_count +sysctl to zero. +.It +For floppies and un-zoned hard disks (ST-506 to +early IDE) set +.Fl b Ar bigsize +to the size of a track. +.El .Sh SEE ALSO .Xr dd 1 , .Xr ada 4 , @@ -143,7 +227,8 @@ recoverdisk -b 0 /dev/ada3 /somewhere The .Nm utility first appeared in -.Fx 7.0 . +.Fx 7.0 +because Somebody™ forgot to make a backup copy. .Sh AUTHORS .An -nosplit The original implementation was done by @@ -151,34 +236,29 @@ The original implementation was done by with minor improvements from .An Ulrich Sp\(:orlein Aq Mt uqs@FreeBSD.org . .Pp -This manual page was written by +This manual page was originally written by .An Ulrich Sp\(:orlein . .Sh BUGS -Reading from media where the sectorsize is not a power of 2 will make all -1 MB reads fail. -This is due to the DMA reads being split up into blocks of at most 128kB. -These reads then fail if the sectorsize is not a divisor of 128kB. -When reading a full raw audio CD, this leads to roughly 700 error messages -flying by. -This is harmless and can be avoided by setting -.Fl b -to no more than 128kB. +If a failing device causes the machine to crash, there is +a risk that a chunk might have been successfully read +and removed from the work-list, but not yet flushed to +the +.Ar destination . .Pp .Nm -needs to know about read errors as fast as possible, i.e., retries by lower -layers will usually slow down the operation. -When using -.Xr cam 4 -attached drives, you may want to set kern.cam.XX.retry_count to zero, e.g.: -.Bd -literal -# sysctl kern.cam.ada.retry_count=0 -# sysctl kern.cam.cd.retry_count=0 -# sysctl kern.cam.da.retry_count=0 -.Ed -.\".Pp -.\"When reading from optical media, a bug in the GEOM framework will -.\"prevent it from seeing that the media has been removed. -.\"The device can still be opened, but all reads will fail. -.\"This is usually harmless, but will send -.\".Nm -.\"into an infinite loop. +calls +.Xr fdatasync 3 +on the destination before writing the work-list to a +temporary file, and calls it again on the temporary +file before renaming it to the specified +.Fl w Ar work-file-list +filename. +But even then things dont always work out. +.Pp +.Nm +should have an option for reconstructing the work-list +from the +.Ar destination +by enumerating the +.Fl u Ar pattern +filled ranges. |